Data-Driven Safe Formation Control for Multi-Agent Systems and Its Applications in Multi-UGV Systems

In this paper, we propose a reinforcement learning-based safe control strategy for uncertain heterogeneous multi-agent systems. The objective is to achieve collision-free time-varying formations under switching topologies and with limited network resources. Without requiring global communication information, an event-triggered observer is designed to decouple the heterogeneous dynamics from switching networks and reduce data transmission frequency. A data-driven off-policy reinforcement learning algorithm is developed for addressing the robust safe formation control problem. The algorithm is capable of solving non-quadratic optimization problems without requiring model information. The proposed strategy is applied to a multi-unmanned ground vehicle (UGV) system for a patrolling mission, and the experimental results verified the effectiveness of the proposed strategy.