CS 289, Spring 1999
Assignment 3: Probabilistic reasoning, due 4/1




1. The inputs to a plan execution simulator are (a) a set of action descriptions in PDDL syntax; (b) a complete initial state description; (c) a plan expressed as a sequence of actions. Write a simulator that computes the final state description resulting from executing the plan in the initial state. The simulator should check all action preconditions.

2. Using a planner of your choice (see Weld survey for planners available on the web), construct a plan for fixing a flat tire given the domain description. (Note: the file contains variants that may be suitable for specific planners; there is also a series of subproblem definitions in case the planner can't handle the whole thing at once.) Show that the plan works using the simulator.

3. Modify the simulator so that with probability 0.1, an action fails -- i.e., one of the effects does not occur. Show an example of a plan not working due to an action failure.

4. Write a simple replanning agent that is able to plan a course of action given a goal, detect action failure, replan accordingly, and resume execution. Show an example of your agent overcoming a failed action.

5. Discuss shortcomings in the representation of the tire world, and explore the extent to which they can be fixed within the syntactic constraints of PDDL-based planners.