Learning Selective Communication for Multi-Agent Path Finding

Learning communication via deep reinforcement learning (RL) or imitation (IL) has recently been shown to be an effective way solve Multi-Agent Path Finding (MAPF). However, existing based MAPF solvers focus on broadcast communication, where agent broadcasts its message all other predefined agents. It is not only impractical but also leads redundant information that could even impair the multi-agent cooperation. A succinct scheme should learn which relevant and influential each agent’s decision making process. To address this problem, we consider a request-reply scenario propose xmlns:xlink="http://www.w3.org/1999/xlink">Decision Causal Communication (DCC), simple yet efficient model enable agents select neighbors conduct during both training execution. Specifically, neighbor determined as when presence of causes adjustment central agent. This judgment learned local observation thus suitable for decentralized execution handle large scale problems. Empirical evaluation in obstacle-rich environment indicates high success rate with low overhead our method.