Tackling Safe and Efficient Multi-Agent Reinforcement Learning via Dynamic Shielding (Student Abstract)

Multi-agent Reinforcement Learning (MARL) has been increasingly used in safety-critical applications but no safety guarantees, especially during training. In this paper, we propose dynamic shielding, a novel decentralized MARL framework to ensure both training and deployment phases. Our leverages Shield, reactive system running parallel with the reinforcement learning algorithm monitor correct agents' behavior. our algorithm, shields dynamically split merge according environment state order maintain decentralization avoid conservative behaviors while enjoying formal guarantees. We demonstrate effectiveness of shielding mobile navigation scenario.