On-line Monitoring and Diagnosing a Team of Service Robots: the RoboCare Experience

The paper addresses the tasks of monitoring and diagnosing the actions concurrently performed by a team of plan executors in a partially observable environment. The paper formalizes a distributed approach where each plan executor is associated with a supervisor agent, responsible for keeping track of the executor state (i.e., a belief state given the partial observability) both under nominal and anomalous conditions. Since plan executors compete for accessing the available resources, interactions among them may arise during plan execution. The paper proposes and fomralizes a methodology, based on the notion of dependency set, for dynamically capturing interactions among executors and deriving local belief states which are globally consistent.