Literature Review

O R I G I N A L A R T I C L E

User perceptions of anthropomorphic robots as monitoring devices

Stuart Moran • Khaled Bachour • Toyoaki Nishida

Received: 15 February 2012 / Accepted: 16 September 2013 / Published online: 20 October 2013

� Springer-Verlag London 2013

Abstract The principle behind anthropomorphic robots is

that the appearance and behaviours enable the pre-defined

social skills that people use with each other each day to be

used as a means of interaction. One of the problems with

this approach is that there are many attributes of such a

robot which can influence a user’s behaviour, potentially

causing undesirable effects. This paper aims to identify and

discuss a series of the most salient behaviour influencing

factors in the literature, related to a number of robot

attributes. A particularly novel aspect of this work is the

study of robots as monitoring or data collection devices,

one of several behaviour influencing factors which has

received insufficient attention. As a first step towards this,

the PSA matrix is produced, which visualises the rela-

tionships between system attributes and user perceptions,

grounded in empirical evidence in the literature. This

matrix highlights gaps in the literature, brings together a

series of salient behaviour influencing factors for the first

time and gives a clear view of the state of the art.

Keywords Anthropomorphic � Data collection � Matrix � Monitoring � Pervasive � Robotics � Ubiquitous

1 Introduction

With continuous advancements in artificial intelligence and

reduction in costs of hardware, the reality of a world where

humans coexist with robots approaches ever closer.

Through ongoing social research, most major computer

technologies have a clear set of design principles and

methods for introducing technology successfully into the

user market. In the case of robots, one direction being

explored is the approach of a humanlike appearance.

Anthropomorphic robots attempt to make the use of the

well-defined social abilities that people use with one

another day-to-day as a means of interaction. The benefits

of this are that there is ideally very little learning time

required to interact with the technology, through a familiar

and intuitive interface. There are many different attributes

of a robot which can help increase the perceptions of

anthropomorphism, including facial features, physical

expressiveness, emotions and personality. One of the

problems with these attributes is that slight variations can

cause a change in the way a robot is perceived and inter-

acted with, including feelings of discomfort [e.g. uncanny

valley (Mori 1970)]. Such variations are not limited to

human–robot interaction (HRI), but can also been found in

other computer systems. For example, many such parallels

exist in monitoring systems, such as CCTV and wearable

computing.

We argue that anthropomorphic robots, by design, are

inherently monitoring (or data collection) devices, but

unlike typical monitoring systems, they have a unique

embodiment and physical presence. In human–human

relationships there is evidence that shows how being

observed can change behaviours, consider social facilita-

tion effects (Woods et al. 2005a), or the infamous, but

often contested, Hawthorne effect (Wickstrom and Bendix

S. Moran (&) � K. Bachour Mixed Reality Lab, University Nottingham, Nottingham, UK

e-mail: [email protected]

K. Bachour

e-mail: [email protected]

T. Nishida

Nishida Lab, Department of Intelligence Science and

Technology, Graduate School of Informatics, Kyoto University,

Kyoto, Japan

e-mail: [email protected]

123

AI & Soc (2015) 30:1–21

DOI 10.1007/s00146-013-0515-6

2000). When we consider the ways in which robots can be

anthropomorphised, questions start to emerge as to how

their humanoid form might lead to similar data collection-

based behavioural effects. Hence, the beginnings of an

intriguing point of intersection starts to emerge between the

fields of humanoid robotics and monitoring systems.

The aim of this paper is to propose, explore and

stimulate discussion on the novel thread of research on

anthropomorphic robots as monitoring systems. Viewing

anthropomorphic robots through a ‘monitoring lens’ may

lead to new understandings, perspectives and questions

regarding their relationship with users. To do this, we

must first understand the state of the art in both fields

individually, and then examine what exists at the point of

their intersection. In particular, we are interested in how

these attributes may change and influence behaviour

(along the lines of Hawthorne). Research in social

robotics has shown a tendency to explore how different

characteristics or attributes of a robot can influence users

through their differing perceptions and attitudes. Through

a survey of the monitoring and robotics literature, a series

of key robot characteristics emerged which can influence

the user during interaction, which frames the discussion.

Similarly, there are a number of key attributes associated

with monitoring systems (Moran and Nakata 2010), and

these were used as a coding scheme when exploring the

literature. We observed a number of established and

hypothesised causal inter-relationships between design

characteristics and user perceptions of both anthropo-

morphic robot and monitoring systems. Building on these

relationships, we constructed a matrix to help visually

portray the relationships between characteristics within

the area. Using this matrix, we are able to uncover how

these two fields intersect and also identify various current

trends and patterns. A benefit of this is the effective

capture and ‘mapping’ of existing research, in addition to

the identification of gaps and future potential directions of

research. Following this, we present a series of challenges

to help guide and frame future research exploring robots

as a monitoring systems.

2 Background

The following sections discuss both anthropomorphic

robots and monitoring systems, with the aim of providing

evidence to support the motivation of this paper.

2.1 Anthropomorphic robots

A humanoid form provides a natural means of interaction

between human and social-based robots (Heerink et al.

2006). Humanoid robots are effective at affording

(suggesting) how they should be controlled, used and

responded to (Dautenhahn 2002), easing the burden on

learning curves for users. It is this affordance which fun-

damentally differentiates the interaction between humans

and robots compared to other technologies in daily use.

This difference stems from the change in a users perception

of a robot from simply being a technology to being a social

actor (Heerink et al. 2010a). This change can be attributed,

in part, to the robot’s appearance, the fact robots will

actively share spaces with people, and the robots display of

intelligence and interest with the user and its surroundings

(Young et al. 2009). While there are many benefits to a

humanoid appearance, there comes a cost to this unique

interaction in the form of undesirable/unexpected effects

on users such as discomfort (Koay et al. 2006) or misplaced

expectations (Dautenhahn 2004). The ‘uncanny valley’ was

proposed by Mori (1970), which describes the different

degrees of anthropomorphism as experienced by the user.

The trough of the ‘valley’ represents the point of human

rejection, while the peaks represent acceptance. Like other

researchers (Bartneck et al. 2007), we take the view that the

‘valley’ should rather be considered a ‘cliff’, as users are

unlikely to knowingly appeal to a machine which demon-

strates increasingly ‘perfect’ anthropomorphism.

There are a number of other factors that may influence

user perceptions and feelings towards a robot (Young et al.

2009), for example existing research has established that a

robots appearance and actions can affect human behav-

iours, including a willingness to comply with robot

instructions or advice (Goetz et al. 2003; Powers and

Kiesler 2006), communication avoidance (Nomura et al.

2006) and the time spent with the robot and whether a user

is prepared to have physical contact (Nomura et al. 2008).

This influence is not limited to physical attributes, but can

also include effects caused by social attributes [e.g. per-

sonality (Takayama and Pantofaru 2009)] or more func-

tional attributes such as its purpose or application (Goetz

et al. 2003). While any influence is likely to always be

carefully considered, some of the human behaviours caused

by human–robot interaction (HRI) can be undesirable. This

may be avoidable through a rigorous identification and

understanding of all robots attributes which may influence

people (Siegel et al. 2009; Kuo et al. 2009). However,

many of these attributes still remain unidentified, unex-

plored or simply remain, as of yet, disjointed and unrelated

in the literature. More work is needed to identify, study

and/or bring together these salient factors and their inter-

relationships, including their associations with new, pre-

viously unconsidered factors (Heerink et al. 2009a). For

example, there is a fundamental aspect of HRI which to

date has received insufficient direct attention; an aspect

which stands to have a significant effect on the user when

robots are introduced into the ‘real world’.

2 AI & Soc (2015) 30:1–21

123

2.2 Data collection

To effectively interact with a user and achieve its intended

purpose/service, a robot must be ‘aware’ of the user, col-

lecting and processing significant amounts of data (Syrdal

et al. 2007b; Glas et al. 2009). Robots must be able to track

and recognise the location, identity and behaviour of

humans for any meaningful social interaction to take place

(Michalowski et al. 2006). Data collection is not unique to

HRI and is adopted in many (if not all) interactive tech-

nologies in some way. As mentioned earlier, HRI differs

heavily from existing technologies, mainly in terms of the

degree of interaction. Existing monitoring technologies,

such as closed-circuit television (CCTV) or radio fre-

quency identification (RFID), collect data and information

in a one-way interaction, restricted to certain areas, space

and data types (Syrdal et al. 2007b). This means that the

technology only collects data about the user, with little

reaction or response. In contrast, robots adopt a two-way

interaction (e.g. request and response) and also have the

opportunity, and capability, to collect larger amounts of

(personal) data than other technologies in private use

(Syrdal et al. 2007b).

One of the problems associated with monitoring of any

kind is that, like robot appearance and actions, it has been

shown to affect the behaviour of the person being observed

(Rosenthal and Jacobson 1968; Larson and Callahan 1990)

and even cause discomfort (Scholtz and Consolvo 2004;

Lyytinen et al. 2004). If the act of monitoring is affecting

the behaviour of the user, arguable their response is no

longer natural. This makes the data collected inaccurate,

potentially leading to a sub-optimal or inappropriate

response from the robot (or any interactive technology).

Moran and Nakata (2010) have conducted work where they

clearly identified a series of salient monitoring system

attributes that influence the interaction with users including

awareness, control, boundaries, trust, intrusion, justifica-

tion and context. It is our view that these are directly

applicable to a robot and may be part of the cause of

behavioural or interactional change, e.g. its placement in

relation to the users or the size of the sensors. For example,

consider a humanoid robot which has eyes and ears. These

are properties similar to a human which afford certain

capabilities, e.g. see and hear in specific directions. There

are likely to be discrepancies between these perceived/

afforded abilities and the actual, where a robot may have

eyes at its front, but may be able to see all around it. The

affordances may also run deeper, whereby humans per-

ceive the robot to have the same perceptibility and judge-

ment as humans, generating Hawthorne effects (Wickstrom

and Bendix 2000). There is also the potential to create

Panoptic effects (Jonsson 2006), whereby humans cannot

know when a robot is, or is not, collecting data about them

and so must assume data are collected constantly. When

this is considered alongside the Hawthorne effect, the types

of problems related to behaviours and coverage that can

occur become highlighted.

