A local consensus based algorithm for distributed multi-agent path finding

Distributed multi-agent path finding (MAPF) is an important research problem in multi-agent systems. However, existing methods struggle to handle complex interactions in high-density scenarios due to insufficient conflict awareness and limited cooperative path-decoupling capability, leading to planning failures or increased path costs. To address these issues, we propose Local Consensus-based Priority Inheritance with Backtracking (LC-PIBT), an efficient rule-based distributed path planning method. LC-PIBT enables on-demand transmission of neighborhood information through an event-triggered state propagation mechanism. Based on shared information, we further introduce a path–topology correlation analysis framework for conflict detection and coordination, which effectively mitigates deadlocks and livelocks. In addition, we analyze cascading conflicts and design a multi-level heuristic decision-making strategy to reduce unnecessary movements and suppress conflict escalation. Experimental results show that LC-PIBT achieves a 100% success rate across four benchmark maps, with an average decision time below 1 ms, and attains the lowest path costs in most test scenarios. LC-PIBT provides an efficient and scalable solution for distributed MAPF.