Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Agent Path Finding via Finite-Horizon Hierarchical Factorization

Jiarui Li, Alessandro Zanardi, Gioele Zardini

TL;DR

This paper tackles scalable multi-agent path finding (MAPF) in dynamic, warehouse-like environments. It introduces a finite-horizon hierarchical factorization framework that plans one step at a time in a receding-horizon manner, enabling online execution. Each cycle computes parallel single-agent paths, detects $H$-step conflicts with spatial hashing to form a two-level grouping, and uses a PIBT-based replanning for non-finalized robots, with congestion-resolution expanding groups when needed. The approach achieves up to a $60\%$ reduction in time-to-first-action and maintains competitive SOC compared to offline baselines such as LaCAM* across different problem sizes and horizons, demonstrating strong practical speedups for real-time large-scale robot coordination.

Abstract

We present a novel algorithm for large-scale Multi-Agent Path Finding (MAPF) that enables fast, scalable planning in dynamic environments such as automated warehouses. Our approach introduces finite-horizon hierarchical factorization, a framework that plans one step at a time in a receding-horizon fashion. Robots first compute individual plans in parallel, and then dynamically group based on spatio-temporal conflicts and reachability. The framework accounts for conflict resolution, and for immediate execution and concurrent planning, significantly reducing response time compared to offline algorithms. Experimental results on benchmark maps demonstrate that our method achieves up to 60% reduction in time-to-first-action while consistently delivering high-quality solutions, outperforming state-of-the-art offline baselines across a range of problem sizes and planning horizons.

Multi-Agent Path Finding via Finite-Horizon Hierarchical Factorization

TL;DR

This paper tackles scalable multi-agent path finding (MAPF) in dynamic, warehouse-like environments. It introduces a finite-horizon hierarchical factorization framework that plans one step at a time in a receding-horizon manner, enabling online execution. Each cycle computes parallel single-agent paths, detects -step conflicts with spatial hashing to form a two-level grouping, and uses a PIBT-based replanning for non-finalized robots, with congestion-resolution expanding groups when needed. The approach achieves up to a reduction in time-to-first-action and maintains competitive SOC compared to offline baselines such as LaCAM* across different problem sizes and horizons, demonstrating strong practical speedups for real-time large-scale robot coordination.

Abstract

We present a novel algorithm for large-scale Multi-Agent Path Finding (MAPF) that enables fast, scalable planning in dynamic environments such as automated warehouses. Our approach introduces finite-horizon hierarchical factorization, a framework that plans one step at a time in a receding-horizon fashion. Robots first compute individual plans in parallel, and then dynamically group based on spatio-temporal conflicts and reachability. The framework accounts for conflict resolution, and for immediate execution and concurrent planning, significantly reducing response time compared to offline algorithms. Experimental results on benchmark maps demonstrate that our method achieves up to 60% reduction in time-to-first-action while consistently delivering high-quality solutions, outperforming state-of-the-art offline baselines across a range of problem sizes and planning horizons.
Paper Structure (4 sections, 3 figures)

This paper contains 4 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Finite-Horizon Hierarchical Factorization
  3. Experimental Results
  4. Discussion and Conclusion

Figures (3)

  • Figure 1: The algorithm plans in a receding-horizon fashion, computing and finalizing robot movements one timestep at a time. This online approach enables immediate execution of each computed step.
  • Figure 2: The ratio of the TNBE between our algorithm and offline baselines; values below 1 indicate that our method achieves faster execution readiness.
  • Figure 3: Our algorithm consistently outperforms LaCAM* in solution quality across all robot counts. In the warehouse map, performance improves monotonically as the planning horizon increases, while in the random map, the optimal planning horizon varies depending on the number of robots.