Contextual Information Based Occluded Pedestrian.

Chapter 1: Introduction

1.1 Background

612,500 pedestrians aged 15-29 were killed in 2013, which was the number one

cause of death among that age group [1]. Fully automated driving systems are seen as

possible remedies for reducing road traffic fatalities due to the fact that they do not

possess the fundamental issues of human drivers, such as failure to comply with the

rules, lack of attention while driving, etc. Furthermore, decision-making for

automated driving systems (ADS) is a challenging area that plays a key role in fully

automated systems. Especially, developing intelligent systems taking precautious

actions for objects that are currently unobservable but interacting with the ego

vehicle in the future has attracted much attention recently.

According to SAE [2], there are five levels of driving automation. Levels 1-2 are

defined as Automated Driving Assistance Systems (ADAS), whereas Levels 3-5 are

considered Automated Driving Systems (ADS). Level 3-5 systems are expected to

participate partially or fully in decision-making. Currently, only up to Level 3 systems

are available in the market, and the National Highway Traffic Safety Administration

(NHTSA) has projected that level 4-5 autonomy would arrive not before 2025 [3].

Figure 1.1: An example of environmental cues and how a human driver would easily

decide a similar scheme for emergence probability

The motion prediction and risk assessment are vital in taking precautious actions.

Lefevre et al. [4] categorized the motion prediction and risk assessment methods into

four categories, and stated that despite being computationally more demanding,

interaction-aware methods are more reliable than other methods. Occlusion-aware

[5] risk assessment methods have been proposed as the alternative interaction-aware

methods. Those alternatives vary from solutions based on partially/mixed observable

Markov decision processes (POMDP/MOMDP) [6–13], to solutions based on setbased

methods [14–19].

On the other hand, there are currently two shortcomings of available methods in

the literature. Firstly, human driver performance under occlusion and limited

visibility conditions have been mostly neglected. Secondly, little attention has been

paid to using visible information and prior knowledge to predict pedestrian

emergence out of occluded areas.

This thesis introduces a novel occlusion-aware risk assessment system for ADSs.

The proposed method can estimate pedestrian emergence probability from occluded

areas and adjust its driving policy accordingly. Our method’s overview is

demonstrated in fig. 1.3, where the red ego vehicle uses information such as visible

pedestrians and parked cars to assess the probability of emerging pedestrians to

derive an optimal driving policy.

1.2 Literature review

Various methods from set-based methods to the exploitation of behavior of other

participants in a rule-based fashion were used [20–24]. Althoff et al. [20] improved

the reachability analysis to obtain PID controllers and implemented their method on

a real car. In a similar fashion, Magdici et al. [21] generated so-called fail-safe optimal

trajectories, some of which were designated as “emergency maneuvers” so as to stop

the ego vehicle without any collision; furthermore, the “emergency-maneuvers” in

[21] generated by predicting all possible trajectories of other traffic participants

within the given time horizon. Kousik et al. [22] utilized the duality of low-fidelity and

highfidelity models to generate safe trajectories in arbitrary environments, in real-

time; safe trajectories were generated by taking into account the mismatch between

the models. Koschi et al. [23] used set-based predictions, contextual information, and

traffic rules. Koschi et al. compared their method against that of Thorton et al. [12].

However, only visible pedestrians were considered in [23]. Kapania et al. [24] first

utilized the gap acceptance behavior of pedestrians and deterministic limitations of

vehicles. The gap acceptance behavior and the limitations were used by an FSM

controller. The controller yielded whenever possible, or chose the action to avoid

collisions; furthermore, the controller exploited the gap acceptance behavior in the

different modes, such as aggressive and conservative. However, none of those works

considered occlusions and invisible traffic participants.

1.2.1 Occlusion-Aware Risk Assessment Methods

On the other hand, occlusion-aware risk assessment methods incorporated

obstructed visibility information. These methods can be categorized into two: the

methods based on POMDP/MOMDP representation and the methods based on reach-

ability analysis.

The methods based on POMDP/MOMDP representation used probabilistic models

and Bayesian filtering [6–13]. These methods could estimate the uncertainties to

incorporate the observation along with reward models. Then, the estimation was used

to extract a sub-optimal driving policy. Bouton et al. [9] focused on extracting the

driving policy at an intersection with POMDP representation; the proposal was

compared against a TTC-based method. However, the results in [9] were rather

ambiguous on the account of the fact that the proposed method was outperformed by

the TTCbased method for denser traffic conditions. Bouton et al. [11] extended the

work of [10,25] on utility decomposition to propose a scalable decision-making

method under sensor occlusions; the proposal could handle multiple road users.

However, the action and state resolution were very low on the account of avoiding

exponentially growing computational demand. Schratter et al. [13] achieved safe

control for occluded pedestrians with POMDP representation and the help of

Automatic Emergency Braking (AEB) systems. However, that method suffered from

the low resolution of statespace, and it could handle one pedestrian. In general,

methods relying on POMDP representation and RL have suffered from real-time

inapplicability. In other words, those methods are computationally demanding;

consequently, they are impractical for real-life implementations as the execution time

is rather long in situations that require immediate actions. Secondly, due to lack of

resolution in action and state-space, the driving policy may result in an uncomfortable

driving experience for passengers.

The methods based on reachability analysis determined the risky situations in

advance by over-approximating the future sets of other traffic participants using

reachability analysis [14–19]. Hoerman et al. [14] used dynamic grid maps in three

distinct dynamic grid map layers, namely object-based, object-free, and unobservable

region layer for risk assessment and collision-avoidance; any cell was considered as

occupied had the cell in any of the three layers detected as occupied. Orzechowski et

al. [16] expanded Althoff et al’s reachability analysis so as to also incorporate

occlusions and unobservable traffic users. Furthermore, Orzechowski et al. stated that

Ta¸s and Stiller [17] prioritized safety, did not consider comfort. Naumann et al. [19]

considered the worst-case presumptions for critically unseen traffic users and

claimed to ensure “provably safe but not over-cautious actions”. Furthermore, most of

these methods adopted the notion of blame from [5], which defined safety as choosing

trajectories that do not cause a collision. However, those trajectories may still end in

a collision due to erroneous actions by other traffic participants; moreover, neglecting

to compensate for other’s faults is a major flaw that may result in morally wrong

situations. Those methods assumed the worst-case scenarios in some subset of

possible actions by other traffic participants; specifically, for those which considered

the pedestrians in over-approximated sets. However, a proof of safety by constraining

physical capabilities of other traffic participants is ill-defined, in that there may exist

some cases that those constraints are either too conservative or not sufficiently

representative.

1.2.2 Pedestrians’ Response to an Incoming Vehicle: Gap

Acceptance

The assumption that pedestrians are not influenced by traffic flow on the road is

inaccurate; pedestrians are also intelligent dynamic systems that are able to react to

changes in their surroundings. Furthermore, pedestrians also consider their safety,

that is, they want to cross the road safely. Therefore, over-approximating their

possible movements may result in an over-conservative driving policy. Moreover, it is

not very time-efficient to over-approximate the unsafe behaviors, whereas it is not

safe to under-approximate them either.

One possible solution is to model the pedestrian crossing decisions as a function

of the remaining distance between a pedestrian and an oncoming vehicle. One model

of the crossing decision is called the gap acceptance model; the decisions are

probabilistically modeled as a likelihood function. The gap is defined as in (1.1):

D(Distance)

G(Gap) = (1.1)

V (Velocity)

where D is the distance between the ego vehicle and the pedestrian and V is the

velocity of the ego vehicle.

A realistic way to obtain a reasonably accurate likelihood function is to observe a

large number of pedestrians. The gap acceptance model has been studied at various

locations, such as midblock or crosswalk intersections from different cities with

various conditions in [26–29]. Brewer et al. [26] modeled the behavior by a logistic

regression function of the gap; the data were collected at 42 study sites in seven states.

Similarly, Cherry et al. [27] also modeled the behavior by a logistic regression function

of the gap; by contrast, more variables, e.g., the effective gap, wait time, vehicle speed,

were used in the model. On the other hand, the influence of illegal parking and car

types were considered in Yannis et al’s work [28,29]. In order to explain the model in

detail, the work of [26] can be used; in their work, a logistic regression fitting tool was

used to estimate gap acceptance with respect to gap between the pedestrian and the

vehicle. The tool fitted the rejection behavior into the (1.2) with βx being a first-order

polynomial function of gap.

p(Gap rejection (1.2)

Conversely, the acceptance behavior was fit into the function in (1.3):

p(Gap acceptance|G = d) = 1 − p(Gap rejection|G = d)

(1.3)

Using (1.3), the gap acceptance characteristics from [26] is demonstrated in Fig. 2.10.

Arguably, this model may not be accurate; however, it is open to further improvement

by more sophisticated deep learning methods. Furthermore, this model has validated

that pedestrians cross rarely should the ego vehicle be very close. On the other hand,

the works on the gap acceptance model have not taken into account the distracted

pedestrians; the absence of such pedestrians is a limitation to the generality of the

model. Accordingly, the proposal should also consider such pedestrians in order to

overcome that limitation.

1.2.3 Occlusion-robust Tracking Methods

Occlusion-robuts tracking methods can be facilitated in order to develop a method

that can navigate successfully in scenario with multiple pedestrians. Marron et al. [30,

31] proposed a modified PF, which is called XPFCP for tracking multiple bodies which

modeled multiple objects in one multimodal distribution. This method could

Figure 1.2: Pedestrian gap acceptance characteristics, from Brewer et al.

be useful to track multiple unseen pedestrians with a small modification because their

work can only track objects that were observed previously.

