Training Small LLMs as Spatial Multi-Agent Policies

Yi Mao, Andrew Perrault

Training LLM-based multi-agent systems with multi-agent reinforcement learning is rapidly gaining traction, and a parallel line of work argues that such systems should be judged by…