Proactive Conflict Area Prediction for Boosting Search-Based Multi-Agent Pathfinding

Abstract

The multi-agent pathfinding aims to compute conflict-free paths for multiple agents in shared environments. Traditional methods, such as conflict-based search (CBS), guarantee optimality but suffer from high computational costs due to constraint tree expansion. Learning-based approaches improve efficiency but often compromise solution quality. We propose proactive conflict-aware prediction (PCAP), which improves CBS by predicting conflict-prone areas based on constraint data. This approach enables a more informed constraint application, reducing unnecessary expansions while preserving optimality. Experimental results show that PCAP reduces computation time by 40% compared to CBS while maintaining solution quality, making it a scalable and effective approach for complex MAPF scenarios.