Decentralized multi-agent control using GNNs for ICRA 2022

Repository containing the code base for evaluating the multi-agent coordination policy used in the paper "A Framework for Real-World Multi-Robot Systems Running Decentralized GNN-Based Policies".

Supplementary video material:

Citation

If you use any part of this code in your research, please cite our paper:

@inproceedings{blumenkamp2022decentralizedgnn,
  title={A Framework for Real-World Multi-Robot Systems Running Decentralized GNN-Based Policies},
  author={Blumenkamp, Jan and Morad, Steven and Gielis, Jennifer and Li, Qingbiao and Prorok, Amanda},
  booktitle={IEEE International Conference on Robotics and Automation},
  year={2022},
  organization={IEEE}
}

Setup

1. Install ROS2
2. Clone with git clone --recursive to include submodules
3. Source the system ROS2 installation (typically . /opt/ros2/galactic/setup.sh)
4. Build from the root of this repository using colcon build --symlink-install. This command has to be executed twice (otherwise the RVO library is not built and included properly), this is a known issue.

Training the policy

We provide pre-trained models, but if re-training the GNN is desired, this can be done as described here. The path to the model has to be specified in the launch file (model_path parameter in the passage_gnn_simple executables).

Evaluation in simulation

1. After the build was successful, source the workspace from the root in two separate terminal windows, using . install/setup.bash
2. Start the simulation in the first window by running ros2 launch launch/simulation_robomasters.launch.py
3. Start the policy in the second window. There are three options:
   1. RVO only: Uses centralized RVO to navigate robots to their goals, can be used as a baseline. Run ros2 launch launch/centralized_rvo_passage_robomaster.launch.yaml
   2. Centralized GNN: Uses centralized RVO for resetting robots to initial positions and centralized GNN policy evaluation to navigate robots to their goal positions. Run ros2 launch src/passage_gnn_simple/launch/centralized_passage_robomaster.launch.yaml.
   3. Decentralized GNN: Uses centralized RVO for resetting robots to initial positions and decentralized GNN policy evaluation to navigate robots to their goal positions. Run ros2 launch src/passage_gnn_simple/launch/decentralized_passage_robomaster.launch.py.

Evaluation on real robots

This will be updated soon.