Learning-based fast alternating direction method of multipliers for multi-agent path finding using temporary variable-fixing

doi:10.1016/j.ejor.2026.04.028

Learning-based fast alternating direction method of multipliers for multi-agent path finding using temporary variable-fixing

2026-04-22

0

PRE

AI

DOI：10.1016/j.ejor.2026.04.028

Original

Original request for help

Save

Abstract

En 中文

• We propose a novel learning-based fast ADMM for MAPF. • Fast ADMM is developed using tailored Dijkstra and soft start. • Imitation learning is used to build a policy network to accelerate fast ADMM. • Our method achieves considerable speedup with only a tiny loss in solution quality.

Keywords:

multi-agent path finding

alternating direction method of multipliers

imitation learning

fast ADMM

temporary variable-fixing

Learning-based fast alternating direction method of multipliers for multi-agent path finding using temporary variable-fixing

Abstract

Journal

Organization