Message-Aware Graph Attention Networks for Large-Scale Multi-Robot Path Planning

The domains of transport and logistics are increasingly relying on autonomous mobile robots for the handling distribution passengers or resources. At large system scales, finding decentralized path planning coordination solutions is key to efficient performance. Recently, Graph Neural Networks (GNNs) have become popular due their ability learn communication policies in multi-agent systems. Yet, vanilla GNNs rely simplistic message aggregation mechanisms that prevent agents from prioritizing important information. To tackle this challenge, letter, we extend our previous work utilizes by incorporating a novel mechanism allow message-dependent attention. Our Message-Aware Attention neTwork (MAGAT) based key-query-like determines relative importance features messages received various neighboring robots. We show MAGAT able achieve performance close coupled centralized expert algorithm. Further, ablation studies comparisons several benchmark models attention very effective across different robot densities performs stably constraints bandwidth. Experiments demonstrate model generalize well previously unseen problem instances, it achieves 47% improvement over success rate, even large-scale instances ×100 larger than training instances.