Title: SPARC: Spatial-Aware Path Planning via Attentive Agent Communication

Abstract: While efficient communication is paramount for decentralized Multi-Robot Path Planning (MRPP), current learned approaches often fail to distinguish between neighbors based on their physical closeness, treating all agents as equally important. This oversight results in diluted attention precisely where coordination is most crucial: in congested areas. To address this, we introduce Relation enhanced Multi Head Attention (RMHA), a novel communication framework that incorporates pairwise Manhattan distances directly into the attention weight calculation. This allows robots to dynamically focus on messages from spatially pertinent neighbors. By integrating a distance-constrained attention mask and GRU-based message fusion, RMHA works seamlessly with MAPPO to ensure stable end-to-end training. Our method demonstrates robust zero-shot generalization, scaling from 8 training robots to 128 test robots on 40x40 grids. At an obstacle density of 30%, RMHA achieves a success rate of roughly 75%, surpassing the strongest baseline by more than 25 percentage points. Furthermore, ablation studies highlight that the encoding of distance relations is the primary driver behind the performance gains observed in high-density scenarios.

Index Terms: Multi-robot path planning, graph attention mechanism, multi-head attention, communication optimization, cooperative decision-making

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC