Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bridging Planning and Execution: Multi-Agent Path Finding Under Real-World Deadlines

Jingtian Yan, Shuai Zhou, Stephen F. Smith, Jiaoyang Li

TL;DR

This work addresses the gap between planning and execution in multi-agent path finding under real-world deadlines by introducing REMAP, a framework that augments planners with execution-time awareness via ExecTimeNet. ExecTimeNet, a Transformer-GNN model, predicts per-agent execution times from ADG-encoded MAPF plans and supports time-point or distribution outputs to compute deadline penalties. REMAP integrates with CBS and MAPF-LNS, updating penalties during search and guiding refinement toward deadline-feasible solutions, achieving up to 20% improvement over baselines on large maps. The approach generalizes across unseen maps and scales to hundreds of agents, offering a practical path toward reliable, time-sensitive multi-agent systems beyond MAPF-RD.

Abstract

The Multi-Agent Path Finding (MAPF) problem aims to find collision-free paths for multiple agents while optimizing objectives such as the sum of costs or makespan. MAPF has wide applications in domains like automated warehouses, manufacturing systems, and airport logistics. However, most MAPF formulations assume a simplified robot model for planning, which overlooks execution-time factors such as kinodynamic constraints, communication latency, and controller variability. This gap between planning and execution is problematic for time-sensitive applications. To bridge this gap, we propose REMAP, an execution-informed MAPF planning framework that can be combined with leading search-based MAPF planners with minor changes. Our framework integrates the proposed ExecTimeNet to accurately estimate execution time based on planned paths. We demonstrate our method for solving MAPF with Real-world Deadlines (MAPF-RD) problem, where agents must reach their goals before a predefined wall-clock time. We integrate our framework with two popular MAPF methods, MAPF-LNS and CBS. Experiments show that REMAP achieves up to 20% improvement in solution quality over baseline methods (e.g., constant execution speed estimators) on benchmark maps with up to 300 agents.

Bridging Planning and Execution: Multi-Agent Path Finding Under Real-World Deadlines

TL;DR

This work addresses the gap between planning and execution in multi-agent path finding under real-world deadlines by introducing REMAP, a framework that augments planners with execution-time awareness via ExecTimeNet. ExecTimeNet, a Transformer-GNN model, predicts per-agent execution times from ADG-encoded MAPF plans and supports time-point or distribution outputs to compute deadline penalties. REMAP integrates with CBS and MAPF-LNS, updating penalties during search and guiding refinement toward deadline-feasible solutions, achieving up to 20% improvement over baselines on large maps. The approach generalizes across unseen maps and scales to hundreds of agents, offering a practical path toward reliable, time-sensitive multi-agent systems beyond MAPF-RD.

Abstract

The Multi-Agent Path Finding (MAPF) problem aims to find collision-free paths for multiple agents while optimizing objectives such as the sum of costs or makespan. MAPF has wide applications in domains like automated warehouses, manufacturing systems, and airport logistics. However, most MAPF formulations assume a simplified robot model for planning, which overlooks execution-time factors such as kinodynamic constraints, communication latency, and controller variability. This gap between planning and execution is problematic for time-sensitive applications. To bridge this gap, we propose REMAP, an execution-informed MAPF planning framework that can be combined with leading search-based MAPF planners with minor changes. Our framework integrates the proposed ExecTimeNet to accurately estimate execution time based on planned paths. We demonstrate our method for solving MAPF with Real-world Deadlines (MAPF-RD) problem, where agents must reach their goals before a predefined wall-clock time. We integrate our framework with two popular MAPF methods, MAPF-LNS and CBS. Experiments show that REMAP achieves up to 20% improvement in solution quality over baseline methods (e.g., constant execution speed estimators) on benchmark maps with up to 300 agents.

Paper Structure

This paper contains 36 sections, 9 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Problem Formulation
  4. Related Works
  5. Multi-Agent Path Finding
  6. MAPF with Deadlines
  7. Bridging Planning and Execution
  8. Travel and Action Time Prediction
  9. Method
  10. REMAP
  11. Penalty Update
  12. MAPF-LNS
  13. CBS
  14. Execution Time Prediction (ExecTimeNet)
  15. ADG encoding
  16. ...and 21 more sections

Figures (9)

  • Figure 1: Overview of REMAP.
  • Figure 2: Model structure of ExecTimeNet. Pink arrows indicate the direction of information propagation.
  • Figure 3: Comparison of the linear and percentage penalty on MAPF-LNS. The terms "Linear" and "Percentage" in parentheses after "Penalty Gap" indicate the penalty types used in the planner.
  • Figure 4: Comparison of the linear penalty on MAPF-LNS with different iterations of refinement.
  • Figure 5: Comparison of the linear penalty on CBS.
  • ...and 4 more figures

Theorems & Definitions (2)

  • Definition 1: Multi-Agent Path Finding
  • Definition 2: MAPF with Real-world Deadlines