Robots can also be used in conjunction with other

technologies, such as RFID tags and floor sensors (Nohara

et al. 2008), sharing information as a part of a wider sys-

tem, leading to augmentation of the robots capabilities

(Koide et al. 2004; Dragone et al. 2010). Through this

augmentation, the coverage of the robots data collection is

increased, and with this increase, the undesirable effects on

users are anticipated to become amplified and more

numerous (Moran and Nakata 2010). When all this is

considered alongside the unique and private way humans

interact with robots, it becomes clear that designers of

these systems should carefully consider the issues of

monitoring robots with social intelligence (Kahn et al.

2006). This is important given the area of ubiquitous

human–robot interaction remains largely unexplored

(Dragone et al. 2010). A user’s response to a system is

predominately based on their perception of reality and not

the reality itself. Their perceptions of the attributes of the

system are more meaningful to them, and hence their

behaviours, than the objective facts (Young et al. 2009). As

such, the relationship between robot attributes and user

perceptions should be explored together, to provide a more

holistic understanding of the effects of the robot/monitor-

ing systems. For example, some researchers have already

established which precise physical attributes of a robot

head can be intentionally varied to effectively control how

humanlike it is perceived by the user (DiSalvo et al. 2002).

Much like the inter-relationships between physical and

social factors, there are a number of directly corresponding

perceptions, but there are also perceptions which comprise

a number of different physical and social factors. It is

through empirically measuring and understanding the

relationship between a robots attributes and a users per-

ceptions of said attributes that researchers will eventually

be able to intentionally encourage or prevent specific

behavioural responses (Nomura et al. 2007).

The research literature on HRI, behavioural change and

perceptions was comprehensively explored in order to

identify a series of salient and recurring factors which

influence user behavioural responses to robots, a similar

approach to Fong et al. (2003). In order to capture as many

different influencing attributes of a robot as possible, the

review explores both relevant hypotheses and empirical

studies, and the impacts of robots of different types (though

a general leaning towards anthropomorphic), in different

contexts and with different types of users. This provides a

holistic view of robotics and the possible areas of influence

in HRI. This is beneficial as it may capture information

which could be overlooked with a narrower search. The

AI & Soc (2015) 30:1–21 3

123

factors identified in the literature are categorised in terms

of physical and social attributes. Physical attributes con-

cern the tangible aspects of a robot, while the social attri-

butes are related to the application and interaction with

humans at a social level. Depending on the user, the

behavioural influence may be stronger in one of these

levels than the other. For example, users may be more

interested in the functional aspects of the robot (e.g. shape

and things which influence physical task) than aspects such

as colour or contour which influence human-likeness.

Equally, a user may be more interested in the way the robot

behaves and talks than what tasks it can carry out and its

physical form. A secondary objective of this study of the

literature is to explore the monitoring aspects of HRI in

terms of the factors identified by Moran and Nakata (2010),

where each factor was used as a part of a coding scheme

during the review.

3 Monitoring attributes

We now consider a series of attributes related to monitor-

ing systems that are relevant in the context of social

robotics as data collecting devices, namely privacy and

invasiveness, control, user understanding and ownership/

trust.

3.1 Privacy and invasiveness

The first set of attributes that have been studied in the

context of monitoring robots deal with privacy and the

physical, social and perceived invasiveness of robots.

3.1.1 Border crossings

Perceived natural border crossing (PNBC) is defined as

the degree to which a person feels that any natural borders

have been crossed, based on Marx (2001). Marx (2001)

talks about natural borders, where physical objects create

borders which help define physical areas of privacy (e.g. a

home is defined by its walls). When information crosses

these borders, privacy is perceived to have been invaded.

However, a user must initially perceive the physical bor-

ders to have been crossed before perceptions of privacy

invasion arise. Another form of natural border crossing is

when a robot physically invades a user’s personal space,

even in the absence of a physical border. When a robot was

closer than 3 m users felt discomfort (Koay et al. 2006;

Dautenhahn 2007), and the further away the robot, the

more comfortable they felt (Pacchierotti et al. 2006). If

robots must be closer than 3 m to achieve their task, then

there are several ways to avoid discomfort including: giv-

ing users control over the robot (Walters et al. 2009a),

reducing its size (Hiroi and Ito 2008) and/or ensuring that

the user feels minimal levels of anxiety (Nomura et al.

2007). Designers should carefully consider the robots they

are introducing into social environments and attempt to

prevent users feeling spatial discomfort (Syrdal et al.

2008). Understanding the relationships between systems

characteristics and perceptions is the first step towards

achieving this.

Perceived social border crossing (PSBC) is defined as

the degree to which a person feels that a social border has

been crossed, based on Marx (2001). Social borders refer to

the borders defined by a person’s expectations of confi-

dentiality in social roles and relationships (Marx 2001).

They are heavily related to natural borders, in that infor-

mation can be shared across them (made available to oth-

ers), resulting in perceptions of privacy invasion. In some

cases, people do not want their information to be shared

with certain social groups (e.g. colleagues at work). How

robots (accidently, or intentionally) store and share data

with a third party is a concern actively voiced by people

who have experienced HRI (Vargas et al. 2009; Syrdal

et al. 2007b). When robots are used to capture data in

public spaces, it is important that the reason for this is

made explicitly clear to all those who directly, and indi-

rectly, interact with it (Syrdal et al. 2007b).

3.1.2 Frequency of data collection and border crossings

Perceived frequency of border crossings (PFBC) is defined

as a person’s understanding of how frequently borders are

crossed, based on Moran and Nakata (2010). In keeping

with the idea that interaction with a robot involves border

crossings (social and natural), user perception of the reg-

ularity of these crossings may influence their behaviour

(Nomura et al. 2008). Furthermore, when determining the

level of social skills of a robot, the nature and frequency of

interaction need to be analysed (Dautenhahn 2007).

Frequency of data collection is defined as the regularity

with which data can be collected by the robot, based on

Moran and Nakata (2010). Robots have the potential to

continuously collect data about users and an environment

(Syrdal et al. 2007b) and repeated experiences with a robot

can change the behaviour of the human involved (Nomura

et al. 2007). A common assumption is that HRI is co-

located and synchronous; however, a user may interact

with the robot at different times (delayed responses) and

even in different places (through use of internet) (Yanco

and Drury 2004). For this reason, it is necessary to consider

not only the time spent within a robot’s coverage, but also

the frequency with which they engage with the robot and

hence the area being monitored. This aspect of HRI

becomes significant when we consider the mobility of a

robot.

4 AI & Soc (2015) 30:1–21

123

3.1.3 Robot obtrusion (physical presence)

Robot obtrusion (RO) is defined as the obtrusiveness of the

robot in terms of its physical presence, based on Moran and

Nakata (2010). RO captures the physical obtrusiveness of a

robot in terms of its position in relation to a user and the

effect of its complete overall appearance. While obtru-

siveness might be considered a subjective phenomena, we

argue that there are likely to be thresholds of physical

obtrusiveness that are consistent across users. Furthermore,

there are attributes and perceptions of a robot that are

discussed later in the paper that capture the more social

aspects related to obtrusiveness. RO is closely related to

PNBC in that robots can have an invasive physical pre-

sence (Young et al. 2009; Kidd and Breazeal 2004) when

crossing the natural border into a user’s personal space

leading to perceptions of anxiety (Hiroi and Ito 2008;

Kanda et al. 2002) or spatial discomfort (Syrdal et al.

2008), particularly when placed within the 3 m range

(Koay et al. 2006; Dautenhahn 2007). However, the mere

knowledge of the presence of a robot collecting data about

a user may increase the risk of perceiving their personal

space to have been invaded, causing discomfort (Scholtz

and Consolvo 2004; Lyytinen et al. 2004).

3.1.4 Perceived privacy invasion

Perceived privacy invasion (PPI) is defined as the degree to

which a person feels that the robot is invasive of their

privacy, based on Moran and Nakata (2010). Privacy

invasion is one of the most important user perceptions to

consider, as it could be the most significant cause of

undesirable behaviours. Dryer et al. (1999) describe pri-

vacy as a function of the type of information users must

give up to attain value from the application and how widely

available this information is disseminated. The nature of a

robot means it interacts with a human in a personal way

that no other technology does, but how does the increased

level of data capture of personal data in this context affect a

person’s privacy (Kahn et al. 2006)? It has been shown that

people may not feel comfortable about the collection of

personal data by a robot (Syrdal et al. 2007b). Another

privacy consideration raised with regard to data collected

by a robot is its long-term storage (Vargas et al. 2009). A

robot’s ability to ‘forget’ certain information may be a way

for a user to control the robot, effectively reducing the

perceptions of privacy invasion by regulating the type and

amount of data stored (Vargas et al. 2009).

Depending on the context and purpose of the technol-

ogy, users may not expect total privacy (e.g. hospital robot)

and accept this fact when making choices in light of its

application (Nguyen et al. 2008). Perceptions of a system’s

capabilities appear to significantly influence PPI (Vallejo

et al. 2008), and the greater the PPI, the less willing a user

will be to accept the technology (Zweig and Webster

2003). Contrary to the above, some research suggests that

privacy is less important in determining acceptance of a

technology than is currently believed (Spiekermann 2008).

According to Spiekermann (2008), this may be due to the

users focus on the delivery of a service, rather than on

tracking a person.

3.1.5 Coverage and data integration

Robot coverage is defined as the physical range within

which a robot is able to collect data or interact with its

users, based on Moran and Nakata (2010). Each type of

sensory device used in a robotic system has a different

range or coverage compared to another. This range is

defined as the distance, or area, within which the robot is

able to collect data. Many sensors are susceptible to

physical objects such as walls and doors limiting their

range, even the most powerful devices can be affected by

such limitations. An interesting consideration is that

physical constraints are unlikely to be an issue should a

robot be a part of a pervasive sensor network, which will

increase its coverage (Moran and Nakata 2010). The ability

for a robot to move will also dramatically increase its

coverage, even if the sensors it uses are relatively weak.

With such increases in coverage and information avail-

ability, it may be difficult for users to find places that they

perceive to be unmonitored (Punie 2003).

