Heterogeneous Multi-Agent Communication Learning via Graph Information Maximization

Communication learning is an effective way to solve complicated cooperative tasks in multi-agent reinforcement learning (MARL) domain. Graph neural network (GNN) has been widely adopt for learning the multi-agent communication and various GNN-based MARL methods have emerged. However, most of these methods are not specially designed for heterogeneous multi-agent scenarios, where agents have heterogeneous attributes or features based on different observation spaces or action sets. Without effective processing and transmission of heterogeneous feature information, communication learning will be useless and even reduce the performance of cooperation.