In this paper, we propose combining a planner with a rea-soner to help an agent handling failures during its plan execution. This is achieved even when the planner cannot come up with a valid plan considering the domain operators due to lack of exact information on the actual outcomes of actions that may resolve the failure. We show that if a reasoning tool is not available to the agent, the problem could be resolved at the expense of irrelevant actions. The intuitive idea that is applied in this case is finding actions that may change the state of the failure. If a reasoner, PROBCOG in our case, is provided, then more efficient solutions can be attained. Our ongoing work includes a method to find actions that have direct relations to the cause of failure. By using this approach, irrelevant actions are not selected anymore.
展开▼