Probabilistic Mission Planning and Analysis for Multi-agent Systems

Mission planning is one of the crucial problems in the design of autonomous Multi-Agent Systems (MAS), requiring the agents to calculate collision-free paths and efficiently schedule their tasks. The complexity of this problem greatly increases when the number of agents grows, as well as timing requirements and stochastic behavior of agents are considered. In this paper, we propose a novel method that integrates statistical model checking and reinforcement learning for mission planning within such context. Additionally, in order to synthesise mission plans that are statistically optimal, we employ hybrid automata to model the continuous movement of agents and moving obstacles, and estimate the possible delay of the agents' travelling time when facing unpredictable obstacles. We show the result of synthesising mission plans, analyze bottlenecks of the mission plans, and re-plan when pedestrians suddenly appear, by modeling and verifying a real industrial use case in UPPAAL SMC.
DPAC - Dependable Platforms for Autonomous systems and Control