On the other hand, there are currently two shortcomings of available methods in

the literature. Firstly, the possible reasons of humans’ successful driving under

occlusion and limited visibility conditions has been mostly neglected. Furthermore,

little attention has been paid to scenarios in which GPS accuracy is degraded due to

buildings, bridges, trees, etc. [32]. None of the methods has taken into account the fact

that the contextual information could be incorporated into the prior knowledge;

hence, the information can be utilized to assess the risk for occluded unseen

pedestrians. In addition, the literature is deficient in formally defining the limitations

of vehicles in terms of safety, and comfort. For example, the distance to stop without

a compromise on safety and the distance to stop without a compromise on

Figure 1.3: An example of environmental cues and how a human driver would easily

decide a similar scheme for emergence probability

passengers’ comfort change with different weather conditions. Consequently, we

believe that considering the limits of safety and comfort for different conditions will

improve robustness and capabilities of ADS.

1.2.4 Contribution of This Thesis

This thesis focuses on whether only contextual information without localization

information can further improve the performance and achieve a human-like behavior.

For example, given that the ego vehicle is the red car in fig. 1.3, could we design a

method that can use visible pedestrians and parked cars to assess emerging

pedestrians’ probability to extract an optimal driving policy? The main contributions

are:

• Using contextual information to assess the risk of emerging pedestrians from

occlusions, then utilize this information to extract the driving policy

• Pedestrian emergence probabilities are calculated using a modified logistic

regression curve on the contextual cues

• The proposed controller utilizes the emergence probabilities by using three

driving modes, converted into a structured finite state machine (FSM) to provide

comfortable and safe driving.

1.3 Outline of the Thesis

This thesis proceeds as follows: in Chapter 2, a hypothetical problem is defined to

serve as a mathematical background to define the interactions with certain

assumptions. In this hypothetical problem, there is an object and multiple dynamic

obstacles that share the same path; the formulas and constraints for the object to

navigate on the road without any accidents are carefully defined.

Later, the formulas and constraints are utilized to design the proposed method for

the occluded/unseen pedestrian scenarios. The proposed method uses Bayesian

filtering, a meticulously designed FSM, and an optimization scheme to extract the

driving policy. Moreover, the design of the proposed method is explained in detail.

Secondly, in Chapter 3, the proposed method’s evaluation and the results are

demonstrated and interpreted carefully. Finally, in Chapter 4 final remarks and

limitations of the study are discussed vigorously, and the thesis is concluded with

some possible future directions.

Chapter 2: Theoretical Background and Methodology

This chapter formulates the hypothetical problem. Formulation of the

hypothetical problem is useful to define the deterministic limits to avoid collisions.

Later, the chapter introduces the problem of interest, namely avoiding emerging

pedestrians from occlusions, stochastically; the deterministic limits are utilized in this

part. The chapter proceeds with risk assessment using contextual information to the

ego vehicle. After that, the chapter explains the model used for the agents, i.e., the ego

vehicle, the pedestrians. The assumptions used in this work are stated. Finally, the

chapter concludes with the utilization of risk assessment in a control architecture and

provides the design of the controller in the proposed method.

2.1 Theoretical Background

2.1.1 Mathematical Definition of the Problem

Assume that there is an object moving along its path, whose longitudinal velocity

is controllable and denoted by v, whose visibility range is denoted by rvisible, and whose

viewing angle is denoted by α. Furthermore, assume that there are also dynamic

obstacles that can walk only in the lateral direction; the moving direction of the

obstacles is perpendicular to that of the object. Therefore, the obstacles and the

Figure 2.1: An object moving along the path with possible options: either with or

without static obstacles which limit the visibility of the object

object should use the same path; however, the object should not collide with any

obstacles.

In order to define the limitations for collision avoidance, assume that at t = t−, the

object cannot observe any dynamic obstacle, i.e., there is no obstacle inside the

object’s visibility region. Provided that the obstacles move with a constant velocity

and assuming that the behavior of the obstacles is not influenced by the object, the

current possible locations of the obstacles that would collide with the object can be

found. Now, assume that the object can detect the obstacles whose center of gravity

(CoG) is inside the visibility region. Then, an example scenario in Fig. 2.2 can be used

to define the rules and limitations of the object. The object has the following

limitations:

• It has point-mass model dynamics so it obeys the rules of Newtonian dynamics

for a point-mass object

• It has a certain delay to identify the obstacles after the very first instance of

obstacle detection; this delay is denoted by tsense

Figure 2.2: At t = t+, there could be some dynamic obstacles that are at the boundary

of detection

• It has a certain delay to actuate any control to alter its velocity using

acceleration/deceleration; this delay is denoted by tsystem delay

• It cannot realize any acceleration/deceleration value instantly after the delay

tsystem delay, but the acceleration ramps up in a certain time before it reaches the

steady value, this amount of time is denoted by tramp

2.1.2 Deterministic Limitations to Avoid Collisions

A collision occurs the object’s future trajectory and at least the future trajectory of

one obstacle coincide for the same instance. A collision can be categorized by the point

of contact at the instance of collision; either the obstacles can hit the object from its

sides or the object can hit the obstacles. The latter category is considered in this thesis

since such collisions can be avoided by the actions of the object. The object can collide

with an obstacle either:

• is already inside the expected path, and is to stay on the path,

Figure 2.3: An obstacle which is inside the object’s expected path, and which is to stay

inside the expected path

• or the obstacle can be outside of the expected path and is to enter the expected

before the object arrives.

In the first case, as can be seen in Fig. 2.3, the object’s limitation is its visibility

range, rvisible. Assume that the object cannot realize arbitrary accelerations instantly,

but tramp seconds is required to reach the steady deceleration: For t ∈ [0,tramp]:

(2.1)

(2.2)

(2.3)

Then, at t = tramp:

(2.4)

(2.5)

(2.6)

Figure 2.4: An obstacle which is outside of the object’s expected path, and which is to

enter inside the expected path

Then, total distance the object travels before stopping from the instance of actuating

brakes:

(2.7)

Finally, including the time to sense obstacles, tsense, and the time to actuate the brakes,

tsystem delay, the total distance that the object travels before stopping is:

(2.8)

Note that, af is a positive acceleration value which is the absolute value of the steady

braking deceleration of the object. The object should always ensure that rvisible >

dstop.

In the second case, as can be seen in Fig. 2.4, the object’s limitation is its visibility

region. Assume that the lateral distance between the CoG of the object and that of an

obstacle is denoted by dp, and the longitudinal distance between the frontal structure

of the object and the CoG of an obstacle is denoted by d0; furthermore, assume that

the width of the object and an obstacle are denoted by wobject, wobstacle and the lengths

are denoted by lobject, lobstacle respectively.

Using (2.1)–(2.8):

The fact that there is a region of danger in which the obstacles outside of this region

do not pose a threat to the object can be deduced from (2.9). Additionally, the danger

of collision is a smaller subset of the aforementioned region in which condition in

(2.10) is satisfied.

2.1.3 Moving to the Real Life Problem: Avoiding Emerging

Pedestrians from Occlusions

The limitations and definitions introduced and the formulas obtained in Section

2.1.1 and 2.1.2 is useful to formulate the proposal to avoid emerging pedestrians from

occlusions. The terminology used in Section 2.1.1 and 2.1.2 should be revised so as to

convert the hypothetical problem into a real-life problem:

• the object to the ego vehicle,

• dynamic obstacles to pedestrians (or VRUs in general),

• static obstacles to parked cars, trees, or buildings.

In order to proceed to risk assessment, rewrite the two conditions for a collision,

namely (2.9) and (2.10), with the new terminology:

(2.11)

Notice that a new variable dsafe is introduced since the ego vehicle is expected to keep

a safe distance from pedestrians provided that collision avoidance is not necessary,

but the ego vehicle should wait for the pedestrians on the road to cross.

Any pedestrian who is do meters away longitudinally, and dp meters away laterally,

with lateral velocity vped satisfying (2.11) poses a risk to the ego vehicle. However, as

long as (2.12) is not satisfied, the risk could be averted.

There may exist other possible definitions or formulations of the pedestrian

collision avoidance. Nonetheless, the possible definitions or formulations can be

converted into the current problem formulation via a transformation.

2.1.3.1 Vehicle Controllability Limitations and Risk Zones

Considering the variable that the ego vehicle has control over, the safety of driving

is related to (2.12). It is assumed that tsense and tsystem delay are variables that the ego

vehicle cannot alter, but only measure or estimate; by contrast, it is also assumed that

vego, af, and tramp are variables that the ego vehicle can control within their physical

limits. The limitations for the maximum possible deceleration is provided in Appendix

The control of these variables can be categorized in three aspects, namely, comfort,

safety, and efficiency. In fact, there may exist various driving policies which may

prioritize one or two of these aspects. However, in a successful policy, these three

aspects should be considered optimally without neglecting any of them. The optimal

consideration should minimally care for the passengers of the vehicle, and ideally for

all road users that the vehicle interacts. Let us denote the positive value for maximum

deceleration of the ego vehicle as amax and the minimum possible time to reach a

steady acceleration/deceleration value as tramp min, and the maximum velocity that ego

vehicle is legally allowed to drive as vego max. Let us also assume that these limitations

are constant values.

Provided that the ego vehicle is in a situation in which it is required to ensure

safety, it is assumed that the ego vehicle will try to reach the steady deceleration of

amax with its limiting time tramp min, let us assume that under this policy the earliest

distance that the ego vehicle can stop is dstop min. On the other hand, when there is no

indication of danger, it is assumed that the ego vehicle will try to reach vego max. On the

other hand, to define an optimal policy in terms of comfort, the experience of comfort

of a passenger inside a car should be investigated. The common sense and personal

experience can be used to undertake the investigation; the passengers are able to

experience both the force and the change in force exerted on their bodies. From

Newtonian mechanics for a constant mass of body, the force exerted on the body is

defined in (2.13):

(2.13) (2.14)

(2.15)

where j denotes the derivative of acceleration in

(2.14) and (2.15), which is also known as jerk. Since both acceleration and jerk

perceived by the human body similarly in any direction, the magnitude of acceleration

and jerk is the main factor for the comfort.

