Rishi Veerapaneni

Multi-agent path finding (MAPF) is the problem of finding collision-free paths for a team of agents to reach their goal locations. State-of-the-art classical MAPF solvers typically employ heuristic search to find solutions for hundreds of agents but are typically centralized and can struggle to scale when run with short timeouts. Machine learning (ML) approaches that learn policies for each agent are appealing as these could enable decentralized systems and scale well while maintaining good solution quality. Current ML approaches to MAPF have proposed methods that have started to scratch the surface of this potential. However, state-of-the-art ML approaches produce “local" policies that only plan for a single timestep and have poor success rates and scalability. Our main idea is that we can improve a ML local policy by using heuristic search methods on the output probability distribution to resolve deadlocks and enable full horizon planning. We show several model-agnostic ways to use heuristic search with learnt policies that significantly improve the policies’ success rates and scalability. To our best knowledge, we demonstrate the first time ML-based MAPF approaches have scaled to high congestion scenarios (e.g. 20% agent density).

Graph search planning algorithms for navigation typically rely heavily on heuristics to efficiently plan paths. As a result, while such approaches require no training phase and can directly plan long horizon paths, they often require careful hand designing of informative heuristic functions. Recent works have started bypassing hand designed heuristics by using machine learning to learn heuristic functions that guide the search algorithm. While these methods can learn complex heuristic functions from raw input, they i) require significant training and ii) do not generalize well to new maps and longer horizon paths. Our contribution is showing that instead of learning a global heuristic estimate, we can define and learn local heuristics which results in a significantly smaller learning problem and improves generalization. We show that using such local heuristics can reduce node expansions by 2-20x while maintaining bounded suboptimality, are easy to train, and generalize to new maps & long horizon plans.