🤖 AI Summary
This work addresses the challenge of enabling safety-critical multi-agent systems to collaboratively maintain dynamic communication links among mobile targets under partial observability. We propose a control-theoretic safety-aware reinforcement learning framework. Its core innovation lies in the first principled integration of Control Barrier Functions (CBFs) with multi-agent deep reinforcement learning, introducing an edge-level safety-activation mechanism that guides message passing in graph neural networks—thereby jointly optimizing local safety constraints and decentralized policy learning. Evaluated on a dynamic bridging task, the method significantly enhances safety and robustness: collision rate decreases by 76% in high-density scenarios, while communication connectivity is maintained at 92.4%. The framework establishes a verifiable and deployable paradigm for partially observable, safety-sensitive multi-agent coordination.
📝 Abstract
Addressing complex cooperative tasks in safety-critical environments poses significant challenges for multi-agent systems, especially under conditions of partial observability. We focus on a dynamic network bridging task, where agents must learn to maintain a communication path between two moving targets. To ensure safety during training and deployment, we integrate a control-theoretic safety filter that enforces collision avoidance through local setpoint updates. We develop and evaluate multi-agent reinforcement learning safety-informed message passing, showing that encoding safety filter activations as edge-level features improves coordination. The results suggest that local safety enforcement and decentralized learning can be effectively combined in distributed multi-agent tasks.