Therefore, it can be assumed that an acceleration value between [−acomfort,acomfort]

values and a jerk value between [−jcomfort,jcomfort] are considered to be comfortable.

Determining these values and the values used in the simulation are provided in

Section 3.1.

Here, the distance that the ego vehicle can stop when its positive value of

deceleration is denoted by acomfort; the time to reach that deceleration value without

exceeding comfortable acceleration or jerk value is denoted by tramp comfort. Let us

assume that under this policy the earliest distance that the ego vehicle can stop is dstop

comfort. The ego vehicle can be considered to have imaginary zones, so-called risk zones

as in Fig. 2.5. Assuming that dstop min ≤ dstop comfort, and a reference frame whose origin is

at the CoG of the frontal structure of the ego vehicle is used to define the risk zones,

the longitudinal ranges of the zones are as follows:

• Danger Zone : [0,dstop min]

• Discomfort Zone : [dstop min,dstop comfort]

• Safety Zone : [dstop comfort,rvisible]

This suggestion to use risk zones can be viewed as a geometrical interpretation of

the capabilities of the ego vehicle. Nevertheless, these three zones may submerge or

disappear completely with changing velocity. For example, given a scenario

demonstrated in Fig. 2.5, the ego vehicle could avoid a collision with all of the

pedestrians except the leftmost one was he or she to step on the road before the ego

vehicle can pass through. Therefore, if it is physically possible for the leftmost

pedestrian to reach the lane at the bottom before the ego vehicle, the ego vehicle

cannot avoid the collision with longitudinal control.

Figure 2.5: Given the limitations of the ego vehicle, the ego vehicle can safely and

comfortably yield to any pedestrians crossing inside the Safety Zone, whereas the ego

vehicle is required to compromise on the passengers’ comfort in order to yield to the

pedestrians inside the Discomfort Zone. Finally, it is not possible to yield or stop before

hitting a pedestrian inside the Danger Zone (Zone sizes are not to scale)

2.1.3.2 Ensuring Comfort for the Passenger inside the Ego Vehicle

Considering the comfort of the passengers and meeting the requirements is crucial

once the safety is ensured. In Section 2.1.3.1, the physical limitations of the vehicle for

the distance traversed before coming to a stop after the instance the braking action

are provided, namely dstop min and dstop comfort; and the disturbances to the comfort of

the passengers are defined. These concepts along with the work of Bae et al. [33], in

which the the lateral and longitudinal acceleration and jerk values are categorized in

terms of comfort limits, can facilitate determining the limitations to the acceleration

and jerk values under different conditions; the details are provided in Appendix A.

Using the reason the following remarks can be deduced:

• Any risk within the safety zone could be averted without compromising the

comfort; therefore, no precautious action is required.

• The maximum risk inside the discomfort zone determines the appropriate

reference speed. Moreover, since a risk that will actually happen inside this zone

is going to disrupt the comfort already, the deceleration action should be

actuated without violating the comfort criterion.

• Similarly, the maximum risk inside the danger zone determines the appropriate

reference speed. However, a risk that will actually happen inside this zone is

going to have consequences (e.g., injuries due to collision); therefore, the

deceleration action should be actuated with relaxed comfort criteria (i.e.,

aggressive driving region in Fig. A.1).

2.1.3.3 Designing the Controller

After choosing the appropriate policy to avoid pedestrians, a controller is required

to actuate the control actions. There are several controllers in the literature, such as

LQR controllers, PID controller sliding mode controllers, MPC controllers, etc., and the

topic is well-studied for point mass objects. Each controller scheme has certain

advantages and disadvantages; for example, an LQR controller optimizes a chosen

unconstrained cost function whereas an MPC controller optimizes a chosen

constrained cost function. To limit the jerk and acceleration to ensure comfort, an MPC

controller could be useful. However, quadratically constrained optimization problems

are generally NP-hard whereas the unconstrained problems are solved in polynomial

time [34]. Furthermore, with a small modification, an unconstrained optimization

problem can be converted into a constrained one. Starting with the vehicle model and

the traditional LQR optimization scheme:

x˙t = f(¯xt,u¯t) (2.16)

where xt is the state vector at time t, ut is the control vector at time t, and ˙xt is the

derivative of the state vector at time t. If we use a linear model, then the equation

becomes:

x¯˙t = Ax¯t + Bu¯t (2.17)

Specifically, since we use the point mass model, and discretize the dynamics we have

the following equations:

vk+1 = vk + ∆t ∗ ak

(2.18)

dk+1 = dk − ∆t ∗ vk

(2.19)

where vk is the velocity of the ego vehicle, ak is the acceleration value of the ego vehicle

dk, in (2.19), is the lateral distance to the imaginary line tangent to the close side of

the visible pedestrian to the vehicle, and also perpendicular to the ego vehicle’s

direction. Using the vector notation again with the discrete time equations:

(2.20)

(2.21)

(2.22)

The state-space when there are at least one visible pedestrian is represented by

(2.22); since the ego vehicle is expected to yield, the distance between the closest

pedestrian and the vehicle is one of the states. By contrast, if there are no visible

pedestrians to be yielded, then the state-space is represented by (2.18). Then, to use

LQR controller, the quadratic cost function is defined as follows:

ck = (¯xk − x¯ref)T Q(¯xk − x¯ref) + ¯uTk Ru¯k

(2.23)

T−1

J = Xck

(2.24)

k=0

(2.25)

(2.26)

)

(2.27)

where K is the Kalman gain, and the optimal control is obtained by minimizing the

cumulative cost J provided that appropriate Q and R matrices are selected.

Moreover, the reference state for cruising with a constant velocity is described in

(2.25), whereas the reference state for yielding to a visible pedestrian on the road is

described in (2.26).

However, one shortcoming of the current state-space modeling is that it is not

possible to limit the jerk since the traditional LQR is not a constrained optimization

formulation. In addition, in the case of implementing this proposal to a real-world

vehicle, a traditional LQR controller will generate unrealistic and high frequency

control actions due to noisy measurements resulting in fluctuations, which the

controller will try to compensate constantly. High-frequency modes of acceleration

can be very uncomfortable, and it may injure the passengers. The equations of LQR

optimization can be revised with the small modification to limit the jerk:

(2.28)

(2.29)

(2.30)

where the state-space for yielding with limited jerk is described in (2.29), and the

state-space for cruising with limited jerk is described in (2.30). Since the jerk of a

vehicle is not a realizable control input, the implementation of the LQR controller with

the modification will be similar; the controller will actuate the control input ak

obtained from (2.29) or (2.30).

Finally, we have chosen appropriate Q and R matrices for both the cruising and

yielding and computed the resulting full-state-feedback coefficients K as follows:

2.1.3.4 Occluded Pedestrian Emergence Probability

The collision-free policy for the ego vehicle could be computed were all the

variables to be known. For example, the body dimensions, the velocity and the

location of an unseen pedestrian are unknown. Using the probability theory, a

stochastic policy can be determined that minimizes the collision risk under given

prior assumptions and the history of observation. The contextual information, such

as the parked cars, the presence of a crosswalk, and the presence of pedestrians, can

be utilized to assess a pedestrian distribution in occluded regions. In this thesis, the

probability density function (PDF) of the pedestrians, who might emerge from the

occlusions, as a function of the longitudinal axis at time t is chosen to model the

uncertainties; the probability function is denoted by fX¯(¯xt). However, that probability

function can only be observed, that is, that function can only be accessed as the

conditional

Figure 2.6: The proposal should utilize visible cues to estimate the emerging

pedestrians’ distribution along the longitudinal axis from occluded areas similar to

what human drivers do, then this risk assessment is used to extract the driving policy

PDF, as described in in (2.31), of emerging pedestrians given the current observations

which is denoted by vector ¯zt.

fX¯(¯xt|z¯t) is distribution of r.v. x conditioned on ¯zt (2.31)

where the observation vector ¯zt can be considered as the contextual information

available at time t, e.g., the number of the parked cars equal to N, the distance to the

crosswalk is d0. For example, the pdf of emerging pedestrians given the contextual

information estimated by the ego vehicle might be as in Fig. 2.6.

2.1.4 Bayesian Filtering

A Bayes filter can be utilized to update a probabilistic function with received

observations. Specifically, a Bayes filter is very useful for applications that it is

possible to collect information to understand the environment better. It is a

mathematical tool that gives the best estimate for the current observation history and

the prior knowledge about the probabilistic variable. This best estimate is also known

as belief or belief state. Let us define the observation vector from the initialization to

time t as ¯z1:t, and the actual state of the pedestrians as xt. Belief or belief state is

defined in (2.32):

Bel(¯xt) = p(¯xt|z¯1:t)

(2.32)

Bel−(¯xt) = p(¯xt|z¯1:t−1)

(2.33)

2.1.4.1 Belief Update

As it is discussed before, belief or belief state is the best answer mathematically to

the question of “given the observation history of ¯z1:t, what is the best estimate on the

actual states?”. This question is formulated as two steps in Bayes filters assuming that

the environment satisfies the Markov assumption, i.e., the current states includes all

necessary information to interpret the current state:

Prediction: (2.34) predicts the current state without the observation using the

prior knowledge on the dynamics, p(¯xt|x¯t−1).

Correction: (2.35) corrects this prediction by considering the prior knowledge on

the measurements, p(¯zt|x¯t).

Bel−(¯xt) = p(¯xt|x¯t−1)Bel(¯xt−1)dxt−1

(2.34)

Bel(¯xt) = αtp(¯zt|x¯t)Bel−(¯xt)

(2.35)

where αt is the normalization constant to ensure that belief is a probability function

(i.e., the integral of belief over the longitudinal axis is equal to 1). Note that, if there is

no prior information on the distribution of the pedestrians for initialization, belief

should be chosen a uniform distribution.

