Reinforcement Learning

SRMT: Shared Recurrent Memory Transformer boosts multi-agent coordination by implicitly sharing information via a global memory, significantly outperforming baselines in complex pathfinding tasks.