Feedback 14-07-12

Progress

• Changed the light-dark domain such that it locates the agent at a randomly selected starting cell at the beginning of an episode
• Implemented a heuristic for the light-dark domain:
• Assumes the agent is on the grid cell for which the belief is the highest
• Takes the action that has the smallest Euclidean distance from that cell to the goal cell

Experimental Set-up

• Light Dark domain with the following layout:

********L*
********L*
********L*
********L*
********L*
***G****L*
********L*
********L*
********L*
********L* 

• At the beginning of each episode, the agent is randomly placed at any cell except the goal G
• Initial belief: uniform over all possible states, except the goal state
• Number of episodes: 1000
• Discount factor: 0.95
  

Results