2.1.4.2 Belief Representations

Some popular types of Bayes filters can be categorized and summarized as in Fig.

2.7. Let us elaborate on each one of them:

Kalman filters have a wide application area that this filter type is the most common

practice for Bayes filters [35]. The fact that multiplying, convolving, or adding two

Gaussian functions creates another Gaussian function facilitate computational

efficiency for Kalman filters. However, computational efficiency advantage is only

available with the assumption that all random variables and noises in the system are

Gaussian distributed random variables, which may be an inaccurate assumptions for

some situations. Although this assumption limits Kalman filter’s generality, together

with extended Kalman filters (EKF), Kalman filters can estimate both linear and

nonlinear dynamics of a system. For example, Kalman filters are the common practice

for distance measurement sensors and GPS tracking systems.

Multi-hypothesis tracking (MHT) extends the unimodality of Kalman filters by

assuming the belief as a mixture of Gaussian distribution (i.e., multi-modal belief

representation) [35]. Each hypothesis constitutes a unimodal distribution such that a

Kalman filter tracks this hypothesis. Then, each hypothesis is weighted based on its

accuracy and success to predict the data.

Figure 2.7: Different types of Bayes filters

Grid-based approaches discretize states spatially and assign a single value of belief

for each grid in the belief space. This piecewise segmentation allows grid-based

approaches to assume arbitrary belief distributions. However, computational

complexity is not scalable with the increasing number of grids. Therefore, these

approaches trade-off grid resolution and accuracy with computational complexity.

Particle filters (PF) which are also called sequential Monte Carlo (SMC) methods,

models the belief by sampling the distribution, weighted according to the distribution

itself [36]. Particle filters are much more efficient compared to grid-based methods as

they re-distribute the resources (hypotheses) to the best estimate using a procedure

called sequential importance sampling (or re-sampling) after making the observation.

Their capability to represent arbitrary distributions without discretizing the

statespace makes particle filters a common choice where Kalman filters cannot

accurately represent the state-space.

2.1.5 Tracking Obstacles Using Bayesian Filtering

Let us utilize the belief and Bayes filtering in order to estimate the distribution of

the dynamic obstacles. Let us assume that ¯xt represents the position and the velocity

of an obstacle, we will further generalize and redefine ¯xt so as it to represent the

position and the velocity of multiple obstacles. From (2.34), we know the belief

update requires the prior knowledge of transition model p(¯xt|z¯1:t). Let us define the

transition model with a simple kinematic model in (2.36):

(2.36)

where ψ¯ is the process noise vector. Furthermore, p(¯zt|x¯t) is the observation model

which is also required to be a prior knowledge, which is defined in Equation (2.37):

(2.37)

where ¯ϕt is the observation noise vector. And note that, we assumed that the object

can only observe the position of the obstacle whereas observation matrix could be an

I4×4 matrix should the object have a sensor to read the velocity of the obstacle.

2.1.6 Prior Probability Distribution Based on Reasoning

2.1.6.1 Presence of Crosswalks

Under the influence of just a crosswalk, one may expect to observe that

pedestrians mostly use the crosswalk to cross the street whereas there still might be

pedestrians that cross outside of the crosswalk, but we expect this behavior to be less

likely as the distance longitudinal distance between the pedestrians and the

crosswalk increases. This distribution can be modeled using a Gaussian distribution,

and the pdf of the pedestrians’ distribution for the example in Fig. 2.8 is as follows:

Figure 2.8: The occluded pedestrian distribution considering only the presence of the

crosswalk

(2.38)

where µcrosswalk is the longitudinal distance between the center of the crosswalk

and the ego vehicle, and the choice of σcrosswalk could vary based on the statistical

behavior of the pedestrians.

2.1.6.2 Presence of Pedestrians

The presence of pedestrians is another indication that it is more likely to observe

emerging pedestrians from occlusions. Even if they stand still, one may expect to

observe that the presence of pedestrians around a pedestrian is more likely. The idea

behind this assumption is that observing a standing pedestrian both means that this

pedestrian may start walking, and there could be other pedestrians occluded/unseen

Figure 2.9: The occluded pedestrian distribution considering only the presence of

other pedestrians

around that pedestrian. This distribution can be modeled using a mixture of Gaussian

distribution, and the pdf of the pedestrians’ distribution for the example in Fig. 2.9 is

as follows:

(2.39)

where µiped is the longitudinal distance between the CoG of ith pedestrian and the

ego vehicle, σped is the standard deviation to incorporate the sensor error. The general

formula for N number of pedestrians is as follows:

(2.40)

2.1.6.3 Presence of Parked Cars

A study on human behavior at a signaled crosswalk states that from their analysis,

”group size and traffic density was related to safer behaviors and the nearby presence

of parked vehicles to riskier ones” ([37]). In other words, although one cannot directly

constitute a correlation between the presence of parked cars and the presence of

pedestrians, pedestrians tend to be more distracted, w when they are near parked

cars. Therefore, one can increase the distribution of the occluded pedestrians around

parked cars in order to increase caution since the pedestrians near parked cars are

less likely to behave safely. One can represent this distribution with a mixture, and the

pdf of the pedestrians’ distribution for the example in Fig. 2.10 is as follows:

(2.41)

where µiparked cars is the longitudinal distances between the ith parked car’s furthest

point to the ego vehicle longitudinally and the ego vehicle, σ value should be chosen

either using statistics or heuristics. The general formula for N number of parked cars:

(2.42)

Figure 2.10: The pedestrian distribution only with the presence of parked cars

2.2 Methodology

2.2.1 Risk Assessment

2.2.1.1 Static Risk Assessment using Logistic Regression Prior

Risk assessment is the most crucial part and the main contribution of this work.

Using the prior knowledge of pedestrian distribution and Bayesian inference as

discussed in Section 2.1.6, the PDF and the CDF can be obtained; for example, the

functions can be obtained for a scenario in which the crosswalk from Fig. 2.8, the

standing pedestrians from Fig. 2.9, and the parked cars from Fig. 2.10 are present at

the same time. Then, the posterior of the total pedestrian emergence probability along

the longitudinal axis given that the observation of the pedestrians, the parked cars,

and the crosswalk is demonstrated in Fig. 2.11a. In order to reduce the computational

cost of the estimation of the posterior probability, a collection of piecewise

(a) The contribution of the observation of (b) The cdf of the total pedestrian emergence

pedestrians, parked cars, and the crosswalk to probability along the longitudinal axis, and the

pdf and cdf of the total pedestrian emer- multiple sigmoid functions (1/(1+e−wx)) with gence

probability varying w from the case in Fig. 2.11a

Figure 2.11: An example distribution of emerging pedestrians from occlusions and a

possible expansion of the distribution with piecewise functions

weighted sigmoid functions can be utilized to estimate the posterior probability. First,

let us define the weighted sigmoid function as:

(2.43)

which has two horizontal asymptotes at lim = 0 and lim

Then, let us demonstrate the similarity of the example CDF and some weighted

sigmoid functions in Fig. 2.11b. In fact let us also assume that the variable x is the

observation vector at time t z¯t and the variable w is the weight vector of the

observation ¯w. Since the proposal uses possible visible cues, e.g., the number of

parked cars, the number of visible pedestrians, the distance to the crosswalk, the

distance to the closest parked car, and the distance to the closest visible pedestrian;

the probability estimation becomes:

Figure 2.12: An example of the probability distribution of occluded pedestrian

emergence

p(Pedestrian emergence risk (2.44)

where ¯zt = [1,n1,n2,d1,d2,d3]T is the observation vector. The definition of the variables

inside the observation vector is given in Table 2.1. A possible pedestrian emergence

probability is demonstrated in Fig. 2.12.

Table 2.1: Elements of ¯zt in (2.44)

Variables

Explanation

the density of parked cars

the density of visible pedestrians

the distance to the crosswalk

the distance to the closest parked car

the distance to the closest visible pedestrian

All measurements are normalized which is achieved by dividing the distance

measurements by rvisible and the quantity measurements (i.e., number of pedestrians,

the number of parked cars) are divided by a predefined maximum number. This way

the number measurements are treated as density information which gives more

useful information than the number information. Furthermore, if any quantitative

measurement such as the number of parked cars is absent, it is considered to be 0;

whereas, if any distance measurement such as the distance to the crosswalk is absent,

it is considered to be 1 which is the normalized value for the farthest distance. The

vector w¯ in (2.44) is the weights of the visible cues, which was chosen by heuristics.

2.2.2 Collision Avoidance

Risk assessment, p(Occluded pedestrian emergence|z¯t, is calculated as explained

in Section 2.2.1; in addition, provided that the ego vehicle determines its risk zones,

and assesses the current risk, then it is also necessary to determine the collision risk,

p(Collision|z¯t). Note that, since the quantity measurements such as the density are

discrete, it causes sudden discrete changes at the critical instance of observing or

losing sight of a parked car or a pedestrian. One possible solution could be to divide

the function into sections, that is, using thresholds to define regions in the function,

as in Fig. 2.13, so as to utilize the function in a control architecture.

Then, one possible control architecture could be to use an FSM with distinct states

for each region in the function. As the risk decreases, the ego vehicle could be more

encouraged to drive the speed limit while it should slow down to a proportion of the

speed limit in a risky situation. Let us name the FSM states for each region the function

from Fig. 2.13:

Figure 2.13: A possible utilization of the risk assessment in Fig. 2.12

• For the green region, it is called Normal Drive in which the ego vehicle is given

the reference velocity of the speed limit, vspeed limit,

• For the yellow region, it is called Steady Drive in which the ego vehicle is given

the reference velocity of some proportion of the speed limit, α1vspeed limit,

• For the red region, it is called Cautious Drive in which the ego vehicle is given the

reference velocity of another proportion of the speed limit, α2vspeed limit.

