Managing Dynamic Multi-Agent Simple Temporal Network

The realization of plans of activities by several agents is usually subject to a set of temporal constraints. The latter are either constraints holding on each agent individually, or constraints allowing the activities of agents to be synchronized. To represent the set of temporal constraints imposed on distributed plans, the framework of Multi-agent Simple Temporal Network (MaSTN) can be used. In this paper, we consider the problem of maintaining the temporal consistency of distributed plans during execution, when temporal constraints may be updated. To this end, we propose new incremental algorithms for managing dynamic MaSTNs. These algorithms help each agent know, as soon as possible, whether the distributed plan to be executed is still temporally consistent. They range from algorithms focusing on keeping as much as possible the privacy between agents to algorithms focusing on improving the response time to updates through more information sharing. We analyze the robustness of these algorithms when communications are intermittent, and we provide experimental results demonstrating the trade-off to be made between performance and privacy.