SkyRover

SkyRover, a modular and extensible simulator tailored for cross-domain pathfinding research.

Official Website: https://sites.google.com/view/mapf3d/home

Changelog

v1.1.0 (2025-03-04)

• Add non-ROS mapf3d demo, base on python panda3d.
• Modularize interfaces.
• Add script to convert MovingAI *.map to *.pcd file.

v1.0.1 (2025-02-25)

• Bug fixes.

v1.0.0 (2025-02-11)

• Initial release of SkyRover.

Environment Setup

To get started with SkyRover, follow the instructions below to set up your environment.

This project is developed on Ubuntu24.04LTS.

Get the code

mkdir -p ~/ros2_ws/src
cd ~/ros2_ws/src
git clone https://github.com/MA-Wenhui/skyrover.git

Install dependencies

conda create -n skyrover python=3.12
conda actiavte skyrover
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
pip install ray tensorboard matplotlib open3d panda3d

Install ROS2

To install ROS2, please follow the official installation guide:

• [ROS2 Installation Guide](https://docs.ros.org/en/jazzy/Installation.html)

Install Gazebo (Harmonic)

Please follow the official installation guide:

• [Gazebo Harmonic Installation](https://gazebosim.org/docs/harmonic/install_ubuntu/)

Install PX4 (Optional)

PX4 is an optional component for SkyRover, primarily for aerial simulations. If you want to integrate PX4 with SkyRover, follow the instructions below:

1. Download PX4 from [PX4 official site](https://px4.io/).
2. Follow the installation steps specific to your system.

Install Navigation2 (Optional)

Follow the installation steps to install Navigation2 Stack:

• [Navigation2 Doc](https://docs.nav2.org/development_guides/build_docs/index.html#install)

Train 3D DCC model (or just use pretrained data)

cd ~/ros2_ws/src/skyrover/skyrover/wrapper
python -m dcc_3d.train_dcc_3d

Run MAPF 3D

For ROS2 users:

# build the skyrover package
cd ~/ros2_ws/
colcon build --symlink-install
source ~/ros2_ws/install/setup.bash

# open the gz sim, it may takes a while for the first time
GZ_SIM_RESOURCE_PATH=~/ros2_ws/src/skyrover/skyrover/world/models/:$GZ_SIM_RESOURCE_PATH gz sim ~/ros2_ws/src/skyrover/skyrover/world/warehouse.sdf

# choose an alg to run
ros2 run skyrover run_mapf3d --alg 3dcbs --pcd ~/ros2_ws/src/skyrover/skyrover/world/map/map.pcd --pub_gz True --min_bound -21 -39 0 --max_bound 21 23 15
ros2 run skyrover run_mapf3d --alg 3dastar --pcd ~/ros2_ws/src/skyrover/skyrover/world/map/map.pcd --pub_gz True --min_bound -21 -39 0 --max_bound 21 23 15
ros2 run skyrover run_mapf3d --alg 3ddcc --pcd ~/ros2_ws/src/skyrover/skyrover/world/map/map.pcd --model ~/ros2_ws/src/skyrover/skyrover/wrapper/dcc_3d/data/65000.pth --pub_gz True --min_bound -21 -39 0 --max_bound 21 23 15

# run rviz to visualize /mapf_3d_pc topic
rviz2 ~/ros2_ws/src/skyrover/default.rviz

For non-ROS users:

conda activate skyrover
cd skyrover/skyrover/data
python map2pcd.py --map warehouse-20-40-10-2-1.map --out map_pointcloud_fine.pcd
cd ../..
python -m skyrover.run_mapf3d_no_ros --alg 3ddcc --pcd skyrover/data/map_pointcloud_fine.pcd --model skyrover/wrapper/dcc_3d/data/65000.pth --min_bound 0 0 0 --max_bound 123 321 9

Generate PointCloud from Gazebo (Optional)

One can use following gazebo plugin to generate pcd directly:

• [gazebo_map_creator](https://github.com/arshadlab/gazebo_map_creator)

Cite Our Work

@misc{ma2025skyrovermodularsimulatorcrossdomain,
      title={SkyRover: A Modular Simulator for Cross-Domain Pathfinding},
      author={Wenhui Ma and Wenhao Li and Bo Jin and Changhong Lu and Xiangfeng Wang},
      year={2025},
      eprint={2502.08969},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2502.08969},
}