Note that as the ego vehicle should drive slower for the values of function within the

red region, 0 < α2 < α1 < 1. Also considering a state for yielding a pedestrian when

necessary, and for engaging in maximum braking to avoid collision under dangers, the

FSM architecture in Fig. 2.14 is constructed with state transition conditions are given

in Table 2.2.

The speed limit vspeed limit is calculated as:

vspeed limit = min{Legal speed limit,Safety speed limit} (2.45)

Figure 2.14: The proposed FSM to utilize the quantified risk

Table 2.2: State Transitions for the FSM in Fig. 2.14

State Transition

Explanation

p(Pedestrian emergence risk|z¯t) > lsteady

p(Pedestrian emergence risk|z¯t) ≤ lsteady

p(Pedestrian emergence risk|z¯t) > lcautious

lsteady ≤ p(Pedestrian emergence risk|z¯t) ≤ lcautious

A visible pedestrian to be yielded

No visible pedestrian to be yielded

No feasible solution to avoid collision

A feasible solution to avoid collision

where safety speed limit is calculated from the roots of the following in equality:

Then the roots of the inequlity in (2.46) is given as:

(2.47)

(2.48)

The root ∆1 is negative (given that ∆1 < ∆2), as can be seen from (2.47); therefore, safety

speed limit is calculated using ∆2. Furthermore, the safe speed limit is determined by

the choice of af. This choice is arbitrary within the limits of this variable; however, one

reasonable choice would be to both consider safety and comfort and choose a speed

which the ego vehicle could anticipate the risks and slow down or stop before a

pedestrian, a speed such that the condition dstop min ≤ dstop comfort ≤ rvisible

is satisfied.

2.2.3 Controller Design

The interpretation of risk inside different risk zones should be different because a

smaller risk of emergence in danger zone should be approached much more carefully

than the same level of risk of emergence in discomfort zone. In other words, the

alertness to the pedestrian emergence probability inside danger zone should be much

higher than the alertness to the pedestrian emergence probability inside discomfort

zone. Therefore, the controller should assign different threshold values to the

FSM architecture per risk zone. The visualization of this process is demonstrated in

Fig. 2.16.

Combining all, the ego vehicle will observe the environment from its sensors to

assess the risk of the emergence of occluded pedestrians. Then, this information is

going to be interpreted differently per risk zone, and this interpretation is going to be

Figure 2.15: Proposed controller architecture

outputted as a reference state. Finally, a longitudinal controller is going to choose the

appropriate control action to reach the reference state within decided control limits.

The flow diagram of the proposed controller architecture is given in Fig. 2.15.

2.2.4 Overall Proposed Control Algorithm

After considering the emergence risk in Section 2.2.1 and the collision avoidance

in Section 2.2.2, and the controller design in Section 2.2.3, now these should be

combined in a logical flow, an algorithm to be executed. As it is explained previously

that the ego vehicle does not require to take precautious action against any

pedestrians entering the expected path from inside the safety zone, the algorithm’s

look-ahead distance should be dstop comfort. Then, it should calculate the risk in the

future by considering visible cues within a predefined window; furthermore, it should

have a spatial resolution ∆d. The maximum risk per zone (i.e., danger zone, discomfort

Figure 2.16: Risk interpretation of proposed method inside different risk zones, the

combination of risk assessment and collision avoidance

zone) will be compared against the thresholds to choose the appropriate

deceleration/acceleration limit alimit and jerk limit jlimit with the appropriate FSM state

in the order the ego vehicle to reach the reference velocity vref within the chosen limits.

In addition, as it is crucial to determine the deceleration limitations, which are limited

by the friction coefficient between the tires and the road, a look-up table can be used

(Table A.1) provided that the ego vehicle could determine the current weather

conditions.

The algorithm for the proposed controller is as follows where the predefined vari-

ables are tsense, tsystem delay, ∆d, ∆t, lcautiousdanger , lsteadydanger, lcautiousdiscomfort, lsteadydiscomfort,

window size,

system and measurement noises, ψ¯t, ¯ϕt, the current state variables for the ego vehicle

x¯t,ego and the observation ¯zt:

Algorithm 1 The proposed algorithm

1: function proposedController

2: amax ← µroad ∗ g, st+1 ← snormal drive

3: dstop min, dstop comfort ← (2.12)

4: if A visible pedestrian is to be inside the path then

5: if TTC < TTCstop then

6: st+1 ← semergency

7: else

8: st+1 ← syielding

9: else

10: current state ← danger

11: max risk ← 0, current risk ← 0

12: repeat

13: # Update current risk zone

14: if d > dstop min then

15: current state ← discomfort

16: max risk ← 0

17: # Check the neighboring visible cues

18: current risk ← from (2.44)

23: ifthen

24:

25:

else if max risk > lcautious then

26:

st+1 ← scautious drive

27: else if current state = discomfort then

28: Set ldiscomfort, lsteady, alimit and jlimit

29: if max risk > lsteady then

30: st+1 ← ssteady drive

31: else if max risk > lcautious then

32: st+1 ← scautious drive

33: d ← d + ∆d

34: until d ≥ dstop comfort

35: return st+1

Chapter 3: Evaluation and results

3.1 Evaluation

In order to evaluate the proposal, a straight road is chosen. Furthermore, with

respect to the pedestrian frequency the scenarios are divided into three:

• Suburban scenarios (sc1) where there are one or two pedestrians, one or two

parked cars, and a crosswalk.

• Mildly crowded urban scenarios (sc2) where there are multiple parked cars,

multiple pedestrians and a crosswalk.

19:

if max risk < current risk then

20:

max risk ← current risk

21:

if current state = danger then

22:

Set lcautious, lsteady, alimit and jlimit

• Very crowded urban scenarios(sc3) where the parking slots are full, and there are

multiple pedestrians and a crosswalk.

The summary of all scenarios can be seen in Table 3.1 This way it is going to be

possible to observe the strengths and weaknesses of the proposal in various possible

scenarios. Specifically, the mildly crowded urban scenarios and the very crowded urban

scenarios are the scenarios to be addressed the most successful since the GPS

localization accuracy and precision deteriorates in urban locations[32].

In addition, some baseline controllers should be introduced in order to compare

the proposed method. Since the proposal is formulated upon three aspects (i.e.,

comfort, Table 3.1: Scenarios used in the Simulation Environment

Scenario

Road Type

Number of parked cars

Crosswalk

Suburban scenario (Scenario 1)

Straight Road

Several (At most 2)

Yes

Mildly crowded street (Scenario 2)

Straight Road

Multiple, all parking slots are full

Yes

Very crowded street (Scenario 3)

Straight Road

Multiple, but not all parking slots are full

Yes

safety, and efficiency), some of these aspects could be disregarded in the baselines to

demonstrate the outcome. Following this idea it is decided that the proposal is to

compared against four baselines:

• Baseline1 (B1): This baseline is occlusion-unaware, and drives the legal speed

limit of the road (30km/h). It yields to the pedestrians that are on the road will

likely enter its expected path.

• Baseline2 (B2): This baseline is also occlusion-unaware, and drives two-third of

the legal speed limit of the road (20km/h). It yields to the pedestrians that are

on the road will likely enter its expected path.

• Baseline3 (B3): This baseline is also occlusion-unaware, and drives one-third of

the legal speed limit of the road (10km/h) only if it observes a crosswalk which

closer than a specific distance; otherwise, it drives the speed limit. It yields to

the pedestrians that are on the road will likely enter its expected path.

3.1.1 Metrics

Defining metrics is as important as choosing the correct baselines to compare

against. A meticulous analysis of the literature was made. Combining the extensive

survey, on important metrics to determine the driving quality, of Jahanagirova et al.

[38] and other additional safety metrics, the metrics that are the most appropriate to

evaluate the proposed method on a straight road are carefully investigated and

selected. The selected metrics are as follows:

• mt1: The total number (successful/unsuccessful) of yields is useful to determine

the pedestrian-friendliness of the methods. A method that can navigate safely

and carefully should be able to yield to more pedestrians.

• mt2: Deceleration (mean, std) is useful to determine the conservativeness of the

methods. A method that can navigate safely and carefully should be able to

resort to smaller magnitudes of deceleration as the precautious actions before

occlusions are in favor of using smaller magnitudes of deceleration.

• mt3: The total number of successful finishes is the most essential metric of

safety because it means the method cannot handle risky situations and causes

an accident if this metric is smaller

• mt4: Time of emergency braking (mean, std) is similar to mt2, but it is more

specific to the readiness of the method under test on the account of the fact that

a longer duration of emergency braking is a good indication of inability to react

unexpected dangerous situations/

The values for the comfortable jerk and acceleration values are chosen to be acomfort

∈ [−2.0,1.47]m/s2, jcomfort ∈ [−0.9,0.9]m/s3 from [33].

3.2 Simulation

The simulation environment is built from scratch in Python. The flow diagram of

the simulation is demonstrated in Fig. 3.1. To generate realistic scenarios, we have

Figure 3.1: The flow diagram of the simulation inside an episode per frame

considered several aspects of the generation of pedestrians, also the decision-making

of pedestrians. The parameters used in the simulation are demonstrated in Table 3.2.

3.1

3.2.1 Modeling Pedestrians

Pedestrians are modeled as point-mass objects with instant velocity change (the

time passed until a pedestrian accelerates to its walking velocity is negligible) and

zero acceleration since the velocity of pedestrians is very small compared to that of

vehicles. Statistical data from Saberi et al. [41] for the velocity of pedestrians also

resulted in a mixture of Gaussian distributions which can be simplified to a Gaussian

distribution with µv ped = 1.5m/s and σv ped = 0.6m/s. In addition, since not all the

pedestrians that a driver sees on the sidewalk will cross the street, we have also

considered generating some pedestrians that will not cross the street at all.

