Final Project Ideas

• Find optimal risk-sensitive policies for racetrack.
• Implement SVI for deterministic domains.
• Use Bayes net algorithm to evaluate a probabilistic STRIPS plan.
• Implement SATplan, compare to stochastic search in plan space.
• Generate good probabilistic policies for unobservable domains like grid world (woods7).
• Find optimal memory-based policies for unobservable deterministic domains like grid world (woods7).
• Solve a huge MDP like racetrack using techniques from numerical analysis.
• Use informed search to find optimal plans for inaccessible domains.
• Use hill-climbing, simulated annealing, and/or genetic algorithms to find good policies.
• Encode and solve Tetris.
• Create a problem that is hard for policy iteration.

Community Service Projects

• Convert ``POMDP'' format into our format.
• Convert ``graphplan'' format into our format.
• Create user interface for animating our format.