Resilient adaptive optimal control of distributed multi-agent systems using reinforcement learning

摘要

This study presents a unified resilient model-free reinforcement learning (RL) based distributed control protocol for leader-follower multi-agent systems. Although RL has been successfully used to learn optimal control protocols for multi-agent systems, the effects of adversarial inputs are ignored. It is shown in this study, however, that their adverse effects can propagate across the network and impact the learning outcome of other intact agents. To alleviate this problem, a unified RL-based distributed control frameworks is developed for both homogeneous and heterogeneous multi-agent systems to prevent corrupted sensory data from propagating across the network. To this end, only the leader communicates its actual sensory information and other agents estimate the leader' state using a distributed observer and communicate this estimation to their neighbours to achieve consensus on the leader state. The observer cannot be physically affected by any adversarial input. To further improve resiliency, distributed H_∞control protocols are designed to attenuate the effect of the adversarial inputs on the compromised agent itself. An off-policy RL algorithm is developed to learn the solutions of the game algebraic Riccati equations arising from solving the H_∞control problem. No knowledge of the agent's dynamics is required and it is shown that the proposed RL-based H_∞control protocol is resilient against adversarial inputs.