Table 3.2: Parameters used in the Simulation Environment

Parameters

Symb

Values

Road length

lroad

96 m

Lane width

wlane

3 m

Number of lanes

N/A

Simulation window size

N/A

1920×5

pixels

lroad 96 m rvisible 50 m vped

5 km/h

3.2.2 Detecting Occlusions

There could be many possible methods to detect sensor occlusions; however, in

order to focus on the proof-of-concept proposal, we have simplified occlusion

detection by making several assumptions:

• The ego vehicle can sense an object (parked cars, pedestrians, crosswalks) if the

object’s CoG is inside the visibility polygon.

• The ego vehicle can identify different objects it senses them (e.g., the ego vehicle

can distinguish the measurements from different objects).

• The ego vehicle knows its visible range rvisible, and its vision angle α.

• The maximum size of parked cars is known.

There could be some cases in which the given assumptions might be wrong.

However, for the straight road in the simulation environment, the occlusion triangle

calculation is demonstrated in Fig. 3.2.

3.3 Results

The results of the simulation, 1000 episodes per scenario, is demonstrated in

Table 3.3 and in Fig. 3.7 and3.8.

In terms of comfort, the proposed method clearly outperforms in terms its yielding

capabilities (mt1) even though all baseline methods have the exact same FSM policy

for yielding, as demonstrated in Fig. 3.7a–3.7c. Specifically, the proposed method

outperforms the baselines in sc2, by 32.46%, 10.37%, 11.00% respectively. This

clearly shows that, the proposed method is very pedestrian friendly with its

yielding capabilities.

In terms of safety, outperformance of the proposed method is clearly visible for

the number of successful finishes (mt3) out of 1000 episodes per scenario, as

demonstrated in Fig. 3.8d–3.8f, where the proposed method outperforms the

baselines in

Figure 3.2: How to determine an occlusion triangle

Figure 3.3: Total successful and unsuccessful yields (mt1) of different controllers in

mildly crowded urban scenarios (sc2)

sc2 by 51.23%, 7.64%, 15.19% respectively (performance in sc2 is demonstrated in

fig. 3.5). This metric is crucial as it is directly related to collision risk of the method.

Furthermore, the proposed method also outperforms all other baselines both in the

average deceleration (mt2) and the average time of emergency braking (mt4) which

could be interpreted as an indication of how successfully a method anticipates the

incoming risk such that it resorts to minimal deceleration value and emergency

braking (performance in sc2 is demonstrated in fig. 3.6). One important remark is that

in fig. 3.6 total number of yields per controller is different due to the fact that this

metric is only available for successful path completions. In other words, an aggressive

driving style may end up in a collision which in return means that the successful yields

in the episode will not be considered.

Figure 3.4: Deceleration values (mt2) of different controllers in mildly crowded urban

scenarios (sc2)

Figure 3.5: The total number of successful finishes (mt3) of different controllers in

mildly crowded urban scenarios (sc2)

Figure 3.6: Average time of emergency braking (mt4) of different controllers in mildly

crowded urban scenarios (sc2)

Table 3.3: Overall Performance of Proposed Controller and Baselines over 1000

episodes

Metrics

Proposed

mt1a(sc1)

(817/94)

(934/24)

(912/33)

(937/23)

mt1 (sc2)

(7482/1811)

(8980/1821)

(8929/1496)

(9911/1294)

mt1 (sc3)

(8012/1754)

(9645/1875)

(9449/1535)

(10310/1404)

mt2b(sc1)

(-2.90, 1.55)

(-1.12, 0.99)

(-1.42, 1.23)

(-1.07, 0.95)

mt2 (sc2)

(-2.13, 1.78)

(-1.31, 1.30)

(-1.05, 1.29)

(-0.79, 0.90)

mt2 (sc3)

(-1.99, 1.75)

(-1.29, 1.28)

(-1.02, 1.27)

(-0.80, 0.91)

mt3c(sc1)

898

992

966

996

mt3 (sc2)

652

916

856

986

mt3 (sc3)

664

956

870

988

mt4d(sc1)

(0.27, 0.35)

(0.08, 0.18)

(0.09, 0.22)

(0.07, 0.16)

mt4 (sc2)

(0.63, 0.57)

(0.32, 0.38)

(0.30, 0.38)

(0.11, 0.18)

mt4 (sc3)

(0.63, 0.60)

(0.30, 0.36)

(0.30, 0.39)

(0.12, 0.19)

a mt1: The total number (successful/unsuccessful) of yields

b mt2: Deceleration (mean, std) c mt3: The total number of

successful finishes d mt4: Time of emergency braking

(mean, std)

(a) mt1 in sc1 (b) mt1 in sc2 (c) mt1 in sc3

Figure 3.7: Comfort related results

(a) mt2 in sc1 (b) mt2 in sc2 (c) mt2 in sc3

(d) mt3 in sc1 (e) mt3 in sc2 (f) mt3 in sc3

(g) mt4 in sc1 (h) mt4 in sc2 (i) mt4 in sc3

Figure 3.8: Safety related results

Chapter 4: Conclusion and future work

This thesis proposed a probabilistic risk assessment and collision avoidance

method for emerging pedestrians from occlusions. First, the theoretical background

is explained to introduce some useful knowledge for the method. It is shown that the

method can be implemented together with a simple FSM controller to choose the

appropriate velocity for the ego vehicle. Furthermore, the proposal is compared

against several baselines. The method is evaluated against the baselines in three

different scenarios, 1000 episodes with randomized initial conditions per scenario

type, in the simulation environment in Python built from scratch. The method

outperformed these baselines in the predefined metrics. Since the proposal does not

rely on accurate map data or accurate and precise localization to achieve occlusion-

aware vehicle control, it could be used in which the localization sensor fidelity is low,

e.g., urban areas, and big metropolitans. Moreover, the proposed method is sensor-

agnostic, in that any sensor type that can provide the required contextual information

can be integrated with the proposed method.

On the other hand, this thesis demonstrates a possible proof-of-concept for the

proposal, and in fact have several limitations that should be overcome before moving

to a real-life implementation of this proposal. One of these limitations is the

assumption of the detection and identification of distinct pedestrians and parked cars,

or other objects without errors. Although the current technology and developments

can provide this assumption for the majority of time, there are also some cases in

which this assumption may not hold, and detection systems fail. The proposed

method might not handle such situations that the detection systems fail, and some

pedestrians and/or some parked cars are not detected; therefore, some occlusions are

missed. Another limitation is due to the simplification of the risk assessment with

another function. The weights of the function that assesses the risk are determined

heuristically, and the optimality of resulting controller have not proven. Therefore,

there may exist a better weight vector, or a better representation of the risk. Lastly,

the simplification such as assessing the risk on a straight road, assuming that the

vehicle is a point-mass object might limit the performance of the proposed method.

Although, we have considered important and realistic phenomena, i.e., the delay in

actuating the control actions and the delay in sensing objects, the aforementioned

assumptions might still deviate the implementation results from the simulated ones.

One future direction could further investigate the other possibilities to assess the

probability using other contextual information such as the age of the visible

pedestrians, the possible actions engaged by the visible pedestrians (e.g., presence of

children playing soccer at the sidewalk, presence of distracted pedestrians due to use

of cellphones or a conversation companion). Another possible direction might collect

the data from expert drivers with the given contextual information to find the

correlations between the information and chosen actions/assessed risk. Given a

collected expert driver data, the proposed method’s risk assessment function can be

trained to obtain the most fitting weights. This may improve the performance of the

method. Finally, the proposed method can be extended and developed by

relaxing/removing the assumption, generalizing the proposal for different coordinate

systems and different road architectures.

In conclusion, this thesis might interest some researchers and the proposed

method may encourage these researchers to further investigate the quite promising

and unexplored topic of ”occluded pedestrian risk assessment using contextual

information by providing a proof-of-concept for avoiding occluded pedestrians using

those information. Given the assumptions and simplified problem definitions, the

results and the performance of the proposed method may be affected; however, with

distinct possible future directions, this work can be improved and/or extended.

Bibliography

[1] W. H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[2] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[3] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[4] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[5] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent, reinforcement

learning for autonomous driving,” arXiv preprint arXiv:1610.03295, 2016.

[6] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus, “Intention-

aware motion planning,” in Algorithmic foundations of robotics X. Springer, 2013,

pp. 475–491.

[7] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[8] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[9] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[10] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[11] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[12] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[13] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian collision

avoidance system for scenarios with occlusions,” 2019.

[14] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[15] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[16] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[17] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-¨

ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[18] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[19] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[20] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[21] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,” in

2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[22] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[23] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of pedestrians

in urban environments considering formalized traffic rules,” in 2018 21st

international conference on intelligent transportation systems (ITSC). IEEE, 2018,

pp. 2704–2711.

[24] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[25] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning agents,”

in Proceedings of the 20th International Conference on Machine Learning (ICML-

03), 2003, pp. 656–663.

[26] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of pedestrian

gap-acceptance behavior at selected locations,” Transportation Research Record:

Journal of the Transportation Research Board, vol. 1982, no. 1, p. 132–140, 2006.

[27] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block pedestrian

crossings in china: gap acceptance, conflict and crossing path analysis,”

International Journal of Injury Control and Safety Promotion, vol. 19, no. 4, p. 320–

330, 2012.

[28] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[29] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[30] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[31] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[32] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[33] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[34] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[35] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[36] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[37] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[38] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[39] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[40] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[41] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[42] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001.

[43] W. H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[44] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[45] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[46] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[47] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent, reinforcement

learning for autonomous driving,” arXiv preprint arXiv:1610.03295, 2016.

[48] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus, “Intention-

aware motion planning,” in Algorithmic foundations of robotics X. Springer, 2013,

pp. 475–491.

[49] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[50] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[51] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[52] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[53] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[54] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[55] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian collision

avoidance system for scenarios with occlusions,” 2019.