Perceived coverage (PC) is defined as the degree to

which person understands the area covered by the robots

sensors, based on Moran and Nakata (2010). Different

users will have a different understanding of the capabilities

of certain robots, brought about in part by the assumptions

they draw from the anthropomorphic form of the robot. As

such users will have different perceptions of the coverage

of the robots sensors, and the effects of the monitoring will

be different. Several hundred, or as few as one or two,

different robots may be needed to interact with users,

collecting data. The number of robots will affect the

interaction with the user (Yanco and Drury 2004). How-

ever, this area is still relatively unexplored and the question

remains: how does the presence of a number of robots

influence the interaction between users (Severinson-Ekl-

undh et al. 2003)?

Data integration is defined as the process of combining

data from different local and/or global sources, based on

Moran and Nakata (2010). With the potential for many

robots and many sensors to be collecting different types of

data at different times, a problem arises when trying to

understand and make use of all the information collected.

Data integration is therefore a necessary step in the sense-

making process. By combining data from different sources,

AI & Soc (2015) 30:1–21 5

123

a bigger picture can be built about those being observed

and more accurate inferences made. This data might be

stored in ‘the cloud’ and be hidden from the user, with the

robot itself potentially acting as a gateway to this infor-

mation alongside being the means by which it is collected.

The question remains whether or not humans would even

want to see/have access to this information. As the

potential for inference about users activities increases, their

perceptions and attitudes towards the robot are anticipated

to change. Participants in one study showed concerns

regarding the ability of the robot to infer meaningful

information about them (Lee et al. 2011). One suggested

way of avoiding this is by having a robot effectively

communicate its data integration/collection abilities, and

how it constructs knowledge about the user (Lee et al.

2011).

3.2 Control

The second set of data monitoring attributes of robots relate

to the extent to which the user has control, or perceives

having control, over the robot itself, the data it collects, and

the way in which this data is used.

3.2.1 Robot control

Robot control is defined as a user’s ability to physically

prevent or allow a robot from performing an action, e.g.

specific movements or data collection, based on Moran and

Nakata (2010). Physical control of a robot can take many

forms: in one experiment, users were given control of

robots movement, so they could signal to stop it when the

felt it came too close (Hiroi and Ito 2008); this is a form of

gesture based or physical control. A related study showed

that users felt ‘in control’ when the robot was fully visible

and approaching from a frontal direction (Dautenhahn et al.

2006). Control could be in a simpler form, where users may

have an option to simply deactivate the robot using an ‘off’

switch. In many cases, users are likely to be required to

forego their control over a robot, for example, a medical

robot may need to frequently touch a human to administer

drugs. Equally, providing a user with too much control may

be overwhelming and intrusive, increasing the complexity

of interaction.

Perceived robot control (PRC) is defined as the degree

to which a person feels they have direct control over the

robot itself, based on Moran and Nakata (2010). Control

can vary with different robots, and depending on users’

knowledge and understanding of the robot, they may per-

ceive themselves to have different levels of robot control.

Additional factors such as both the users and robots per-

sonality must be considered, as these have been shown to

affect user perceptions of control (Syrdal et al. 2006;

Meerbeek et al. 2008). For example, it has been shown that

if a human perceives the robot to be a living entity (extreme

anthropomorphism), they may show reluctance to switch it

off (Bartneck et al. 2009b). Thus, overly high levels of

anthropomorphism may actually cause users to feel some

control has been removed from them.

3.2.2 Data and application control

Data access is defined as the degree to which the data

collected is made available to individuals, groups and/or

the public, based on Moran and Nakata (2010), while data

sharing is defined as the process of distributing information

across different robotic systems, based on Moran and

Nakata (2010). Due to the nature of interaction with a robot

and the number of potential users, the disclosure of data is

an important consideration (Syrdal et al. 2007b). Take for

example a workplace, it may be possible for a robot to

share personal work-related information (e.g. the length

and number of breaks) with your employer, potentially

becoming a tool of surveillance (Kahn et al. 2006). This

has led to people showing concerns about how their data

are shared by robots (Vargas et al. 2009) and how they can

control this information (Enz et al. 2011). An interesting

situation is where an individual in a public space is inter-

acting with the robot, but due to its sensory range, the robot

is also passively, or indirectly, interacting with all those

secondary users around it (i.e. recording data about

bystanders). An important question in HRI is how do we

give those who are not directly interacting with the robot

control over it? In some cases, such as for public security

or safety, the designer of the robot may not want any users

to have control over it.

Application control is defined as the users’ ability to

manage how and by whom their data is used, based on

Dryer et al. (1999). What differs between application

control and robot control is that the data have been col-

lected, and now the control is aimed at the use and avail-

ability of the data. Users want robots to be controllable, but

some autonomy will always be required as few people want

to carry out constant supervision or intervention (Dau-

tenhahn et al. 2005; Yanco and Drury 2004; Enz et al.

2011). Under what situations is a user able and willing to

give up control (Enz et al. 2011; Meerbeek et al. 2008)? If

the right balance is not met, then issues related to mis-

placed trust may occur (Desai et al. 2009).

Perceived data control (PDaC) is defined as the degree

to which a person feels they have control over the use of,

and access to, the data collected by the robot, based on

Theofanos and Scholtz (2005). Depending on how well the

users are informed or how much experience they have with

a robot, their perceptions of how much control they have

over data access and use may differ. Their perceptions of

6 AI & Soc (2015) 30:1–21

123

robot control may also influence their overall perceptions

of data control.

Privacy invasion in robotics research is related to the

sensitivity of the information collected (Syrdal et al.

2007b). In one study, a robot storing information related to

a users personality and other psychological characteristics

was shown to make them feel uncomfortable (Syrdal et al.

2007b). People have also shown concerns regarding what

data were being stored by the robots and for how long

(Vargas et al. 2009). One suggested approach to mitigate

some of these concerns is for a robot to have the ability to

‘forget’ the data it has collected (Vargas et al. 2009). As

new sensors are developed capable of collecting new types

of data and robots are connected to networks of such

sensors, the data, information and knowledge a robot has

access to become an important consideration to both users

and designers.

3.3 User understanding

Next, we describe attributes relating to the user’s under-

standing of how and why the robot functions, whether this

understanding is accurate or not.

3.3.1 Informedness and assumptions

User informedness is defined as the degree to which a user

is informed as to the workings or purpose of the robot,

based on Moran and Nakata (2010). This is partially related

to the concept of awareness in HRI, which is an under-

standing of the location and status of the robot (Drury et al.

2003). As is the case with many current robot systems,

there may be a legal, ethical and/or social requirement for

owners or designers to ensure that the observed are made

aware of the inner workings and purpose of the robot. This

may require the use of large signs detailing the information

in simplistic, easy to understand terms, or in the small print

of a complex terms and agreement document (Nguyen and

Mynatt 2002). Users should be informed of the types of

data the robot is collecting, particularly as privacy invasion

is related to the sensitivity of the information collected

(Syrdal et al. 2007b).

Application assumptions (AA) is defined as the degree

to which a person understands the purpose and workings of

the robot, including what inferences can be drawn about

them, based on Dryer et al. (1999). This is strongly linked

to user informedness, but deals more specifically with the

users understanding of the system. There are two main

ways users can understand the purpose and workings of the

robot: through directly (or indirectly, i.e. affordance) being

informed by people or other media (e.g. forms, signs,

presentations) or through their own personal assumptions

based on previous experiences. Depending on how well

users understand the robot, their perceptions of the purpose

and workings of the system will change. The attributes of

the system, such as the distribution and frequency of

interaction with the robot, are likely to influence their

perceptions of what inferences can be drawn about them

from the data it collects. Their assumptions about the

application of the robot may also influence their level of

trust they have in the robot (Desai et al. 2009). Perceived

anthropomorphism is an indirect form of information

sharing and can lead to assumptions regarding the abilities

to a robot. These assumptions can lead to expectations

which cannot be fulfilled (Dautenhahn 2002; Bartneck

et al. 2009a) and result in a break down in the interaction

(Dautenhahn et al. 2005). Users with negative reactions

towards a robot can affect their assumptions about its

capabilities (Nomura et al. 2006), and assumptions about

capabilities can in turn influence attitudes (Nomura et al.

2005). Without an understanding of the monitoring capa-

bilities of the robot, e.g. inability to comprehend robot

sensors (Lee et al. 2011), users are not capable of under-

standing the privacy trade-offs they are making in their part

in its use (Nguyen et al. 2008).

Perceived benefit (PB) is defined as the degree to which a

person feels the robot is of benefit to themselves, others or the

observer, based on Moran and Nakata (2010). A users will-

ingness to adopt any new technology is dependent on their

perceptions of its capabilities and functions (Vallejo et al.

2008; Beckwith 2003), and acceptance of robotic systems is

no different (Heerink et al. 2008a, 2009b). Ultimately, the

data collected by a robot should be fully justified (Syrdal

et al. 2007b; Kidd and Breazeal 2005); and if perceived as

sufficiently beneficial by the user, then any negativity

towards the robot will be tolerated and regarded as a ‘nec-

essary evil’ (Syrdal et al. 2007b). There may also be a suf-

ficient ‘reward’ that outweighs the ‘damage’ potentially

caused by disclosing specific information from the user. In

some cases, simply understanding and being aware of a robot

may remove concerns regarding invasion (Lee et al. 2011).

3.3.2 Affordance

Perceived affordance (PA) is defined as a person’s

understanding of the capabilities of the robot based on its

appearance, based on Moran and Nakata (2010). Affor-

dance is a term originally coined by James Gibson (1977),

and the term (perceived) affordance was popularised in

human–computer interaction by Don Norman (1988).

Norman’s affordance has been further extended to the verb

‘afford’, which means ‘to suggest’. The appearance of a

robot acts as an important means of sharing information

about its abilities and functionality with its users (Salvini

et al. 2010; Dautenhahn 1999; Austermann et al. 2010;

Goetz et al. 2003; Dautenhahn 2002).

AI & Soc (2015) 30:1–21 7

123

For example, a robot with human features will lead to

expectations of humanlike capabilities (Weiss et al. 2009;

Austermann et al. 2010), e.g. facial features such as eyes

can lead to perceptions of the ability to see (Blow et al.

2006a). However, information collected by other data

collection devices in a system may be shared with the

humanoid robot, augmenting its capabilities beyond what

can be perceived (Koide et al. 2004; Dragone et al. 2010).

