Learning-based formal synthesis of cooperative multi-agent systems with an application to robotic coordination

In our previous work [4], we proposed a learning-based formal top-down design framework to automatically synthesize coordination and control strategies for cooperative multi-agent systems. Our main idea is to decompose a given team mission into individual local tasks and synthesizing local supervisors while guarantee the multi-agent performance by incorporating supervisor synthesis with compositional verification techniques. In this paper, we apply the top-down design framework to a multi-robot scenario that involves both request-response services and multi-robot coordination. Modified L* algorithms are adapted to both the local synthesis and the compositional verification to ensure that the collective behavior of the robots will eventually guarantee the satisfaction of the global specification. Computational and software tools are developed to integrate automatic supervisor synthesis and interrobot communication.