[56] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[57] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[58] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[59] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-¨

ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[60] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[61] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[62] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[63] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,” in

2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[64] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[65] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of pedestrians

in urban environments considering formalized traffic rules,” in 2018 21st

international conference on intelligent transportation systems (ITSC). IEEE, 2018,

pp. 2704–2711.

[66] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[67] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning agents,”

in Proceedings of the 20th International Conference on Machine Learning (ICML-

03), 2003, pp. 656–663.

[68] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of pedestrian

gap-acceptance behavior at selected locations,” Transportation Research Record:

Journal of the Transportation Research Board, vol. 1982, no. 1, p. 132–140, 2006.

[69] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block pedestrian

crossings in china: gap acceptance, conflict and crossing path analysis,”

International Journal of Injury Control and Safety Promotion, vol. 19, no. 4, p. 320–

330, 2012.

[70] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[71] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[72] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[73] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[74] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[75] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[76] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[77] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[78] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[79] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[80] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[81] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[82] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[83] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[84] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[85] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[86] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[87] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[88] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent, reinforcement

learning for autonomous driving,” arXiv preprint arXiv:1610.03295, 2016.

[89] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus, “Intention-

aware motion planning,” in Algorithmic foundations of robotics X. Springer, 2013,

pp. 475–491.

[90] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[91] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[92] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[93] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[94] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[95] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[96] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian collision

avoidance system for scenarios with occlusions,” 2019.

[97] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[98] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[99] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[100] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[101] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[102] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[103] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[104] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[105] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[106] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[107] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[108] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[109] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[110] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[111] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[112] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[113] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[114] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[115] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[116] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[117] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[118] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[119] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[120] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[121] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[122] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[123] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[124] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[125] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[126] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[127] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[128] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[129] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[130] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[131] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[132] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[133] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[134] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[135] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[136] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[137] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[138] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[139] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[140] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[141] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[142] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[143] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[144] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[145] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[146] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[147] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[148] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[149] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[150] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[151] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[152] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[153] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[154] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[155] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[156] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[157] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[158] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[159] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[160] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[161] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[162] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[163] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[164] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[165] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[166] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[167] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[168] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[169] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[170] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[171] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[172] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[173] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[174] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[175] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[176] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[177] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[178] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[179] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[180] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[181] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[182] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[183] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[184] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[185] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[186] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[187] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[188] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[189] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[190] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[191] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[192] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[193] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[194] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[195] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[196] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[197] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[198] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[199] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[200] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[201] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[202] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[203] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[204] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[205] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[206] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[207] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[208] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[209] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[210] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[211] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[212] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[213] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[214] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[215] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[216] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[217] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[218] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[219] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[220] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[221] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[222] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[223] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[224] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[225] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[226] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[227] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[228] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[229] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[230] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[231] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[232] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[233] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[234] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[235] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[236] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[237] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[238] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[239] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[240] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[241] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[242] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[243] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[244] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[245] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[246] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[247] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[248] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[249] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[250] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[251] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[252] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[253] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[254] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[255] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[256] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[257] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[258] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[259] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[260] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[261] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[262] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[263] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[264] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[265] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[266] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[267] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[268] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[269] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[270] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[271] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[272] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[273] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[274] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[275] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[276] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[277] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[278] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[279] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[280] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[281] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[282] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[283] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[284] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[285] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[286] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[287] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[288] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[289] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[290] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[291] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[292] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[293] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[294] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[295] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[296] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[297] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[298] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[299] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[300] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[301] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[302] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[303] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[304] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[305] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[306] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[307] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[308] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[309] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[310] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[311] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[312] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[313] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[314] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[315] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[316] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[317] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[318] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[319] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[320] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[321] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[322] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[323] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[324] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[325] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[326] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[327] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[328] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[329] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[330] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[331] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[332] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[333] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[334] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[335] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[336] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[337] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[338] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[339] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[340] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[341] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[342] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[343] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[344] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[345] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[346] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[347] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[348] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[349] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[350] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[351] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[352] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[353] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[354] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[355] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[356] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[357] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[358] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[359] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[360] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[361] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[362] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[363] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[364] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[365] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[366] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[367] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[368] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[369] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[370] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[371] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[372] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[373] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[374] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[375] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[376] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[377] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[378] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[379] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[380] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[381] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[382] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[383] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[384] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[385] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[386] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[387] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[388] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[389] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[390] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[391] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[392] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[393] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[394] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[395] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[396] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[397] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[398] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[399] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[400] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[401] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[402] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[403] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[404] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[405] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[406] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[407] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[408] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[409] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[410] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[411] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[412] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[413] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[414] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[415] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[416] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[417] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[418] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[419] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[420] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[421] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[422] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[423] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[424] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[425] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[426] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[427] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[428] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[429] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[430] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[431] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

[432] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[433] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[434] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[435] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[436] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[437] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[438] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[439] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[440] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[441] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[442] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[443] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[444] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[445] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[446] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[447] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[448] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[449] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[450] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[451] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[452] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

100

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[453] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[454] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[455] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[456] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[457] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[458] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[459] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[460] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[461] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[462] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

101

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[463] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[464] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[465] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[466] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[467] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[468] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[469] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[470] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[471] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[472] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

102

[473] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[474] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[475] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[476] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[477] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[478] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[479] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[480] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[481] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[482] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

103

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[483] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[484] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[485] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[486] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[487] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[488] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[489] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[490] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[491] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[492] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[493] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

104

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[494] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[495] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[496] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[497] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[498] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[499] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[500] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[501] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[502] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[503] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

105

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[504] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[505] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[506] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[507] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[508] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[509] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[510] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[511] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[512] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[513] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

106

[514] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[515] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[516] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[517] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[518] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[519] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[520] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[521] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[522] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[523] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

107

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[524] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[525] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[526] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[527] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[528] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[529] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[530] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[531] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[532] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[533] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[534] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

108

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[535] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[536] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[537] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[538] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[539] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[540] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[541] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[542] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[543] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[544] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

109

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[545] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[546] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[547] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[548] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[549] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[550] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[551] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[552] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[553] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[554] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

110

[555] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[556] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[557] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[558] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[559] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[560] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[561] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[562] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[563] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[564] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

111

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[565] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[566] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[567] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[568] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[569] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[570] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[571] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[572] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[573] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[574] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[575] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

112

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[576] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[577] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[578] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[579] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[580] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[581] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[582] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[583] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[584] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[585] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

113

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[586] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[587] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[588] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[589] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[590] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[591] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[592] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[593] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[594] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[595] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

114

[596] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[597] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[598] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[599] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[600] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[601] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[602] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[603] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[604] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[605] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

115

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[606] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[607] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[608] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[609] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[610] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[611] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[612] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[613] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[614] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[615] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[616] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

116

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[617] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[618] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[619] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[620] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[621] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[622] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[623] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[624] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[625] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[626] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

117

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[627] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[628] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[629] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[630] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[631] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[632] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[633] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[634] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[635] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[636] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

118

[637] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[638] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[639] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[640] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[641] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[642] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[643] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[644] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[645] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[646] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

119

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[647] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[648] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[649] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[650] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[651] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[652] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[653] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[654] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[655] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[656] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[657] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

120

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[658] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[659] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[660] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[661] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[662] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[663] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[664] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[665] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[666] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[667] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

121

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[668] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[669] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[670] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[671] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[672] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[673] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[674] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[675] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[676] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[677] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

122

[678] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[679] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[680] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[681] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[682] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[683] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[684] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[685] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[686] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[687] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

123

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[688] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[689] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[690] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[691] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[692] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[693] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[694] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[695] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[696] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[697] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[698] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

124

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[699] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[700] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[701] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[702] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[703] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[704] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[705] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[706] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[707] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[708] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

125

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[709] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[710] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[711] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[712] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[713] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[714] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[715] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[716] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[717] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[718] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

126

[719] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[720] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[721] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[722] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[723] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[724] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[725] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[726] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[727] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[728] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

127

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[729] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[730] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[731] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[732] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[733] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[734] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[735] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[736] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[737] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[738] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[739] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

128

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[740] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001. W.

H. Organization, Global status report on road safety 2015. World Health

Organization, 2015, pp. ix, x.

[741] “Taxonomy and definitions for terms related to cooperative driving automation

for on-road motor vehicles.” [Online]. Available: https://doi.org/10.4271/ j3216

202005

[742] “Automated vehicles for safety,” Mar 2021. [Online]. Available: https:

//www.nhtsa.gov/technology-innovation/automated-vehicles-safety

[743] S. Lef`evre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk

assessment for intelligent vehicles,” ROBOMECH journal, vol. 1, no. 1, pp. 1–14,

2014.

[744] S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent,

reinforcement learning for autonomous driving,” arXiv preprint

arXiv:1610.03295, 2016.

[745] T. Bandyopadhyay, K. S. Won, E. Frazzoli, D. Hsu, W. S. Lee, and D. Rus,

“Intention-aware motion planning,” in Algorithmic foundations of robotics X.

Springer, 2013, pp. 475–491.

[746] S. Brechtel, T. Gindele, and R. Dillmann, “Solving continuous pomdps: Value

iteration with incremental learning of an efficient space representation,” in

Proceedings of the 30th International Conference on Machine Learning, ser.

Proceedings of Machine Learning Research, S. Dasgupta and D. McAllester, Eds.,

vol. 28, no. 3. Atlanta, Georgia, USA: PMLR, 17–19 Jun 2013, pp.

370–378. [Online]. Available: http://proceedings.mlr.press/v28/brechtel13.html

[747] S. Brechtel, T. Gindele, and R. Dillman, “Probabilistic decision-making under

uncertainty for autonomous driving using continuous pomdps,” in 17th

international IEEE conference on intelligent transportation systems (ITSC). IEEE,

2014, pp. 392–399.