This presents a dilemma for users, who perceive the robot

as humanoid, with humanlike capabilities, but the robot

actually has capabilities that are not afforded by its phys-

ical appearance (Dragone et al. 2010). In this sense, it is

interesting to explore what monitoring abilities the robot

affords. The more humanlike it is viewed, then the more its

capabilities will be seen as humanlike. For example, a user

assumes that as a human cannot see through a wall, or see

through the back of its head then a robot with eyes cannot

either (Lee et al. 2011). Consider a robot which is con-

nected to additional sensors and may be able to see or know

things beyond what it affords. A user may walk behind the

robot, but the robot can still ‘see’ the user because of

additional sensors. When the true capabilities of an

anthropomorphic robot are experienced, and they are

beyond what a human is capable of, then it is anticipated

that perceptions of anthropomorphism will decrease.

3.4 Ownership/trust

The following two attributes are closely related as they deal

with user’s perception of whether or not they own the robot

and whether or not they can trust it.

3.4.1 Perceived ownership

Perceived ownership is defined as the degree to which the

user perceives that they have ownership of the robot. One

of the benefits of perceived ownership, or the perception

that the robot is a consumer item, is that the robot will be

viewed as non-threatening and under the users control

(Blow et al. 2006b). This control could be in the form of a

choice in using the device, or the user’s ability to tailor

aspects of a robot, such as appearance and behaviour, in

accordance with their own preferences (Dautenhahn 2007).

It is also interesting to consider this from the point of view

of data collection; if the user perceives ownership of the

robot, then their control, in principle, extends beyond the

robot itself to the data it collects. All issues related to the

monitoring aspects of a robot may be reduced if the user

perceives ownership of the robot and therefore any data

collected. Another view of ownership could be in terms of

a robot’s direct owner with an independent user interacting

with it. For example, consider a medical robot, the patient

does not own it, but still has to share sensitive information

with it, information which the owner of the robot may have

access to. The interplay and dynamics between the three

stake holders: robot(s), robot owner(s) and user(s) will be

an important consideration in how the factors identified in

this paper influence the behaviour of the users. This three-

way interaction is more akin to the relationship between a

human observer and an observed human, facilitated by a

humanoid robot which may be perceived as another

human—a complex but interesting scenario.

3.4.2 Perceived trustworthiness

Perceived trustworthiness (PT) is defined as the degree to

which a person feels the robot, or observer, is trustworthy,

based on Moran and Nakata (2010). For a robot to be

successful in its implementation, users must have trust in it

and its operation (Looije et al. 2006; Heerink et al. 2009b).

A users placement of trust in the robot can be influenced by

the robot’s gender (Siegel et al. 2009), physical and social

presence (Bainbridge et al. 2008), behaviour, autonomy

and adaptability (Desai et al. 2009). The less humanlike the

robot’s appearance and voice, the less trustworthy it will be

perceived (Tamagawa et al. 2011); however, it has also

been claimed that the more humanlike, the less trustworthy

robots become (Schermerhorn et al. 2008). Like the

uncanny valley, there may be a ‘peak of trust’ in a robot,

where there is a threshold of anthropomorphism which

achieves the maximum possible trust. The users own

characteristics will also influence their trust in the robot,

including user personality (Severinson-Eklundh et al.

2003), gender (Siegel et al. 2009), cultural background (Li

et al. 2010), application assumptions (Desai et al. 2009)

and how informed they are about the robot (Desai et al.

2009). The more a user trusts a robot, then the more it will

be perceived as a sociable entity (Heerink et al. 2009b,

2010b).

Robots are not always autonomous and can often be

controlled by a human. This means that perceived trust can

be expanded to include the human observer, whether this is

the controller or simply a user who has access to the

information collected. The observer can be considered

trustworthy if they collect and use a person’s personal data

(as collected by the robot) honestly, responsibly and with

no intent to cause harm (Scholtz and Consolvo 2004).

According to Theofanos and Scholtz (2005), the concept of

trust can be measured in terms of a user’s understanding of:

what data are recorded and what inferences are made about

them (awareness); the user’s ability to manage how and by

whom their data are used (control); what type of infor-

mation they must give up to receive the service and how

widely available the information collected is to other users

(privacy). A balance of trust must be found between the

observed/user and the observer/robot, as too much trust can

8 AI & Soc (2015) 30:1–21

123

be just as problematic as too little (Desai et al. 2009). It

should be noted that insufficient research has investigated

the issue of trust in robotics (Desai et al. 2009; Kidd and

Breazeal 2005).

4 Anthropomorphic attributes

We now move on to consider a series of attributes related to

anthropomorphic robots that are relevant in the context of

behavioural change, namely intelligence, autonomy and

adaptability, anthropomorphism and user-centric attributes.

4.1 Intelligence, autonomy and adaptability

The first set of anthropomorphic properties deal with the

level of autonomy the robots have as well as its ability to

reason within and adapt to the context it is placed in.

4.1.1 Intelligence

Robot intelligence is defined as the level of intelligence

potential of the robot. The intelligence of a robot will play

an important part in its interaction with users, ranging from

socially acceptable movement around a space to under-

standing and communicating socially through use of nat-

ural language (Beer et al. 2011). Robot intelligence may

not yet be at a sufficiently effective level, but there are

ways to emulate intelligence, through animacy and

appearance; though this effect is not retained over time

(Bartneck et al. 2009b).

Perceived intelligence is defined as the degree to which

a robot is seen as a knowledgeable and competent agent

[based on Bartneck et al. (2008)]. In many ways, the per-

ceived intelligence of the robot will be the predominant

determinant of its ability to be perceived as anthropomor-

phic. Existing research has established that perceptions of

intelligence are influenced by the adaptability (Austermann

et al. 2010), anthropomorphism (Walters et al. 2009b) and

animacy (Bartneck et al. 2009b) of a robot.

4.1.2 Autonomy

Robot autonomy is defined as the robots ability to function

independently of human control or intervention. The ability

for a robot to function without constant supervision from

users is critical to their adoption ‘in the wild’. There will of

course need to be constraints on the robots autonomy, as

users may begin to perceive they have reduced control over

the robot, potentially leading to perceptions of anxiety. A

balance should be found between the autonomous func-

tioning of a robot, and a user’s ability to intervene, when

they so choose.

Perceived autonomy is defined as the degree to which a

robot is perceived to demonstrate an ability to function

independently without the need for human intervention or

supervision. In many ways, this is one of the fundamental

principles of a robot and is often an assumption in many

instances of HRI. Different degrees of perceived autonomy

have been found to influence the social presence of a robot

(Syrdal et al. 2008).

4.1.3 Adaptability

Robot adaptability is defined as the robots ability to adapt

and respond to a variety of situations. A robot will be

presented with many scenarios where it will need to adapt,

perhaps dynamically changing attributes such as person-

ality and behaviour. A robot must know how to act in a

socially acceptable way, particularly as the social situation

changes, e.g. moving from formal to informal environ-

ments or showing authority when a person is not acting

correctly (e.g. damaging something).

Perceived adaptability is defined as the degree to which

a robot is perceived as adapting to the needs and situation

of its users (Heerink et al. 2008a). This has been shown to

affect user attitudes towards a robot (Torrey et al. 2006)

and ultimately affect their acceptance of the robot (Heerink

et al. 2009b; Broadbent et al. 2009). Other aspects of a

robot, such as its social presence (Syrdal et al. 2008),

intelligence (Austermann et al. 2010) and perceived use-

fulness (Heerink et al. 2008a, 2009b, 2010b), have been

established as determinants of perceived adaptability.

4.2 Anthropomorphism

The next set of properties relate to the robot’s humanlike

physical appearance, its expressiveness, behaviour and

personality and how they all affect how the robot is viewed

as anthropomorphic as well as how this affects user

expectation and interaction.

4.2.1 Anthropomorphic form

Robot form is defined as the complete physical appear-

ance of a robot, ranging from purely functional to

humanlike characteristics. A humanlike physical form

exhibited by a robot has been linked to a host of different

effects. These effects are created by the user’s tendency to

anthropomorphise a humanlike robot. A first step to

achieving this is to emulate the structure of a human,

including a torso, a head, arms, legs, and even fingers and

toes (Beer et al. 2011); as opposed to the physical

structure of an animal (e.g. zoomorphic (Yanco and Drury

2004)). The details of these components are important, for

example, the dimensions, specific features and total

AI & Soc (2015) 30:1–21 9

123

number of features of a robot face (Blow et al. 2006b;

DiSalvo et al. 2002; Broadbent et al. 2009) have been

shown to influence perceptions of human-likeness, robot

sociability, capabilities and even knowledge (Powers and

Kiesler 2006; DiSalvo et al. 2002; Blow et al. 2006a).

Together these attributes, and their size, can determine the

physical presence a robot holds in an environment.

Variations in size and presence can be quite problematic

for an interaction, so much so that in one experiment,

researchers consciously reduced the size of their robot to

less than the height of an adult male so that it would not

be perceived as threatening (Hiroi and Ito 2008). This

change in size demonstrates the ways in which designers

can control the way a robot is perceived. One of the

problems with giving a robot a too strong human-likeness

is that it may give false expectations about the robots

abilities (Dautenhahn 2004). If these are not met, then

there may be a break down in the interaction (Dautenhahn

et al. 2005; Velonaki 2010). To compensate for this, it

may be possible to dynamically change the appearance of

a robot to coincide with user expectations in a specific

context or scenario (Siegel et al. 2009). For example,

changing a property such as gender has been shown to

affect acceptance (Broadbent et al. 2009; Beer et al.

2011). The form of a humanoid robot, as would be

expected, influences a significant number of perceptions

including likeability (Bartneck 2008), familiarity (Dau-

tenhahn 2002), empathy (Riek et al. 2009), emotional

experience (Zhang et al. 2010), anthropomorphism

(Bartneck 2008; Austermann et al. 2010), trustworthiness

(Bainbridge et al. 2008), invasiveness, (Young et al. 2009;

Kidd and Breazeal 2004; Walters et al. 2005; Kanda et al.

2008), anxiety (Kanda et al. 2002), expectations (Dau-

tenhahn 2004) and human-likeness, robot sociability,

capabilities and even knowledge (Powers and Kiesler

2006; DiSalvo et al. 2002; Blow et al. 2006a).

Perceived anthropomorphism (PA) is defined as the

degree of robot human-likeness based on both physical

appearance and user interaction experience (Zhang et al.

