Graph-based, sparse radar-inertial odometry m-estimation with barometer support and zero-velocity tracking.
Paper: https://arxiv.org/pdf/2408.05764
@inproceedings{girod2024brio,
author = {Rik Girod and Marco Hauswirth and Patrick Pfreundschuh and Mariano Biasio and Roland Siegwart},
title = {A robust baro-radar-inertial odometry m-estimator for multicopter navigation in cities and forests},
booktitle={IEEE Int. Conf. Multisensor Fusion Integration Intell. Syst.},
year={2024}
}
Install ROS noetic.
sudo apt install git build-essential python3-rosdep python3-catkin-tools ros-noetic-rqt-multiplot -y
mkdir catkin_ws
cd catkin_ws
mkdir src
catkin init
cd src
git clone https://github.com/ethz-asl/rio.git
git clone https://github.com/ethz-asl/lpp.git
git clone https://github.com/rikba/gtsam_catkin.git
cd ..
rosdep install --from-paths src --ignore-src -r -y
catkin build
Start RIO in one terminal
source ~/catkin_ws/devel/setup.bash
roslaunch rio rio.launch visualize:=true
Download an example dataset sequence.
Replay the bag
rosbag play 01_urban_night_H_raw.bag
| Sensor | Default topic | Message type | Required | Note |
|---|---|---|---|---|
| IMU | /imu/data_raw | sensor_msgs/Imu | Yes | Calibrate gyro turn-on-bias! |
| IMU filtered | /imu/data | sensor_msgs/Imu | Yes | for initialization |
| Radar | /radar/cfar_detections | sensor_msgs/PointCloud2 | Yes | |
| Barometer | /baro/pressure | sensor_msgs/FluidPressure | No | Activate in cfg |
Radar point cloud format, see also mav_sensors_ros.
| Field name | Size |
| ---------- | ------- |
| x | FLOAT32 |
| y | FLOAT32 |
| z | FLOAT32 |
| doppler | FLOAT32 |
| snr | INT16 |
| noise | INT16 |
The dataset contains 15 sequences of urban night, forest path, field, and deep forest handheld and flown data. It can be downloaded with and without camera images.
dataset_quadrotor_examples.mp4
| Scenario | No. | Link w/o camera |
|---|---|---|
| Urban Night | 01 | 01_urban_night_H_raw_no_img.bag |
| 02 | 02_urban_night_H_raw_no_img.bag | |
| 03 | 03_urban_night_H_raw_no_img.bag | |
| 04 | 04_urban_night_H_raw_no_img.bag | |
| 05 | 05_urban_night_F_raw_no_img.bag | |
| 06 | 06_urban_night_F_raw_no_img.bag | |
| 07 | 07_urban_night_F_raw_no_img.bag | |
| Forest Path | 08 | 08_forest_path_H_raw_no_img.bag |
| 09 | 09_forest_path_H_raw_no_img.bag | |
| 10 | 10_forest_path_H_raw_no_img.bag | |
| 11 | 11_forest_path_F_raw_no_img.bag | |
| 12 | 12_forest_path_F_raw_no_img.bag | |
| Flat Field | 13 | 13_flat_field_F_raw_no_img.bag |
| 14 | 14_flat_field_F_raw_no_img.bag | |
| Tree Slalom | 15 | 15_tree_slalom_F_raw_no_img.bag |
| All | 01-15 | all_raw_no_img |
| Field name | IMU to radar | radar to cam |
| ---------- | ------------ | ------------ |
| x | 0.122 | -0.020 |
| y | 0.000 | -0.015 |
| z | -0.025 | 0.000 |
| qx | 0.67620958 | 0.000 |
| qy | 0.67620958 | 0.7071068 |
| qz | -0.20673802 | 0.7071068 |
| qw | -0.20673802 | 0.000 |
<node name="imu_to_radar_broadcaster" pkg="tf2_ros" type="static_transform_publisher" args="0.122 0.000 -0.025 0.67620958 0.67620958 -0.20673802 -0.20673802 'bmi088' 'awr1843aop'" />
<node name="radar_to_cam_broadcaster" pkg="tf2_ros" type="static_transform_publisher" args="-0.020 -0.015 0.0 0.0 0.7071068 0.7071068 0.0 'awr1843aop' 'cam'" />
rio_calibration_node.cpp allows you to calibrate the rigid transformation between IMU and radar.
The calibration procedure:
- Move the device in a loop, recording RIO odometry output (
/rio/odometry_optimize), IMU (/imu/data_raw) and radar (/radar/cfar_detections). The device has to be located in the same position and orientation at start and goal to make the loop closure. - Set the initial guess in calibration.yaml.
roslaunch rio calibration.launch
This should refine your initial guess. Note you need to excite roll and pitch in your dataset. I noticed that you could also relax the gyro bias process noise, as it is well observable through loop closure.
| Package | Description | Link |
|---|---|---|
| mav_sensors | Linux user space sensor drivers | mav_sensors |
| mav_sensors_ros | ROS sensor interface | mav_sensors_ros |