[748] M. Bouton, A. Cosgun, and M. J. Kochenderfer, “Belief state planning for

autonomously navigating urban intersections,” in 2017 IEEE Intelligent Vehicles

Symposium (IV). IEEE, 2017, pp. 825–830.

[749] M. Bouton, K. Julian, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Utility

decomposition with deep corrections for scalable planning under uncertainty,”

129

in Proceedings of the 17th International Conference on Autonomous Agents and

MultiAgent Systems, 2018, pp. 462–469.

[750] M. Bouton, A. Nakhaei, K. Fujimura, and M. J. Kochenderfer, “Scalable decision

making with sensor occlusions for autonomous driving,” in 2018 IEEE

International Conference on Robotics and Automation (ICRA). IEEE, 2018, pp.

2076–2081.

[751] S. M. Thornton, F. E. Lewis, V. Zhang, M. J. Kochenderfer, and J. C. Gerdes, “Value

sensitive design for autonomous vehicle motion planning,” in 2018 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1157–1162.

[752] M. Schratter, M. Bouton, M. J. Kochenderfer, and D. Watzenig, “Pedestrian

collision avoidance system for scenarios with occlusions,” 2019.

[753] S. Hoermann, F. Kunz, D. Nuss, S. Renter, and K. Dietmayer, “Entering crossroads

with blind corners. a safe strategy for autonomous vehicles,” in 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2017, pp. 727–732.

[754] M. Lee, K. Jo, and M. Sunwoo, “Collision risk assessment for possible collision

vehicle in occluded area based on precise map,” in 2017 IEEE 20th International

Conference on Intelligent Transportation Systems (ITSC). IEEE, 2017, pp. 1–6.

[755] P. F. Orzechowski, A. Meyer, and M. Lauer, “Tackling occlusions & limited sensor

range with set-based safety verification,” in 2018 21st International Conference

on Intelligent Transportation Systems (ITSC). IEEE, 2018, pp. 1729–1736.

[756] O. S¸. Ta¸s and C. Stiller, “Limited visibility and uncertainty aware motion plan-

¨ ning for automated driving,” in 2018 IEEE Intelligent Vehicles Symposium (IV).

IEEE, 2018, pp. 1171–1178.

[757] M.-Y. Yu, R. Vasudevan, and M. Johnson-Roberson, “Occlusion-aware risk

assessment for autonomous driving in urban environments,” IEEE Robotics and

Automation Letters, vol. 4, no. 2, pp. 2235–2241, 2019.

[758] M. Naumann, H. Konigshof, M. Lauer, and C. Stiller, “Safe but not overcautious

motion planning under occlusions and limited sensor range,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 140–145.

[759] M. Althoff and J. M. Dolan, “Online verification of automated road vehicles using

reachability analysis,” IEEE Transactions on Robotics, vol. 30, no. 4, pp.

903–918, 2014.

130

[760] S. Magdici and M. Althoff, “Fail-safe motion planning of autonomous vehicles,”

in 2016 IEEE 19th International Conference on Intelligent Transportation Systems

(ITSC). IEEE, 2016, pp. 452–458.

[761] S. Kousik, S. Vaskov, M. Johnson-Roberson, and R. Vasudevan, “Safe trajectory

synthesis for autonomous driving in unforeseen environments,” in ASME 2017

Dynamic Systems and Control Conference. American Society of Mechanical

Engineers Digital Collection, 2017.

[762] M. Koschi, C. Pek, M. Beikirch, and M. Althoff, “Set-based prediction of

pedestrians in urban environments considering formalized traffic rules,” in 2018

21st international conference on intelligent transportation systems (ITSC). IEEE,

2018, pp. 2704–2711.

[763] N. R. Kapania, V. Govindarajan, F. Borrelli, and J. C. Gerdes, “A hybrid control

design for autonomous vehicles at uncontrolled crosswalks,” in 2019 IEEE

Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1604–1611.

[764] S. J. Russell and A. Zimdars, “Q-decomposition for reinforcement learning

agents,” in Proceedings of the 20th International Conference on Machine Learning

(ICML-03), 2003, pp. 656–663.

[765] M. A. Brewer, K. Fitzpatrick, J. A. Whitacre, and D. Lord, “Exploration of

pedestrian gap-acceptance behavior at selected locations,” Transportation

Research Record: Journal of the Transportation Research Board, vol. 1982, no. 1, p.

132–140, 2006.

[766] C. Cherry, B. Donlon, X. Yan, S. E. Moore, and J. Xiong, “Illegal mid-block

pedestrian crossings in china: gap acceptance, conflict and crossing path

analysis,” International Journal of Injury Control and Safety Promotion, vol. 19, no.

4, p. 320–330, 2012.

[767] D. S. Pawar and G. R. Patil, “Pedestrian temporal and spatial gap acceptance at

mid-block street crossing in developing world,” Journal of Safety Research, vol. 52,

p. 39–46, 2015.

[768] G. Yannis, E. Papadimitriou, and A. Theofilatos, “Pedestrian gap acceptance for

mid-block street crossing,” Transportation Planning and Technology, vol. 36, no.

5, p. 450–462, 2013.

[769] M. Marron, J. Garcia, M. Sotelo, D. Fernandez, and D. Pizarro, ”XPFCP”:

131

an extended particle filter for tracking multiple and dynamic objects in complex

environments, 2005.

[770] M. Marr´on, J. C. Garc´ıa, M. A. Sotelo, D. Pizarro, I. Bravo, and J. L. Mart´ın, A

Bayesian Solution to Robustly Track Multiple Objects from Visual Data. Studies in

Computational Intelligence, 2008, p. 531–547.

[771] “Gps accuracy.” [Online]. Available:

https://www.gps.gov/systems/gps/ performance/accuracy/

[772] I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim,

“Self-driving like a human driver instead of a robocar: Personalized comfortable

driving experience for autonomous vehicles,” arXiv preprint arXiv:2001.03908,

2020.

[773] M. K. Kozlov, S. P. Tarasov, and L. G. Khachiyan, “Polynomial solvability of convex

quadratic programming,” in Doklady Akademii Nauk, vol. 248, no. 5. Russian

Academy of Sciences, 1979, pp. 1049–1051.

[774] Y. Bar-Shalom and X. Li, “Multitarget-multisensor [2] y. bar-shalom and xr li,”

Multitarget-Multisensor Tracking: Principles and Techniques, 1995.

[775] A. Doucet and N. de Freitas, “N. gordon, editors: Sequential monte carlo in

practice,” 2001.

[776] A. Dommes, M.-A. Grani´e, M.-S. Cloutier, C. Coquelet, and F. HugueninRichard,

“Red light violations by adult pedestrians and other safety-related behaviors at

signalized crosswalks,” Accident Analysis & Prevention, vol. 80, pp. 67–75, 2015.

[777] G. Jahangirova, A. Stocco, and P. Tonella, “Quality metrics and oracles for

autonomous vehicles testing.”

[778] R. Schram, A. Williams, M. van Ratingen, S. Ryrberg, and R. Sferco, “Euro ncap’s

first step to assess autonomous emergency braking (aeb) for vulnerable road

users,” in Proceedings of 24th Enhanced Safety of Vehicles (ESV) conference, 2015.

[779] M. Edwards, A. Nathanson, and M. Wisch, “Estimate of potential benefit for

europe of fitting autonomous emergency braking (aeb) systems for pedestrian

protection to passenger cars,” Traffic Injury Prevention, vol. 15, no. sup1, p. S173–

S182, 2014.

[780] M. Saberi, K. Aghabayk, and A. Sobhani, “Spatial fluctuations of pedestrian

velocities in bidirectional streams: Exploring the effects of self-organization,”

Physica A: Statistical Mechanics and its Applications, vol. 434, p. 120–128,

OOF

132

2015. [Online]. Available: https://dx.doi.org/10.1016/j.physa.2015.04.008

[781] C.-G. Wallman and H. ˚Astro¨m, “Friction measurement methods and the

correlation between road friction and traffic safety: A literature review,” 2001.

133

Appendix A: Generalizing the Policy for Different

Conditions and for Driving Comfort

Assuming that one can obtain the road friction coefficient, either by measuring it

or estimating it from the road conditions, and that one can calculate amax using (A.1).

Then one can also calculate dstop value from (2.8). However, there is also another

variable necessary in order to calculate dstop, which is tramp. It is a challenging task to

actually obtain limitation for a vehicle, and it is different for different vehicles.

However, one can use the AEB standards for tramp as in a situation in which the

maximum braking is required, the AEB system is going to be activated. One can see

Appendix to see the conditions required for the activation of AEB systems. Edwards

et al. [40] provided a typical tramp value in AEB systems for 2018+ generation vehicles

to be 0.35 secs.

amax = µroad ∗ g (A.1)

where µroad is the road friction coefficient, and g is the gravitational acceleration value.

However, as one can expect, the road friction coefficient varies with different weather

conditions and road material. [42] gathered the variations due to weather conditions,

which is listed in Table A.1.

Table A.1: Value range for the road friction coefficient for different weather

conditions

from Wallman and Astrom [42]

134

Condition

Value Range

Dry bare surface

[0.80, 1.0]

Wet bare surface

[0.70, 0.80]

Loose snow

[0.20, 0.50]

Packed snow

[0.20, 0.30]

Black ice

[0.15 0.30]

Loose snow on black ice

[0.15, 0.25]

Wet black ice

[0.05, 0.10]

Figure A.1: Comfortable driving limitations in terms of jerk and acceleration, both in

lateral and longitudinal axes

As stated in Section 2.1.2, dstop for amax should always be strictly smaller than visible

range rvisible, that is, the vehicle should never drive a velocity that results in a dstop value

greater than rvisible. For the comfortable driving, [33] provided values for both

acceleration and jerk values as in Fig. A.1.

from Bae et al. [33]

Students also viewed