Reasoning with Conditional Plans in the Presence of Incomplete Knowledge

We present a simple set of inference rules for reasoning about the effects of actions in a conditional plan. The rules allow us to make additional conclusions about a plan at every stage of its execution, and to augment an agent’s knowledge state in the presence of incomplete knowledge. We can use the rules to refine an agent’s knowledge after a particular execution of a plan has completed, and to improve an agent’s ability to generate plans. Furthermore, this enhancement gives us the ability to plan for certain types of temporally oriented goals, e.g., goals that require some initial state condition be restored by the end of the plan. We have implemented this mechanism inside of a planner, and we demonstrate the planner’s increased ability to solve a variety of interesting planning problems.