The CU-Multi Dataset

The CU-MULTI Dataset: A dataset aimed to support multi-robot map-merging, inter-robot place recognition and loop-closure detection.

Download Dataset

If you would like to download the dataset, you can do so here. A Globus account is required, which you can create using either your institutional or personal email address.

Citation

If you find CU-Multi and/or the provided tools useful in your research, please cite our paper:

@misc{albin2025cumultidatasetmultirobotcollaborative,
      title={CU-Multi: A Dataset for Multi-Robot Collaborative Perception}, 
      author={Doncey Albin and Daniel McGann and Miles Mena and Annika Thomas and Harel Biggie and Xuefei Sun and Steve McGuire and Jonathan P. How and Christoffer Heckman},
      year={2025},
      eprint={2509.19463},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2509.19463}, 
}

Dataset File Structure

CU-Multi/
├── calib/
│   └── description/
│       ├── meshes/
│       │   ├── cu_hunter_body.dae
│       │   └── cu_hunter_wheel.dae
│       └── robot.urdf
├── <environment>/
│   ├── <env>.osm.xml
│   └── <robot>/
│       ├── <robot>_gt_utm_poses.csv
│       ├── <robot>-<env>_cam_rgb/...
│       ├── <robot>-<env>_cam_depth/...
│       ├── <robot>-<env>_imu_gps/...
│       ├── <robot>-<env>_poses/...
│       └── <robot>-<env>_lidar/
│           ├── metadata.yaml
│           └── <robot>-<env>_lidar_0.db3

ROS2 topics, message types, and associated sensors

Topic Name	Message Type	Sensor	Publish Rate	Description
<robot>_<env>_ground_truth.db3
<robot>/ground_truth/path	nav_msgs/msg/Path	-	-	.
<robot>/ground_truth/odometry	nav_msgs/msg/Odometry	-	20 Hz	.
/tf	tf2_msgs/msg/TFMessage	-	20 Hz	Publishes tf from 'world' frame to robot lidar frame 'base_link' at each lidar timestamp.
<robot>_<env>_lidar.db3
<robot>/ouster/metadata	std_msgs/msg/String	LiDAR	-	.
<robot>/ouster/points	sensor_msgs/msg/PointCloud2	LiDAR	20 Hz	.
<robot>/ouster/imu	sensor_msgs/msg/Imu	LiDAR	100 Hz	.
<robot>_<env>_imu_gps.db3
<robot>/imu/mag	sensor_msgs/msg/MagneticField	IMU	500 Hz	.
<robot>/imu/data	sensor_msgs/msg/Imu	IMU	500 Hz	Raw IMU measurements.
<robot>/imu/pressure	sensor_msgs/msg/FluidPressure	IMU	500 Hz	Air pressure measurements.
<robot>/gnss_1/llh_position	sensor_msgs/msg/NavSatFix	GPS	2 Hz	Rear left RTK GPS measurements.
<robot>/gnss_2/llh_position	sensor_msgs/msg/NavSatFix	GPS	2 Hz	Rear right RTK GPS measurements.
<robot>_<env>_camera_rgb.db3
<robot>/camera/color/camera_info	sensor_msgs/msg/CameraInfo RGB-D	10 Hz	Camera intrinsic/distortion parameters.
<robot>/camera/color/image_raw	sensor_msgs/msg/Image	RGB-D	10 Hz	Images captured by front camera.
<robot>/camera/color/metadata	realsense2_camera/msg/Metadata	RGB-D	10 Hz	._
<robot>_<env>_camera_depth.db3
<robot>/camera/depth/camera_info	sensor_msgs/msg/CameraInfo	RGB-D	10 Hz	Camera intrinsic/distortion parameters.
<robot>/camera/depth/image_raw	sensor_msgs/msg/Image	RGB-D	10 Hz	Images captured by front camera.
<robot>/camera/depth/metadata	realsense2_camera/msg/Metadata	RGB-D	10 Hz	._

We’ll keep this section updated as the dataset evolves. Feel free to open an issue if anything seems unclear or incomplete.

Converting to ROS1/ROS2 bag formats

Please Note: The order of this is likely to change with a cleaner layout, but this will quickly get you started.

To convert the dataset to ROS1 bag files, you can use the provided Docker containers in the dataset_tools directory following the steps below. Before proceeding further, please make sure you have Docker installed. You can install Docker following the instructions at this link.

Update the SCRIPTS_DIR and DATASETS_DIR variables run_and_enter_container.sh bash script with the path of the conversion_scripts directory (there's one in the ros1 and ros2 folders) you will be using and where you have saved the CU-Multi dataset, respectively.

Convert to ROS1 rosbag format

Install and enter the Docker container

1. cd dataset_tools

2. cd ros1

3. sudo bash docker/build_docker.sh

4. sudo bash run_and_enter_container.bash

5. source /opt/ros/noetic/setup.bash

While you are in the container, run ros_core. If you want to do this in a tmux session, even better. Open a new window and repeat entering the container like above if not using a tmux session. Now enter /root/conversion_scripts directory. In this directory is the Python file for converting the unstructured CU-Multi dataset into a ROS1 rosbag file.

First, edit the Python file with the environment and corresponding robot you would like to convert.

Now run the Python file with:

python3 make_ros1_bag.py.

Convert to ROS2 rosbag format

Install and enter the Docker container

1. cd dataset_tools

2. cd ros2

3. sudo bash docker/build_docker.sh

4. sudo bash run_and_enter_container.bash

Enter the /root/conversion_scripts directory. In this directory is the Python file for converting the unstructured CU-Multi dataset into a ROS2 db3 file.

First, edit the Python file with the environment and corresponding robot you would like to convert.

Now run the Python file with:

python3 make_ros2_bag.py.