2010). The different degrees of anthropomorphism arise

from a users perceptions of humanoid appearance, form

and skills (Bartneck 2008; Beer et al. 2011). This includes

attributes such as a robots vocal and facial characteristics

(DiSalvo et al. 2002; Powers and Kiesler 2006), animacy

(Goetz et al. 2003; Severinson-Eklundh et al. 2003; Di-

Salvo et al. 2002), communication (DiSalvo et al. 2002),

behaviour and emotions (DiSalvo et al. 2002; Sabanovic

et al. 2006; Bartneck et al. 2009a), and gender (Scher-

merhorn et al. 2008), in addition to the number of these

features (Hegel et al. 2009). The more, or less, a robot is

perceived as anthropomorphic will likely affect other per-

ceptions a user holds about the robot (Lee 2010). For

example, an anthropomorphic robot would not afford the

ability to see through objects, which is another area users

may be misled (Lee et al. 2011). It also affects how users

interact with the robot; for example, research has shown

that when robots are viewed as more humanlike, users are

likely to be less truthful (Schermerhorn et al. 2008). Like

robot form, perceived anthropomorphism has been estab-

lished as an indicator of perceived robot intelligence

(Walters et al. 2009b), ease of use (Hegel et al. 2009),

social presence (Schermerhorn et al. 2008), personality

(Goetz et al. 2003), ability (Goetz et al. 2003) and trust-

worthiness (Tamagawa et al. 2011).

4.2.2 Personality

Robot personality is defined as the humanlike persona

displayed by the robot. Syrdal et al. (Syrdal et al. 2007a)

make use of the Big Five Personality Factors which consist

of Emotional Stability, Extraversion, Agreeableness, Con-

scientiousness and Intellect. These personality factors can

define personality types such as friendly, shy, aggressive or

bossy (Woods et al. 2005b). Perceptions of trustworthiness

(Severinson-Eklundh et al. 2003), anthropomorphism

(Fussell et al. 2008) and level of control (Meerbeek et al.

2008) are influenced by decisions regarding robot

personality.

Perceived personality is defined as the degree to which

the robot is perceived to display humanlike personality

traits. A robot’s perceived personality type has been

shown as an anchor point upon which many other per-

ceptions are based (Powers et al. 2005). Robot form can

lead to marked differences in perceived robot personali-

ties. Perceptions of robot personality will shape a user’s

expectations and social responses and will influence their

perceptions of control and trustworthiness. It has also

been suggested that a user will anthropomorphise a polite

robot more than an impolite one (Fussell et al. 2008).

Some research has attempted to control perceptions of

personality through designed personality (i.e. controlled

system attributes), but their experiments showed little

change in perceived personality through variations in the

designed personality (Meerbeek et al. 2008). This sug-

gests that controlling for personality alone is insufficient

and that variations in other factors such as appearance,

which influence perceptions of personality, may be

required. A robot’s form (Meerbeek et al. 2006; Walters

et al. 2008a; Severinson-Eklundh et al. 2003) and ani-

macy/mobility (Severinson-Eklundh et al. 2003) have

been observed to influence user perception of personality.

In turn, this perception can influence a user’s expectations

and social responses (Goetz et al. 2003), control (Meer-

beek et al. 2006), trustworthiness (Desai et al. 2009;

Severinson-Eklundh et al. 2003) and anthropomorphism

(Fussell et al. 2008).

10 AI & Soc (2015) 30:1–21

123

4.2.3 Robot expression

Robot physical expression is defined as the ability for the

robot to use physical actions in order to communicate

through expression. Physical expression, in the form of

animated gestures, is important when establishing com-

munication between robots and humans (Kanda et al.

2002). Having the ability to express and interact with

humans in a more natural way has been shown to affect

user acceptance and perceptions of a robot (Broadbent

et al. 2009; Blow et al. 2006b). Another aspect of physical

expression is the animacy of the robot, defined as the

method and manner by which the robot moves. A robot

may move in a ‘natural’ and fluid way, or it may move in a

rigid and mechanical fashion which can affect its

Humanlike Appearance (DiSalvo et al. 2002) and Per-

ceived robot intelligence (Bartneck et al. 2009b).

Robot Social Expression is defined as the ability for a

robot to express itself socially through communication and

emotion. Humans have shown a preference to communicate

with robots in a humanlike manner, with speech being a

key natural means of interaction (Dautenhahn et al. 2005;

Sabanovic et al. 2006). Specific features of speech, such as

accent and pitch, can also directly influence trustworthiness

(Tamagawa et al. 2011). Emotion is another tool for a

robots social expression (Beer et al. 2011) and affects

perceptions of robot dominance (Kim et al. 2011), human-

likeness (DiSalvo et al. 2002; Sabanovic et al. 2006),

knowledge and sociability (Powers and Kiesler 2006),

obtrusiveness (Walters et al. 2008b), comfort (Nomura and

Kanda 2003), personality (Woods et al. 2005b) and trust-

worthiness (Tamagawa et al. 2011).

Perceived robot social expression is defined as the

degree to which a robot is perceived to display an ability to

express itself through verbal communication and emotions.

While many would consider personality as a form of social

expression, in this research it is viewed as a separate factor

due to its relative complexity, though perceived personality

and perceived robot social expression are anticipated to be

closely related. The task, or purpose, of a robot effectively

determines the need for social expressiveness. For exam-

ple, a factory-based robot does not necessarily need to

express itself in order to function correctly, while a

humanoid robot placed in a service role (e.g. assistant,

guide) needs social expression in order to interact with

users in a natural way, to also create a sense of social

presence (Heerink et al. 2008b, 2009b, 2010b) and inter-

action enjoyment (Heerink et al. 2009b, 2010b).

4.2.4 Social presence

Perceived social presence is defined as the degree to which

a user feels that they are in the presence of a socially

interactive entity. Furthermore, a consistency between

robot voice and personality has been empirically demon-

strated to positively affect perceived social presence (Lee

and Nass 2003), and without such positive reinforcement,

perceptions of social presence are expected to decrease

over time (Leite et al. 2009). Factors which can affect

perception of social presence include robot Mobility

(Syrdal et al. 2008), anthropomorphism (Schermerhorn

et al. 2008), adaptability (Syrdal et al. 2008), physical

presence (Kidd and Breazeal 2004; Bainbridge et al. 2008)

and social expressiveness (Heerink et al. 2008b). Social

presence is an important consideration as it influences a

number of different perceptions including: engagement

(Leite et al. 2009), enjoyment (Heerink et al. 2009a, b,

2010b, 2008b), social expressiveness (Heerink et al. 2010a)

and boundaries (Walters et al. 2005).

4.2.5 Mobility

Robot mobility is defined as the means and functionality of

robot locomotion. The movement speed and approach

direction of a robot have been shown to increase percep-

tions of anxiety (Hiroi and Ito 2008; Nomura et al. 2007)

and threat. The distance a robot is able to travel and how

quickly is also a point of consideration and may be a factor

which actually leads to a reduction in perceptions of

anthropomorphism. Its mobility may also, in certain situa-

tions, be considered obtrusive or intrusive, such as inter-

acting with a robot in a tight corridor (Nakauchi and

Simmons 2002) or crossing/blocking a users path (Dau-

tenhahn 2007). It can also affect anxiety (Hiroi and Ito

2008; Nomura et al. 2007) and perceptions of feeling

threatened (Dautenhahn et al. 2006). Additional locomo-

tive functions of robot mobility are likely to influence

perceptions, for example if the robot can swim/fly.

Perceived mobility is defined as the degree to which

users perceive the robots ability to transport itself freely.

Depending on how the robot is perceived to be mobile,

perceptions of personality, anthropomorphism (Severinson-

Eklundh et al. 2003) and social presence (Syrdal et al.

2008) will change.

4.2.6 Behaviour

Robot behaviour is defined as the manner with which the

robot acts during interaction. Design choices made about

the behaviour of a robot will affect how people perceive it

(Young et al. 2009; Zia-ul-Haque et al. 2007). For exam-

ple, users prefer a robot which is considerate, polite, and

which demonstrates a passive behaviour (Dautenhahn et al.

2005). Even if a robot is perceived as useful, it will only be

adopted if socially acceptable behaviour is displayed

(Walters et al. 2007), which is context and culturally

AI & Soc (2015) 30:1–21 11

123

specific. The way a robot acts can also affect perceptions of

its capabilities (Goetz et al. 2003; Lohse 2009) and its

anthropomorphism (DiSalvo et al. 2002; Sabanovic et al.

2006).

Perceived robot behaviour (PRB) is defined as the

degree to which a user perceives the manner with which a

robot acts as natural and socially acceptable. A robots

appearance will likely play an important part in influencing

a users perceptions of robot behaviour (Walters et al.

2007). Perceptions of anthropomorphism are determined,

in part, by a robots behaviour (DiSalvo et al. 2002) and

therefore also arguably determined by the way a user

perceives that behaviour. If there is a discrepancy between

human form and human behaviour, then negative user

responses are likely (Dautenhahn et al. 2005; Walters et al.

2007; Velonaki 2010). The behaviour of the robot will also

affect users expectations (Lohse 2009), where the behav-

iour is an effective way of communicating information to

users such as the robots intentions (Salvini et al. 2010).

There is still more work to be conducted exploring how

variations in the behaviour of a robot affects a users

responses (Goetz et al. 2003), particularly their emotional

and cognitive responses (Broadbent et al. 2007). The

influence of behaviour on HRI is also difficult to isolate

from the effects of the robot appearance (DiSalvo et al.

2002; Walters et al. 2007; Minato et al. 2004), but has been

found to affect user expectations (Lohse 2009).

4.3 User-centred

The last set of attributes relate to the user’s experience of

interacting with the robot, rather than the intrinsic prop-

erties of the robot itself.

4.3.1 Enjoyment

Perceived likeability/enjoyment is defined as the degree to

which a user perceives the robot to be likeable and the

interaction enjoyable (Bartneck et al. 2008). Likeability

(enjoyment) has also been shown to influence acceptance

of a robot (Heerink et al. 2008b, 2009b), but in contrast,

some research has found no evidence for the importance of

hedonic factors in robot acceptance (Klamer and Ben

Allouch 2010). Factors such as social presence (Heerink

et al. 2008b), appearance (Bartneck 2008) and acceptance

(Heerink et al. 2008b, 2009b) should be considered when

designing for likeability.

