Plan Execution and Recovery in Multi Agent Systems for Space Applications

The paper addresses the problem of supervising the execution of actions concurrently executed by a team of cooperating agents in a partially observable environment. Since the agents cooperate by exchanging services or by executing joint-actions, causal dependencies among actions arise. As a consequence a failure in the action of an agent can have harmful effects on the execution of the actions assigned to other agents. The paper formalizes a distributed approach to the supervision task where each agent supervises its own actions (sub-plan). In presence of an action failure, the supervision task aims at singling out a subset of goals (missing goals) which an agent is no longer able to achieve and which have to be achieved for overcoming the harmful effects of the failure.