Hierarchical Large Scale Multirobot Path (Re)Planning

TL;DR

The paper tackles real-time, large-scale multi-robot path planning in cluttered environments by introducing a hierarchical framework that partitions the workspace into convex cells and coordinates robots both between and within cells. A congestion-aware high-level planner based on Multi-Commodity Flow with Optimal Detour (MCF/OD) regulates inter-cell traffic under influx limits $\boldsymbol{\theta}$ with a suboptimality bound $w_{mcf} \ge 1$, while an anytime low-level MAPF operates inside each cell to compute collision-free local paths. A cell-crossing protocol enables continuous, non-stop execution across cell boundaries, and precomputation reduces replanning to high- and low-level updates. The approach demonstrates real-time performance with up to 142 simulated robots and 32 physical nano-quadrotors, highlighting significant speedups over baseline MAPF methods and applicability to UAV delivery, warehouse, and large-scale robotics scenarios.

Abstract

We consider a large-scale multi-robot path planning problem in a cluttered environment. Our approach achieves real-time replanning by dividing the workspace into cells and utilizing a hierarchical planner. Specifically, we propose novel multi-commodity flow-based high-level planners that route robots through cells with reduced congestion, along with an anytime low-level planner that computes collision-free paths for robots within each cell in parallel. A highlight of our method is a significant improvement in computation time. Specifically, we show empirical results of a 500-times speedup in computation time compared to the baseline multi-agent pathfinding approach on the environments we study. We account for the robot's embodiment and support non-stop execution with continuous replanning. We demonstrate the real-time performance of our algorithm with up to 142 robots in simulation, and a representative 32 physical Crazyflie nano-quadrotor experiment.

Figures (4)

• Figure 1: Long exposure of 32 quadrotors navigating a cluttered environment.
• Figure 2: The user inputs a map, start, and goal locations. Our approach generates a geometric partition, distributes robots among cells (hierarchical planner), and coordinates them within each cell in parallel.
• Figure 3: (a) inter-robot collision configuration $\mathcal{C}_{col}$ and buffered hyperplanes. (b)-(d) vertex-vertex, edge-edge, and edge-vertex generalized conflicts across cells.
• Figure 4: Geometric partitioning with $Q\!\!=\!\!5$, before spatial linear separation (a), and after (b). Each cell is a convex polytope in the workspace.