Scalable Constrained Multi-Agent Reinforcement Learning via State Augmentation and Consensus for Separable Dynamics

In three linesDistributed approach for constrained multi-agent reinforcement learning combining state-augmented policy learning with consensus over Lagrange multipliers. Agents learn offline policies and coordinate via local communication. Linear scalability to thousands of agents, demonstrated on smart grid demand response.

Summary generated by Claude — human-verified