GRAND: Guidance, Rebalancing, and Assignment for Networked Dispatch in Multi-Agent Path Finding

2025-12-02

AI Summary

This paper addresses the lifelong multi-agent pickup and delivery (MAPD) scheduling problem for large-scale robot fleets (up to 500 robots) in high-density warehouses. We propose a learning-optimization hybrid framework: (1) a reinforcement learning–driven graph neural network models warehouse topology and generates idle-robot distribution policies; (2) region-level rebalancing is achieved via minimum-cost flow optimization; and (3) real-time task assignment is performed using a lightweight local allocation module. Our approach innovatively integrates graph-structured learning guidance with tractable combinatorial optimization, ensuring scalability and real-time performance within a 1-second computational budget. On congestion-prone benchmark instances, our method achieves a 10% throughput improvement over the 2024 championship scheduler, significantly alleviating traffic congestion and enhancing overall system efficiency.

Technology Category

• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Intelligent Robots: Learning & Optimization for ROB
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large robot fleets are now common in warehouses and other logistics settings, where small control gains translate into large operational impacts. In this article, we address task scheduling for lifelong Multi-Agent Pickup-and-Delivery (MAPD) and propose a hybrid method that couples learning-based global guidance with lightweight optimization. A graph neural network policy trained via reinforcement learning outputs a desired distribution of free agents over an aggregated warehouse graph. This signal is converted into region-to-region rebalancing through a minimum-cost flow, and finalized by small, local assignment problems, preserving accuracy while keeping per-step latency within a 1 s compute budget. On congested warehouse benchmarks from the League of Robot Runners (LRR) with up to 500 agents, our approach improves throughput by up to 10% over the 2024 winning scheduler while maintaining real-time execution. The results indicate that coupling graph-structured learned guidance with tractable solvers reduces congestion and yields a practical, scalable blueprint for high-throughput scheduling in large fleets.

Problem

• Develops a hybrid method for lifelong multi-agent pickup-and-delivery task scheduling
• Couples learning-based global guidance with lightweight optimization to reduce congestion
• Achieves higher throughput in large robot fleets while maintaining real-time execution

Innovation

• Graph neural network policy trained via reinforcement learning
• Minimum-cost flow for region-to-region rebalancing
• Small local assignment problems for real-time execution