High Level Interpretation of Action Failures in Multi-Robot Systems

The paper addresses the problem of on-line monitoring and diagnosis of multi-robot systems. These tasks are performed by a Diagnostic Module whose high-level architecture is introduced: the on-line monitoring exploits a model of the system (robots + environment) based on automata. An assessment of the system status is performed during the monitoring phase, while a high-level interpretation of the observed behavior of the system (in particular failures in completing the actions assigned to the robots) is done by a Diagnostic Interpretation module. The interpretation process is performed by exploiting a Bayesian network which allows us to relate system observations with abstract concepts in terms of faults and troublesome interactions. Finally, the paper reports some preliminary experimental results concerning the performance and the competence of the Diagnostic Module in the RoboCare domain.