4.3.2 Ease of engagement

Perceived engagement is defined as the degree to which a

person perceives the robot to have drawn them into an

interaction and to successfully negotiate that interaction,

based on Kidd and Breazeal (2005). To function as inten-

ded, a robot must be engaging, encouraging users to will-

ingly have regular interactions (Kidd and Breazeal 2005).

Without this engagement, the robot is unlikely to retain

high degrees of social presence (Leite et al. 2009).

4.3.3 Anxiety

Perceived anxiety is defined as the degree to which a user

feels a sense of anxiety towards interacting with the robot.

A sense of familiarity, generated through previous expe-

riences, may also reduce the anxiety felt by certain users

(Bartneck 2008). Perceived anxiety has also been shown to

change with time (Nomura et al. 2007) and incite com-

munication avoidance (Nomura et al. 2006) and reduce

perceptions of usefulness (Heerink et al. 2009b, 2010b). To

prevent this, aspects of mobility (Hiroi and Ito 2008),

boundaries (Nomura et al. 2007), approach direction

(Dautenhahn et al. 2006) and familiarity (Bartneck 2008)

could be controlled.

4.3.4 Safety

Perceived safety is defined as the user’s perception of the

level of danger, or threat, when interacting with a robot

(Bartneck et al. 2008). It has been suggested that when a

robot is highly obtrusive (perhaps due to its size), the users

will perceive themselves to be secure (Dautenhahn et al.

2006), which in turn can lead to acceptance of the tech-

nology (Bartneck et al. 2008). Perceptions of safety may

also be created through the purpose of the robot, e.g. a

robot which collects physiological data about a person who

has a life-threatening illness may give the person a sense of

safety. However, if the purpose of the robot is to observe a

visitor as they simply walk around a hospital, the user may

feel there is a reason to perceive the situation as unsafe.

Previous experience with a robot (Blow et al. 2006b) and

its size (Hiroi and Ito 2008; Bartneck et al. 2008) can affect

the perceived safety of a user during their interaction with

the robot.

5 Perceptions of system attributes matrix

The research literature on HRI and monitoring was com-

prehensively explored in order to identify a series of salient

and recurring factors which are important to the interaction

and influence user behavioural responses to robots. In order

to capture as many different influencing attributes of a

robot as possible, the review explores both relevant

hypotheses and empirical studies, and the impacts of robots

of different types (though a general leaning towards

anthropomorphic), in different contexts and with different

12 AI & Soc (2015) 30:1–21

123

types of users. This provides a holistic view of robotics and

the possible areas of influence in HRI. This is beneficial as

it may capture information which could be overlooked with

a narrower search. While research from other fields is

likely to be of interest, for example software agents and

autonomous systems, the survey was limited to monitoring

and HRI in order to explicitly highlight the ways in which

these two fields specifically intersect.

It is difficult to get a sense of all of the relationships

identified above, and so the most effective way of repre-

senting and analysing the inter-relationships between the

different themes is through use of a matrix, as shown in Fig. 1.

The aim of this matrix is to clearly visualise the state of the art,

and identify areas where more research is needed. The high-

level groupings of factors are used for the columns and rows,

with references to the literature exploring that relationship.

Anthropomorphic Data Collection

Intelligence Anthropomorphism User Centr. Invasiveness Control Understanding Ownership

A nt

hr op

om or

ph ic

In te

ll ig

en ce

(Syrdal et al. 2008) (Heerink et al. 2010a;

Syrdal et al. 2008;) (Desai et al. 2009)

A n

th ro

p om

or p

h is

h m

(Bainbridge et al. 2008; Bartneck et al. 2009b;)

(Bartneck 2008; Bartneck et al. 2009a; Beer et al. 2011; Desai et

al. 2009; Fussell et al. 2008; Goetz et al. 2003; Hegel et al.

2009; Kim et al. 2011; DiSalvo et al. 2002; Meerbeek et al. 2006;

Powers and Kiesler 2006; Sabanovic et al. 2006;

Schermerhorn et al. 2008; Severinson-Eklundh et al. 2003; Syrdal et al. 2008; Walters et al.

2007; Walters et al. 2008a; Woods et al. 2005; Heerink et al.

2008; Heerink et al. 2009a; Heerink et al. 2009b; Heerink et

al. 2010a;)

(Bartneck et al. 2008; Bartneck 2008; Dautenhahn

et al. 2006; Hiroi and Ito 2008; Kanda et al. 2002; Li et al. 2010; Nomura et al. 2007; Nomura and Kanda

2003; Heerink et al. 2009b; Heerink et al. 2010a; Heerink

et al. 2010b; Walters et al. 2005;)

(Bainbridge et al. 2008; Dautenhahn 2007; Hiroi and Ito 2008; Kanda et al. 2008;

Kidd and Breazeal 2004; Nakauchi and Simmons

2002; Schermerhorn et al. 2008; Tamagawa et al. 2011; Walters et al. 2005; Walters

et al. 2008b; Young et al. 2009;)

(Meerbeek et al. 2006; Meerbeek et al. 2008;

Syrdal et al. 2006;)

(Austermann et al. 2010; Bartneck et al. 2008; Bartneck et al. 2009a;

Dautenhahn 2002; Dautenhahn 2004; Goetz et al. 2003; Lohse 2009;

Walters et al. 2007; )

(Bainbridge et al. 2008; Fussell et al. 2008; Goetz et al. 2003; Lohse 2009;

Meerbeek et al. 2006; Severinson-Eklundh et al. 2003; Tamagawa et al. 2011; Desai et al.

2009)

U se

r C

en te

re d

(Dautenhahn et al. 2006; Nomura et al. 2007;)

(Heerink et al. 2009b; Heerink et al. 2010a)

(Leite et al. 2009)

“Y” Influences…

“X ”

In fl

ue nc

ed b

y …

D at

a C

ol le

ct io

n

In va

si ve

n es

s

(Hiroi and Ito 2008; Kanda et al. 2002; Syrdal et al. 2008;)

(Dautenhahn 2007; Koay et al. 2006; Pacchierotti et al.

2008;)

(Bainbridge et al. 2008;Dautenhahn et al. 2006;Kidd and Breazeal

2004; Syrdal et al. 2008;Young et al. 2009;)

(Dautenhahn et al. 2006; Zweig 2005);

(Bohn et al. 2007; Marx 2001; Mancini et al. 2010 ;Punie 2003 ;)

C on

tr ol

(Hiroi and Ito 2008; Syrdal et al. 2007; Walters et al. 2009;)

(Desai et al. 2009)

U n

d er

st a.

(Desai et al. 2009; Melenhorst et al. 2004; Stanton 2000; Syrdal et al. 2007; Vallejo et al. 2008; Zweig and Webster 2002;)

(Desai et al. 2009; Zweig 2005;)

O w

n .

(Blow et al. 2006) (Melenhorst et al. 2004;

Zweig 2005;) (Blow et al. 2006; Dautenhahn 2007;)

Fig. 1 Relation-based matrix for anthropomorphic and data collection factors

AI & Soc (2015) 30:1–21 13

123

Moving horizontally in the matrix examines the factor

that influences the others, while moving vertically top to

bottom, examine what influences the factor (horizontal:

influences, vertical: influenced by). Initially, we will

explore the matrix factor by factor, considering both the

influence it has on other factors (horizontal) and the

influence of other factors on it (vertical).

Starting at the top left and looking from left to right, we

see that robot intelligence shows a small number of

established influences, only research related to ownership

and user-centred factors. Examining the first column, we

see that only anthropomorphism (as an external factor) has

influence on perceptions of robot intelligence. The gaps in

the matrix may be attributed to the fact that robot intelli-

gence is an inherently difficult aspect of robotics to study

effectively. This is likely due to the level of progress in the

field towards truly convincing and adaptable artificial

intelligence; while strides are being made towards a suit-

able level, the work still remains limited in the convinc-

ingness of the AI appropriate to anthropomorphic robotics.

One technique that can help with studies of the missing

relationships, particularly in relation to monitoring sys-

tems, is Wizard of Oz (Woz). This approach introduces a

human into the loop and suggests to the user that the

machine is artificially responding to them, when in fact it is

a human in control. Assuming the interaction is believable

and convincing, users are unlikely to question the efficacy

of the interaction, nor consider that they are being

deceived.

The next horizontal row shows that anthropomorphism

has the most established influence/relationships with all the

anthropomorphic and data collection factors. This suggests

that it is a well-studied aspect of social robotics and re-

affirms the clear intersection of research between social

robotics and monitoring systems. However, when looking

at anthropomorphism vertically, we see that only inva-

siveness has been identified as a point of influence. This

raises questions as to the impact of the other factors on

perceived anthropomorphism; for example, can different

degrees of control or understanding or the robots ability to

collect data impact user perceptions of its anthropomor-

phism? How might knowledge that a robot can only see

through cameras (analogous to eyes) influence the way it is

perceived? Evidence in the literature suggests that people

anthropomorphise computers in their possession and it

would be interesting to consider how this might expand to

ownership of a social robot. Related to this is also the users

emotional attachment to an object, how might this influ-

ence the way they anthropomorphise it?

Research on the influence of user-centred attributes

appears to be limited to relationships within the data col-

lection attributes. Interestingly, little work has explored the

influence of these on anthropomorphic factors. This may be

related to the fact that a user has to experience the system

to have an emotional response to it, and so detaching the

emotional response from this experience is difficult.

Examining the user-centred attributes column, we see a

different picture of the literature, where a number of the

factors have explored the ways in which user-centred

attributes are influenced by the other factors. This may be

due to the fact that users can be easily exposed to a char-

acteristic such as a robot with a large torso and have their

emotional response measured. However, to have a user in a

specific emotional state prior to this interaction is more

difficult and an understudied area.

Invasiveness appears to be the most studied in the

matrix both in terms of its influence on other factors and

what it is influenced by (horizontal and vertical). It is one

of the predominant data collection-based factors to have

been studied within the context of social robotics. When

considering the personal and intimate ways in which

social robots will be used in the future, it becomes clear

why this factor is important in terms of physical space

and privacy issues. Research examining the effect of

control on other factors is one of the most limited in the

matrix, with a focus on user-centred attributes and

ownership. This raises interesting questions on how

control might influence perceptions of intelligence

(including autonomy) and anthropomorphism: does being

able to control a robot impact the illusion of anthropo-

morphism by breaking the social norm of ‘not control-

ling’ a human? From the column of control, we can see

that a number of factors influence perception of it,

including the obvious component of control, but most

interestingly invasiveness and anthropomorphism.

