Learning to Select Nodes in Bounded Suboptimal Conflict-Based Search for Multi-Agent Path Finding

Abstract

Multi-Agent Path Finding is an NP-hard problem that is difficult for current approaches to solve optimally. Research has shown that bounded suboptimal solvers, such as Enhanced Conflict-Based Search (ECBS), are more efficient than optimal solvers in finding a feasible solution with suboptimality guarantees. ECBS is a tree search algorithm that repeatedly selects nodes from a focal list to expand the tree. In this work, we propose to use imitation learning and curriculum learning to learn node-selection strategies for different grid maps and agent sizes. We then deploy the learned models in ECBS and test their solving performance on unseen instances drawn from the same distribution as the one used in training. Our approach shows substantial improvement over the baselines on different grid maps.

Tasks

Reproductions