A probabilistic safety-aware strategy in stochastic multi-agent reach-avoid games

In this paper, a defender-focused stochastic model for multi-agent systems is considered in a reach-avoid game. During the pursuit-evasion process between defenders and attackers, the problem of collision avoidance with both obstacles and other defenders is addressed. First, a safety-aware predictive planning method is proposed for stationed attackers. Obstacle avoidance and finite-time stability are guaranteed in the probabilistic game based on the strategy from zero-sum game analysis. Then, for pursuit of maneuvering attackers, a collision-free cross-pursuit strategy is used to achieve successful pursuit and safety of the defenders at the same time. A high-probability pursuit region is computed based on stochastic reachable sets. When there is a potential of conflict between two defenders, a collision-free re-allocation is initiated to guarantee the completion of the pursuit under the influence of flow-field disturbances. The feasibility and performance of the proposed strategy is demonstrated by simulation results.