DMRP-Bench: A Benchmark for Multi-Robot Motion Planning in Dynamic Environments

DMRP-Bench is a unified benchmarking framework for evaluating multi-robot motion planning in high-fidelity dynamic environments, built on NVIDIA Isaac Sim and ROS. It supports various global and local planners, and introduces fine-grained interaction metrics (STO, IBTR, LDF, PFD).

Quick Start

1. Prerequisites

• Ubuntu 20.04/22.04
• ROS Noetic (installation)
• NVIDIA Isaac Sim 2022.2.1+ (download)
• Python 3.8+, tmux

2. Installation

# Clone the repository
git clone https://github.com/NKU-MobFly-Robotics/DMRP-Bench.git
cd DMRP-Bench

# Install ROS navigation packages
sudo apt install ros-noetic-navigation ros-noetic-teb-local-planner

# Install Python dependencies
pip install -r requirements.txt

# Install system dependencies via rosdep
rosdep install --from-paths src --ignore-src -r -y

# Build the workspace
cd ~/DMRP-Bench
catkin_make -DCMAKE_BUILD_TYPE=Release
source devel/setup.bash

3. Run a Benchmark

• Step 1: Launch the simulation Start Isaac Sim:

/path/to/isaac_sim/isaac-sim.sh

In Isaac Sim, open File → Open and select one of the provided scene files in the scenes/ directory, e.g., Library.usd, OpenOffice.usd, or Shopping.usd.

• Step 2: Start the ROS socket server

cd scripts
python3 pose_socket_server.py

• Step 3: Send pedestrian poses from Isaac Sim In Isaac Sim, open Script Editor (Window → Script Editor) and run the script from scripts/isaac_send_pedestrian_poses.py (copy the code from that file).

The system will automatically start planning and recording data. Bag files are saved in the bags/ directory inside the repository.

4. Analyze Results

python3 scripts/analyze_multi_robot_bag.py --bag ./bags/your-bag-file.bag

The analysis report will be saved as your-bag-file_analysis.txt.

Configuration

Local planners: To adjust parameters for TEB, DWA, MPC, or Ceres, modify the corresponding YAML files in move_base_benchmark/params/. To switch between different local planners, edit the launch file move_base_benchmark/launch/aa.launch.

Global planners: To change the global planner (CBS, ECBS-8, SIPP, EECBS), modify the launch file mapf_base/launch/mapf_example.launch and specify the desired planner type.

Pedestrian models: The actors in the scene already have collision properties; no extra setup needed.

Citation

If you use this benchmark in your research, please cite:

@inproceedings{hu2026dmrp,
  title={DMRP-Bench: An Integrated, Unified Multi-Robot Motion Planning Benchmark in Dynamic Environments},
  author={Hu, Zhijie and Zhang, Xuebo and Wang, Runhua and Li, Yichen and Bi, Qingchen},
  booktitle={2026 IEEE International Conference on Robotics and Automation (ICRA)},
  year={2026}
}