Enhancing Plan Execution in Dynamic Domains Using Model-Based Reasoning

If an autonomous mobile robot has to perform a really complex task like setting the table for dinner, it has to have capabilities for planning and reasoning in order to be able to successfully finish the task. For the calculation of a plan for a given goal there exist a number of suitable algorithms. But if such a plan is executed on an autonomous mobile robot in a dynamic environment, a number of problems are likely to occur. Beside the problems caused by the assumption used in the planning phase problems arise trough inaccurate sensing, acting and events which are not under control of the robot. All these problems have in common that they cause an inconsistency between the intentions of the plan and the observed world. In this paper we propose model-based diagnosis as a method for the detection and the categorization of such inconsistencies. The obtained knowledge about failures in plan execution and about their root causes can be used to monitor plan execution. Such monitoring together with appropriate repair actions improves the robustness of the execution of plans in dynamic environments and thus improves robustness of autonomous mobile robots.