Like control, the impact of understanding on other

factors is limited, with a focus on invasiveness and

ownership. These factors are predominately related to

data collection, which promotes the need for exploring

these factors in terms of humanoid robotics. Alternately,

understanding is influenced by anthropomorphism and

user-centred emotions factors, however, does not influ-

ence data attributes generally, this again suggests that the

technology is not in a sufficient state of deployment that

we can deploy it to users who can then experience it and

understand it, then informing their views of the other

factors. Ownership also has limited studies, showing

influence on invasiveness, control and user-centred attri-

butes. Even though the influence of ownership has yet to

be explored, again likely due to lack of ownership of

robots in the real world, it is interesting to note that all

the other factors have been found to influence sense of

ownership in some way.

While using the high-level groupings for factors and

their relationships in Fig. 1 helps more easily visualise the

state-of-the-art, much of the detail is lost. For this reason,

14 AI & Soc (2015) 30:1–21

123

we also include a more detailed matrix in Fig. 2, which

identifies specific relationships and the associated litera-

ture, categorised in terms of physical and social attributes.

Physical attributes concern the objective, tangible and

physical aspects of a robot/monitoring system, while the

social attributes are related to the application and interac-

tion with humans at a social level. For example, users may

be more interested in the functional aspects of the robot

(e.g. shape and things which influence physical task) than

aspects such as colour or contour which influence human-

likeness. Equally, a user may be more interested in the way

the robot behaves and talks, than what tasks it can carry out

and its physical form. Each highlighted co-ordinate/square

in the matrix represents a relationship which has either

been empirically tested or hypothesised in the literature.

Those squares which are highlighted in black represent a

relationship supported by evidence in the literature, with

the white number corresponding to a cell in Table 1.

Squares which contain the letter M are relationships which

are hypothesised or tested related to Moran and Nakata’s

(2010) work on monitoring technology. Those squares with

a ‘H’ have been hypothesised as a part of this research, and

those which are have a diagonal line or highlighted grey

represent relationships which are not logically possible.

The far left hand column of the matrix shows the physical

attributes, social attributes and user perceptions as identi-

fied in previous sections. Each is categorised into

humanoid robot aspects (HR, lighter colour), and data

collection aspects (DC, darker colour).

6 Contributions

The main contribution of this paper is in promoting dis-

cussion around anthropomorphic robots as monitoring

devices as a new thread of research. As a part of this, the

most salient attributes of a humanoid robot have been

identified, both purely in terms of robotics and also in

terms of monitoring systems which were identified through

a substantial literature review. Using this review, we con-

structed a matrix which allows us to easily visualise the

state of the art, which also helps us identify trends and

patterns for future directions for research.

7 Future work

Based on the discussions and ideas presented in this

paper, we propose a number of areas for future work.

For example, context is a critical consideration when

understanding behavioural change in robotics. A change

in context can significantly influence the perceptions of

users about a robot and its tracking abilities (Lee et al.

2011). Hence, one consideration for future research

R F

R P

E

R M

O B

R C

O N

R C

O V

R O

B

N O

R

R S

E

R P

E R

R B

E H

R IN

T

R A

D A

R A

U T

F O

D C

D I

U I

D T

A C

O N

D A

C C

D S

H A

P A

N T

P IN

T

P A

D A

P A

U T

P E

O E

P S

P

P R

B

P P

E R

P R

S E

P O

W N

P A

N X

P S

A F

P E

N J

P M

O B

A A

P N

B C

P S

B C

P F

B C

P P

I

P B

E N

P T

R U

P A

F F

P C

O V

P D

C

P R

C

Robot Form RF H H 1 H H 2 H 3 4 5 6 7 8 9 Robot Physical Expresion RPE 4131211101 H H H H

Robot Mobility RMOB H H 15 16 H Robot Control RCON M M 17 M

Robot Coverage RCOV M M Robot Obtrusion ROB 18 19 20 21 M

Number of Robots NOR MMMM Robot Social Expression RSE 22 H 23 H 24 25 26

Robot Personality RPER 27 H 28 29 30 Robot Behaviour RBEH 31 H 32

Robot Intelligence RINT H H Robot Adaptability RADA HH

Robot Autonomy RAUT H Frequency of Data Collection FODC M M

Data Integration DI M User Informedness UI M M 33 34

Data Types DT 6353 Application Control ACON M M M M 37 M

Data Access DACC M 38 M Data Sharing DSHA M

Percieved Anthropormorphism PANT 54443424140493 Percieved Intelligence PINT H

Percieved Adaptability PADA 847464 Percieved Autonomy PAUT H 49 H 50

Percieved Ease of Engagement PEOE H 51 Persieved Social Presence PSP 52 53 54 55 56

Perceived Robot Behaviour PRB 57 58 Percieved Personality PPER 59 HH 60 61 62

Percieved Robot Social Expression PRSE 4636 Perseived Ownership POWN 86766656

Percieved Anxiety PANX 69 70 Perceived Saftey PSAF

Perceived Likeability/Enjoyment PENJ Percieved Mobility PMOB 71 72 73

Application Assumptions AA 74 75 M M M Percieved Natural Border Crossings PNBC H 76 77 M 78 79 80

Percieved Social Border Crossings PSBC M M Percieved Frequency of Border Crossings PFBC M M

Percieved Privacy Invasion PPI M Perceived Benefit/Justification PBEN M 81 82 83

Percieved Trustwrothiness PTRU 84 85 Percieved Affordance PAFF M M M M

Percieved Coverage PCOV MMM Perceived Data Control PDC MMMMH

Perceived Robot Control PRC M M M 654321

F

set u

birtt A

lai c o

S s

n o it

pecre P

res U

P h

y si

ca l

A tt

ri b

u te

s H R

D C

H R

D C

H R

D C

A

B

C

D

E

snoitpecrePresUsetubirttAlaicoSsetubirttAlacisyhP

CDRHCDRHCDRH

Fig. 2 The PSA matrix displays the relationships and inter-relationships between system characteristics and user perceptions

AI & Soc (2015) 30:1–21 15

123

T a

b le

1 R

e se

a rc

h w

h ic

h c o

rr e sp

o n

d s

to th

e n

u m

b e r

c o

o rd

in a te

s in

th e

P S

A m

a tr

ix (F

ig .

2 )

N o

. R

e se

a rc

h li

te ra

tu re

N o

. R

e se

a rc

h li

te ra

tu re

N o

. R

e se

a rc

h li

te ra

tu re

N o

. R

e se

a rc

h li

te ra

tu re

1 .

B a rt

n e c k

(2 0

0 8

); B

e e r

e t

a l.

2 0

1 1

; D

iS a lv

o e t

a l.

(2 0

0 2

); H

e g

e l

e t

a l.

(2 0

0 9 );

P o

w e rs

a n

d K

ie sl

e r

(2 0

0 6

); S

a b

a n

o v

ic e t

a l.

(2 0

0 6 )

2 4

. N

o m

u ra

a n

d K

a n

d a

(2 0

0 3

) 4

7 .

H e e ri

n k

e t

a l.

(2 0

0 8

a )

7 0

H e e ri

n k

e t

a l.

(2 0

0 9

b ,

2 0

1 0

b )

2 .

W a lt

e rs

e t

a l.

(2 0

0 7

) 2

5 .

W a lt

e rs

e t

a l.

(2 0

0 8

b )

4 8

. H

e e ri

n k

e t

a l.

(2 0

0 9

b ,

2 0

1 0

b );

S y

rd a l

e t

a l.

(2 0

0 8

)

7 1

. S

y rd

a l

e t

a l.

(2 0

0 8

)

3 .

H ir

o i

a n

d It

o (2

0 0

8 );

K a n

d a

e t

a l.

(2 0

0 2

); N

o m

u ra

a n

d K

a n

d a

(2 0

0 3

); N

o m

u ra

e t

a l.

(2 0

0 7

)

2 6

. T

a m

a g

a w

a e t

a l.

(2 0

1 1 )

4 9

. D

e sa

i e t

a l.

(2 0

0 9 )

7 2

. S

e v

e ri

n so

n -E

k lu

n d

h e t

a l.

(2 0

0 3

); S

y rd

a l

e t

a l.

(2 0

0 8

)

4 .

B a rt

n e c k

e t

a l.

(2 0

0 8

); H

ir o

i a n

d It

o (2

0 0

8 )

2 7

. F

u ss

e ll

e t

a l.

(2 0

0 8 )

5 0

. D

e sa

i e t

a l.

(2 0

0 9 )

7 3

. S

e v

e ri

n so

n -E

k lu

n d

h e t

a l.

(2 0

0 3

)

5 .

B a rt

n e c k

(2 0

0 8

); L

i e t

a l.

(2 0

1 0

) 2

8 .

S e v

e ri

n so

n -E

k lu

n d

h e t

a l.

(2 0

0 3

) 5

1 .

L e it

e e t

a l.

(2 0

0 9

) 7

4 .

V a ll

e jo

e t

a l.

(2 0

0 8

)

6 .

D a u

te n

h a h

n (2

0 0

4 );

W a lt

e rs

e t

a l.

(2 0

0 7

) 2

9 .

M e e rb

e e k

e t

a l.

(2 0

0 8

); S

y rd

a l

e t

a l.

(2 0

0 6

) 5

2 .

H e e ri

n k

e t

a l.

( 2 0

1 0

a )

7 5

. D

e sa

i e t

a l.

(2 0

0 9

)

7 .

H ir

o i

a n

d It

o (2

0 0

8 );

K a n

d a

e t

a l.

(2 0

0 8 );

K id

d a n

d

B re

a z e a l

(2 0

0 4 );

W a lt

e rs

e t

a l.

(2 0

0 5

); Y

o u

n g

e t

a l.

(2 0

0 9 )

3 0

. M

e e rb

e e k

e t

a l.

(2 0

0 8

); S

y rd

a l

e t

a l.

(2 0

0 6

) 5

3 .

