Deploying Ten Thousand Robots: Scalable Imitation Learning for Lifelong Multi-Agent Path Finding
This work addresses the problem of efficient and high-quality path planning for large-scale multi-agent systems, such as in warehouse robotics, with incremental improvements over existing methods.
The paper tackles the challenge of matching the performance of search-based algorithms in lifelong multi-agent path finding (LMAPF) by proposing SILLM, a scalable imitation-learning-based solver that achieves average throughput improvements of 137.7% over learning-based and 16.0% over search-based baselines in large-scale settings with up to 10,000 agents.
Lifelong Multi-Agent Path Finding (LMAPF) repeatedly finds collision-free paths for multiple agents that are continually assigned new goals when they reach current ones. Recently, this field has embraced learning-based methods, which reactively generate single-step actions based on individual local observations. However, it is still challenging for them to match the performance of the best search-based algorithms, especially in large-scale settings. This work proposes an imitation-learning-based LMAPF solver that introduces a novel communication module as well as systematic single-step collision resolution and global guidance techniques. Our proposed solver, Scalable Imitation Learning for LMAPF (SILLM), inherits the fast reasoning speed of learning-based methods and the high solution quality of search-based methods with the help of modern GPUs. Across six large-scale maps with up to 10,000 agents and varying obstacle structures, SILLM surpasses the best learning- and search-based baselines, achieving average throughput improvements of 137.7% and 16.0%, respectively. Furthermore, SILLM also beats the winning solution of the 2023 League of Robot Runners, an international LMAPF competition. Finally, we validated SILLM with 10 real robots and 100 virtual robots in a mock warehouse environment.