ntnu-arl/rig_dataset

Dataset Release

for "On the Characterization and Limits of 4D Radar for Aided Inertial Navigation"

This repository contains the dataset released alongside the publication of our paper "On the Characterization and Limits of 4D Radar for Aided Inertial Navigation". The paper and code release can be found:

• Paper: https://arxiv.org/abs/2605.01773
• Code: https://github.com/ntnu-arl/rig

The data is provided as ROS bags, with the topics described in the Sensors section below. The data was collected with aerial and handheld platforms, across a variety of environments, for almost 9 km of cumulative trajectories. The platforms used for the data collections can be seen below:

Aerial platform

Handheld platform

The following table summarizes the dataset content:

Sequence	Radar Type	Radar Chirp	Length [m]
Corridor
hel_3mps_align	IWR6843AOP	RC2	88
hel_4mps_align	IWR6843AOP	RC1	90
rec_3mps_align	IWR6843AOP	RC2	457
rec_3mps_const	IWR6843AOP	RC2	457
rec_4mps_align	IWR6843AOP	RC1	479
rec_4mps_const	IWR6843AOP	RC1	457
handheld_1	AWR1843BOOST	RC4	201
handheld_2	AWR1843BOOST	RC4	211
Gym
lem_3mps_align	IWR6843AOP	RC2	403
lem_3mps_const	IWR6843AOP	RC2	400
lem_4mps_align	IWR6843AOP	RC1	404
lem_4mps_const	IWR6843AOP	RC1	400
manual_1	IWR6843AOP	RC1	343
manual_2	IWR6843AOP	RC1	339
Mine
flight_4mps	IWR6843AOP	RC1	92
flight_5mps	IWR6843AOP	RC1	89
flight_6mps	IWR6843AOP	RC1	86
flight_7mps	IWR6843AOP	RC1	90
flight_8mps	IWR6843AOP	RC1	84
flight_9mps	IWR6843AOP	RC1	85
flight_10mps	IWR6843AOP	RC1	87
flight_11mps	IWR6843AOP	RC1	86
manual_3	IWR6843AOP	RC1	139
Forest
manual_4	IWR6843AOP	RC1	453
manual_5	IWR6843AOP	RC1	551
manual_6	IWR6843AOP	RC1	482
manual_7	IWR6843AOP	RC1	521
manual_8	IWR6843AOP	RC1	779
Basement
manual_9	uRAD Automotive	RC3	174
manual_10	uRAD Automotive	RC3	205
manual_11	uRAD Automotive	RC3	252

Sensors

The following table covers the different sensor models used in this dataset collection, as well as their respective topics.

Sensor	Model	Topic
IMU	VectorNav VN-100	/vectornav_driver_node/imu/data, /vectornav_driver_node/pressure
Radar 1	Texas Instruments IWR6843AOP	/radar/cloud
Radar 2	uRAD Automotive	/radar/cloud
Radar 3	Texas Instruments AWR1843BOOST	/radar/cloud
LiDAR	Ouster OS0-128 Rev7	/ouster/imu_packets, /ouster/lidar_packets, /ouster/metadata
Camera	FLIR Blackfly S 0.4 MP	/cam0/cam0/compressed, /cam0/camera_info

Only a single radar is used at a time, therefore they share the topic (/radar/cloud).

In some sequences of the Forest, visual data from a FLIR Blackfly camera is also available.

Extracting LiDAR Points

LiDAR data is stored as packets for smaller file sizes, we provide scripts to decompress these packets into the full LiDAR point clouds. This is done by

• Cloning the feature/rosbag_packet_unpacking branch of the ntnu-arl/ouster_ros repository.
• Build the package according to its README.
• Use rosrun ouster_ros bag_converter INPUT_FILE.bag OUTPUT_FILE.bag ouster to convert the INPUT_FILE.bag rosbag.
  • The resulting OUTPUT_FILE.BAG should now have the /ouster/points topic.

Calibration

Extrinsics, sourced either from CAD or via calibration, are provided for the aforementioned sensors in calibration/.

All extrinsics, except for the camera, are given as the pose of the sensor coordinate frame expressed in the IMU coordinate frame. For example, l_BR_B is the translation from the body (IMU) frame to the radar frame, expressed in the body frame. Likewise q_R_B is the rotation from the radar frame to the body frame.

Intrinsics and extrinsics for the camera can be found in camera.yaml, following kalibr.

Ground Truth

Ground truth is provided by LiDAR-inertial odometry, recorded to the /lio/pose topic of each sequence's rosbag.

Citation

If you use this data in your research, please cite the following publication:

@misc{nissov2026characterization,
      title={On the Characterization and Limits of 4D Radar for Aided Inertial Navigation}, 
      author={Morten Nissov and Kostas Alexis},
      year={2026},
      eprint={2605.01773},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
}