Deploying Ten Thousand Robots: Scalable Imitation Learning for Lifelong Multi-Agent Path Finding

2024-10-28

arXiv.org

AI Summary

Addressing the ultra-large-scale lifelong multi-agent path finding (LMAPF) problem, this paper proposes the first imitation learning framework integrating distributed communication, single-step systematic collision resolution, and global guidance. The method models local observations using graph neural networks, augmented with a lightweight communication module and GPU-accelerated inference, enabling real-time, collision-free path planning for up to 10,000 agents under dynamic goal conditions. Compared to optimal learning and search-based baselines, our approach achieves average throughput improvements of 137.7% and 16.0%, respectively, across six large-scale maps—surpassing the 2023 International LMAPF Competition champion. Furthermore, end-to-end validation is demonstrated in a hybrid warehouse environment comprising 10 physical robots and 100 simulated agents.

Technology Category

• Multiagent Systems: Multiagent Planning
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

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.

Problem

• Scalable imitation learning for lifelong multi-agent path finding
• Matching performance of search-based algorithms in large-scale settings
• Improving throughput in large-scale multi-agent path planning

Innovation

• Novel communication module for multi-agent coordination
• Systematic single-step collision resolution technique
• Global guidance techniques for scalable pathfinding