RIO

Optimization Based and Point Uncertainty Aware Radar-inertial Odometry for 4D Radar System

Prerequisites

Please refer to the docker/Dockerfile for the detailed dependencies.

Dcoker

Start the docker container

./docker/docker.sh -b # build the docker image
./docker/docker.sh -r # run the docker container

Outside the docker container

# Allow the docker container to connect to the X server
xhost +

Inside the docker container

# In the first terminal
roscore &
rviz # config file: rio/config/RIO.rviz

# In the second terminal
cd /ws
catkin_make

# Run the RIO with the sample dataset
python3 /ws/src/docker/run.py -a -n rio -c /ws/src/rio/config/ars548.yaml -d /ws/src/dataset/exp/Sequence_1.bag -r 1 -p 1

Then you can see the odometry and the point cloud in rviz.

System Overview

Experiment Platform

Our platform consists of a 4D FMCW Radar ARS548RDI manufactured by Continental and an IMU BMI088 manufactured by Bosch. The radar sensor is mounted on the front of the platform, while the IMU is mounted on the bottom.

Trajectories on self-collected and ColoRadar dataset

Red trajectory is the proposed full system. Blue one is the system without point uncertainty model, and black one is the ground truth trajectory. We present the results on four sequences in two different datasets.

Self-collected dataset

• Sequence 1 : involved relatively low-speed movements.

• Sequence 2 : involved relatively high-speed movements.

• Sequence 3 : involved relatively high-speed movements with high-speed rotations.

Data Format

Field Name	Data Type	Count	Offset (Bytes)	Remarks
azimuth	sensor_msgs::PointField::FLOAT32	1	0	Angle in the horizontal plane
azimuthSTD	sensor_msgs::PointField::FLOAT32	1	4	Standard deviation of azimuth
elevation	sensor_msgs::PointField::FLOAT32	1	8	Angle in the vertical plane
elevationSTD	sensor_msgs::PointField::FLOAT32	1	12	Standard deviation of elevation
range	sensor_msgs::PointField::FLOAT32	1	16	Distance to the target
rangeSTD	sensor_msgs::PointField::FLOAT32	1	20	Standard deviation of range
velocity	sensor_msgs::PointField::FLOAT32	1	24	Speed of the target
velocitySTD	sensor_msgs::PointField::FLOAT32	1	28	Standard deviation of velocity
rcs	sensor_msgs::PointField::INT8	1	32	Radar cross-section

ColoRadar dataset

It consists of 52 sequences, recorded in mines, built environments, and in an urban creek path, totaling more than 145 minutes of 3D FMCW radar, 3D lidar, and IMU data. The full dataset, including sensor data, calibration sequences, and evaluation scripts. It is available at ColoRadar.

Citation

If you find our work useful in your research, please consider citing:

@article{huang2024morephysicalenhancedradarinertialodometry,
      title={Less is More: Physical-enhanced Radar-Inertial Odometry},
      author={Qiucan Huang and Yuchen Liang and Zhijian Qiao and Shaojie Shen and Huan Yin},
      booktitle={ICRA},
      year={2024},
}

@misc{xu2024modelingpointuncertaintyradar,
      title={Modeling Point Uncertainty in Radar SLAM},
      author={Yang Xu and Qiucan Huang and Shaojie Shen and Huan Yin},
      year={2024},
      eprint={2402.16082},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2402.16082},
}

License

MIT License (see LICENSE).