Anytime Multi-Agent Path Finding with an Adaptive Delay-Based Heuristic
Thomy Phan, Benran Zhang, Shao-Hung Chan, Sven Koenig
Abstract
Anytime multi-agent path finding (MAPF) is a promising approach to scalable and collision-free path optimization in multi-agent systems. MAPF-LNS, based on Large Neighborhood Search (LNS), is the current state-of-the-art approach where a fast initial solution is iteratively optimized by destroying and repairing selected paths of the solution. Current MAPF-LNS variants commonly use an adaptive selection mechanism to choose among multiple destroy heuristics. However, to determine promising destroy heuristics, MAPF-LNS requires a considerable amount of exploration time. As common destroy heuristics are stationary, i.e., non-adaptive, any performance bottleneck caused by them cannot be overcome by adaptive heuristic selection alone, thus limiting the overall effectiveness of MAPF-LNS. In this paper, we propose Adaptive Delay-based Destroy-and-Repair Enhanced with Success-based Self-learning (ADDRESS) as a single-destroy-heuristic variant of MAPF-LNS. ADDRESS applies restricted Thompson Sampling to the top-K set of the most delayed agents to select a seed agent for adaptive LNS neighborhood generation. We evaluate ADDRESS in multiple maps from the MAPF benchmark set and demonstrate cost improvements by at least 50% in large-scale scenarios with up to a thousand agents, compared with the original MAPF-LNS and other state-of-the-art methods.
BibTeX
@article{Phan_Zhang_Chan_Koenig_2025, title={Anytime Multi-Agent Path Finding with an Adaptive Delay-Based Heuristic}, volume={39}, url={https://ojs.aaai.org/index.php/AAAI/article/view/34495}, DOI={10.1609/aaai.v39i22.34495}, abstractNote={Anytime multi-agent path finding (MAPF) is a promising approach to scalable and collision-free path optimization in multi-agent systems. MAPF-LNS, based on Large Neighborhood Search (LNS), is the current state-of-the-art approach where a fast initial solution is iteratively optimized by destroying and repairing selected paths of the solution. Current MAPF-LNS variants commonly use an adaptive selection mechanism to choose among multiple destroy heuristics. However, to determine promising destroy heuristics, MAPF-LNS requires a considerable amount of exploration time. As common destroy heuristics are stationary, i.e., non-adaptive, any performance bottleneck caused by them cannot be overcome by adaptive heuristic selection alone, thus limiting the overall effectiveness of MAPF-LNS.
In this paper, we propose Adaptive Delay-based Destroy-and-Repair Enhanced with Success-based Self-learning (ADDRESS) as a single-destroy-heuristic variant of MAPF-LNS. ADDRESS applies restricted Thompson Sampling to the top-K set of the most delayed agents to select a seed agent for adaptive LNS neighborhood generation. We evaluate ADDRESS in multiple maps from the MAPF benchmark set and demonstrate cost improvements by at least 50% in large-scale scenarios with up to a thousand agents, compared with the original MAPF-LNS and other state-of-the-art methods.}, number={22}, journal={Proceedings of the AAAI Conference on Artificial Intelligence}, author={Phan, Thomy and Zhang, Benran and Chan, Shao-Hung and Koenig, Sven}, year={2025}, month={Apr.}, pages={23286-23294} }