H e e ri

n k

e t

a l.

(2 0

0 8

b ,

2 0

0 9

a )

7 6

. B

a in

b ri

d g

e e t

a l.

(2 0

0 8

);

K id

d a n

d B

re a z e a l

(2 0

0 4 )

8 .

B a in

b ri

d g

e e t

a l.

(2 0

0 8

); T

a m

a g

a w

a e t

a l.

(2 0

1 1

) 3

1 .

B a rt

n e c k

e t

a l.

(2 0

0 9

a );

D iS

a lv

o e t

a l.

(2 0

0 2

); S

a b

a n

o v

ic

e t

a l.

(2 0

0 6

)

5 4

. H

e e ri

n k

e t

a l.

(2 0

0 9

b ,

2 0

1 0

b );

W a lt

e rs

e t

a l.

(2 0

0 5 )

7 7

. D

a u

te n

h a h

n e t

a l.

(2 0

0 6

)

9 .

D a u

te n

h a h

n (2

0 0

4 );

W a lt

e rs

e t

a l.

(2 0

0 7

) 3

2 .

G o

e tz

e t

a l.

(2 0

0 3

); L

o h

se (2

0 0

9 )a

5 5

. B

a in

b ri

d g

e e t

a l.

(2 0

0 8 )

7 8

. S

y rd

a l

e t

a l.

(2 0

0 8

)

1 0

. D

a u

te n

h a h

n (2

0 0

7 );

N a k

a u

c h

i a n

d S

im m

o n

s (2

0 0

2 )

3 3

. Z

w e ig

a n

d W

e b

st e r

(2 0

0 2

) 5

6 .

D e sa

i e t

a l.

(2 0

0 9 )

7 9

. B

o h

n e t

a l.

(2 0

0 7

); M

a n

c in

i

e t

a l.

(2 0

1 0

); M

a rx

(2 0

0 1

); P

u n

ie (2

0 0

3 )

1 1

. D

iS a lv

o e t

a l.

(2 0

0 2

); G

o e tz

e t

a l.

(2 0

0 3

);

S e v

e ri

n so

n -E

k lu

n d

h e t

a l.

(2 0

0 3

)

3 4

. D

e sa

i e t

a l.

(2 0

0 9

) 5

7 D

iS a lv

o e t

a l.

(2 0

0 2

) 8

0 .

Z w

e ig

(2 0

0 5

)

1 2

. B

a rt

n e c k

e t

a l.

(2 0

0 9

b )

3 5

. S

y rd

a l

e t

a l.

(2 0

0 7

b )

5 8

. F

u ss

e ll

e t

a l.

(2 0

0 8

) 8

1 .

M e le

n h

o rs

t e t

a l.

(2 0

0 4

);

S y

rd a l

e t

a l.

(2 0

0 7

b )

1 3

. B

a in

b ri

d g

e e t

a l.

(2 0

0 8

) 3

6 .

S y

rd a l

e t

a l.

(2 0

0 7

b )

5 9

. D

e sa

i e t

a l.

(2 0

0 9 );

S e v

e ri

n so

n -

E k

lu n

d h

e t

a l.

(2 0

0 3 )

8 2

. S

ta n

to n

(2 0

0 0

)

1 4

. S

e v

e ri

n so

n -E

k lu

n d

h e t

a l.

(2 0

0 3 )

3 7

. D

e sa

i e t

a l.

(2 0

0 9

) 6

0 .

M e e rb

e e k

e t

a l.

(2 0

0 6 )

8 3

. Z

w e ig

(2 0

0 5

)

1 5

. S

e v

e ri

n so

n -E

k lu

n d

h e t

a l.

(2 0

0 3 )

3 8

. V

a rg

a s

e t

a l.

(2 0

0 9 )

6 1

. M

e e rb

e e k

e t

a l.

(2 0

0 6 )

8 4

. Z

w e ig

(2 0

0 5

)

1 6

. D

a u

te n

h a h

n e t

a l.

(2 0

0 6 );

H ir

o i

a n

d It

o (2

0 0

8 );

N o

m u

ra a n

d K

a n

d a

(2 0

0 3

); N

o m

u ra

e t

a l.

(2 0

0 7

)

3 9

. W

a lt

e rs

e t

a l.

(2 0

0 9

b )

6 2

. M

e e rb

e e k

e t

a l.

(2 0

0 6 )

8 5

. M

e le

n h

o rs

t e t

a l.

(2 0

0 4

)

16 AI & Soc (2015) 30:1–21

123

exploring the effects of data collection on users is the

context within in which it takes place. Variations in the

types of contexts studied and explored can only be a

good thing. In relation to the context is the task of a

robot, where it might be able to support users, instruct

them or act autonomously. In fact, the design of a robot

and its use is predominately determined by its task

(Yanco and Drury 2004).

The majority of studies and research examined in this

paper have mainly explored the social aspects of robotics

in experimental conditions. There is a current drive

towards exploring the impact of technology ‘in the wild’,

and understanding how the technology actually influences

people in situ can help improve our understanding.

Researchers need to move beyond the laboratory and begin

to see the technology they develop directly in use; which

will likely reveal rich and unexpected interactions.

One key aspect to HRI which is clearly missing re-

longitudinal studies (Kanda et al. 2004; Walters et al.

2008b), which are important to get past the novelty effect

(Sung et al. 2009). It is important to understand how

longitudinal effects change over time, and this could be in

terms of experimental studies or in situ studies. The idea

that humanoid robots afford certain capabilities in terms of

monitoring characteristics is an understudied area. More

work is needed to understand what people think robots are

capable of perceiving and judging, and the benefits and

disadvantages that come with this. For example, consider

the affordances that derive from a robot with human eyes.

Interestingly, robots can be augmented with other systems

(such as smart homes and surveillance systems) which give

them abilities beyond what is afforded.

How do robots collect data and whom and what do they

share with? What about passersby, where there is a sense

of passive collection of personal data? Environment the

robots are in, what about that? There are many consider-

ations to made regarding robots as a monitoring system.

The existing factors were derived from a survey of the

robotics and monitoring literature, and the areas that they

intersect. This survey was consciously restricted to these

areas so as to not dilute. The list of influential factors is

unlikely to be exhaustive and is expected to change; hence,

one direction is to look to other relevant areas such as

interactions with autonomous and expert systems. It will

expand and develop as progress is made.

The final area is to understand and examine how, and if

progress is made in this area of interest. One proposition is

the development of an online repository along the lines of

the matrix outlined in this paper. This would allow

researchers to identify areas they might contribute towards

and to also keep track of where other contributions. A form

of evaluating the advancement of this area might then be to

observe progress within such a repository.T a

b le

1 c o

n ti

n u

e d

N o

. R

e se

a rc

h li

te ra

tu re

N o

. R

e se

a rc

h li

te ra

tu re

N o

. R

e se

a rc

h li

te ra

tu re

N o

. R

e se

a rc

h li

te ra

tu re

1 7

. H

ir o

i a n

d It

o (2

0 0

8 );

W a lt

e rs

e t

a l.

(2 0

0 9

a )

4 0

. S

c h

e rm

e rh

o rn

e t

a l.

(2 0

0 8

) 6

3 .

H e e ri

n k

e t

a l.

(2 0

0 8

b ,

2 0

0 9

b ,

2 0

1 0

b )

1 8

. D

a u

te n

h a h

n e t

a l.

(2 0

0 6 )

4 1

. G

o e tz

e t

a l.

(2 0

0 3

); M

e e rb

e e k

e t

a l.

(2 0

0 6

);

S c h

e rm

e rh

o rn

e t

a l.

(2 0

0 8

); S

e v

e ri

n so

n -E

k lu

n d

h e t

a l.

(2 0

0 3

); W

a lt

e rs

e t

a l.

(2 0

0 8

a )

6 4

H e e ri

n k

e t

a l.

(2 0

0 9

b ,

2 0

1 0

b )

1 9

. H

ir o

i a n

d It

o (2

0 0

8 );

K a n

d a

e t

a l.

(2 0

0 2

); S

y rd

a l

e t

a l.

(2 0

0 8 )

4 2

. A

u st

e rm

a n

n e t

a l.

(2 0

1 0

); B

a rt

n e c k

e t

a l.

(2 0

0 8

);

B a rt

n e c k

e t

a l.

(2 0

0 9

a );

D a u

te n

h a h

n (2

0 0

2 ,

2 0

0 4

);

G o

e tz

e t

a l.

(2 0

0 3

); W

a lt

e rs

e t

a l.

(2 0

0 7

)

6 5

. B

lo w

e t

a l.

(2 0

0 6

b )

2 0

. D

a u

te n

h a h

n (2

0 0

7 );

K o

a y

e t

a l.

(2 0

0 6

); P

a c c h

ie ro

tt i

e t

a l.

(2 0

0 6 )

4 3

. S

c h

e rm

e rh

o rn

e t

a l.

(2 0

0 8

); T

a m

a g

a w

a e t

a l.

(2 0

1 1 )

6 6

. B

lo w

e t

a l.

(2 0

0 6

b )

2 1

. K

id d

a n

d B

re a z e a l

(2 0

0 4

); Y

o u

n g

e t

a l.

(2 0

0 9

) 4

4 .

G o

e tz

e t

a l.

(2 0

0 3

); L

o h

se (2

0 0

9 )a

6 7

. B

lo w

e t

a l.

(2 0

0 6

b );

D a u

te n

h a h

n (2

0 0

7 )

2 2

. B

a rt

n e c k

e t

a l.

(2 0

0 9

a );

D iS

a lv

o e t

a l.

(2 0

0 2

),

S a b

a n

o v

ic e t

a l.

(2 0

0 6

)

4 5

. A

u st

e rm

a n

n e t

a l.

(2 0

1 0

) 6

8 .

B lo

w e t

a l.

(2 0

0 6

b );

D a u

te n

h a h

n (2

0 0

7 )

2 3

. K

im e t

a l.

(2 0

1 1

); W

o o

d s

e t

a l.

(2 0

0 5

b )

4 6

. S

y rd

a l

e t

a l.

(2 0

0 8

) 6

9 .

D a u

te n

h a h

n e t

a l.

(2 0

0 6 );

N o

m u

ra

e t

a l.

(2 0

0 7

)

AI & Soc (2015) 30:1–21 17

123