Research Analyzer
← Back ICRA 2026

DMRP-Bench: An Integrated, Unified Multi-Robot Motion Planning Benchmark in Dynamic Environments

zhijie hu, Xuebo Zhang, Runhua Wang, Yichen Li, Qingchen Bi

PDF

AI summary

Key figure (auto-extracted from paper)
DMRP-Bench provides a standardized, high-fidelity testbed that reveals critical efficiency-safety trade-offs and quantifies how local planning strategies directly impact multi-robot coordination in dynamic environments.
Multi-robot motion planning benchmarking dynamic environments NVIDIA Isaac Sim global-local planning robot coordination

Problem

Existing multi-robot motion planning benchmarks lack standardized interfaces for integrating global and local planners, rely on coarse system-level metrics, and fail to capture fine-grained inter-robot interactions in realistic dynamic simulations.

Approach

The authors developed a modular ROS and NVIDIA Isaac Sim framework that integrates global pathfinding algorithms with local controllers via a standardized interface and full-path subscription, enabling systematic evaluation across high-fidelity indoor scenarios with dynamic pedestrians.

Key results

  • Modular benchmark architecture with standardized global-to-local planner integration
  • Novel fine-grained metrics quantifying coordination risks, congestion, and flow stability
  • High-fidelity simulation scenarios featuring dynamic pedestrians in NVIDIA Isaac Sim
  • Comprehensive evaluation of 16 planner combinations revealing efficiency-safety trade-offs

Why it matters

Provides researchers and developers with a reproducible, standardized testbed to systematically compare and optimize multi-robot planning strategies in complex, dynamic environments.

Abstract

Multi-robot motion planning in dynamic environ- ments poses challenges to safe, efficient coordination, yet a fair, unified testbed for evaluating diverse algorithms remains absent. We present DMRP-Bench, a comprehensive framework to address this gap. It features a layered architecture integrating global and local planners, enabling a comprehensive analysis of their combinations across both macroscopic system-level outcomes and fine-grained inter-robot interactions. High-fidelity indoor scenarios (e.g., library, mall, office) simulating varied spatial layouts and pedestrian dynamics are built within the NVIDIA Isaac Sim environment. Extensive experiments on sixteen planner combinations reveal not only critical trade-offs between trajectory efficiency and safety, but also facilitate a deeper assessment of inter-robot coordination. By correlating path execution fidelity with interaction outcomes, these exper- iments enable quantitative diagnosis of how local behaviors influence interactive dynamics and holistic system performance.

Index terms

Multi-Robot Systems Performance Evaluation and Benchmarking Motion and Path Planning

Related papers