Artificial Intelligence

def env2statespace(env): """ This simple parser demonstrates how you can extract the state space from the Open AI env We *assume* full observability, i.e., we can directly ignore Hole states. Alternatively, we could place a very high step cost to a Hole state or use a directed representation (i.e., you can go to a Hole state but never return). Feel free to experiment with both if time permits. Input: env: an Open AI Env follwing the std in the FrozenLake-v0 env Output: state_space_locations : a dict with the available states state_space_actions : a dict of dict with available actions in each state state_start_id : the start state state_goal_id : the goal state These objects are enough to define a Graph problem using the AIMA toolbox, e.g., using UndirectedGraph, GraphProblem and astar_search (as in AI (H) Lab 3) Notice: the implementation is very explicit to demonstarte all the steps (it could be made more elegant!) """ state_space_locations = {} # create a dict for i in range(env.desc.shape[0]): for j in range(env.desc.shape[1]): if not (b'H' in env.desc[i,j]): state_id = "S_"+str(int(i))+"_"+str(int(j) ) state_space_locations[state_id] = (int(i),int(j)) if env.desc[i,j] == b'S': state_initial_id = state_id elif env.desc[i,j] == b'G': state_goal_id = state_id #-- Generate state / action list --# # First define the set of actions in the defined coordinate system actions = {"west": [-1,0],"east": [+1,0],"north": [0,+1], "south": [0,-1]} state_space_actions = {} for state_id in state_space_locations: possible_states = {} for action in actions: #-- Check if a specific action is possible --# delta = actions.get(action) state_loc = state_space_locations.get(state_id) state_loc_post_action = [state_loc[0]+delta[0],state_loc[1]+delta[1]] #-- Check if the new possible state is in the state_space, i.e., is accessible --# state_id_post_action = "S_"+str(state_loc_post_action[0])+"_"+str(state_loc_post_action[1]) if state_space_locations.get(state_id_post_action) != None: possible_states[state_id_post_action] = 1 #-- Add the possible actions for this state to the global dict --# state_space_actions[state_id] = possible_states return state_space_locations, state_space_actions, state_initial_id, state_goal_id