Continuous Error Map for Autonomous Vehicles in Urban Areas

This repo will contain the implementation for our ITSC 2023 paper: Adaptive Multi-Sensor Integrated Navigation System Aided by Continuous Error Map from RSU for Autonomous Vehicles in Urban Areas. It is part of the project V2X Cooperative Navigation.

Videos:

Checkout our demo at Video Link

Sensor Setup

We collect 10 Hz LiDAR point cloud, 20 Hz images, 100 Hz IMU and 100 Hz ground truth positioning that explicitly considers the real-world noise in urban scenarios under a maximum speed of 30 km/h. Furthermore, we simulated the GNSS measurements using the realist RUMS simulator, which considers the impact of signal reflection and diffraction based on the 3D building model, as the CARLA can only simulate GNSS solutions with Gaussian noise. You can also collect your data using the tools here

Sensor	Description
1x LiDAR	360° Horizontal FOV, +10°~-30° vertical FOV, 100 meters, Velodyne HDL 32 noise model, 5cm standard deviation on range measurement
1 x RGB Camera	FoV of 90°, 960x600, forward, default noise setting in CARLA
1 x IMU	100 Hz, 9-axis, Xsens MTi 10 noise model according to UrbanNav
1 x GNSS	10 Hz, Ublox M8T noise model with sign reflection from buildings

Data Details

The dataset is released as rosbag while the ground truth is released both in TUM format and the raw output from CARLA simulator using CarlaFLCAV

name	Description	size	link
urban-noon 1	simulation data to generate error maps at noon	32.0 GB	ROSBAG, GT
urban-noon 2	test data for error map-aided sensor fusion at noon	32.0 GB	ROSBAG, GT
urban-sunset 1	simulation data to generate error maps at sunset	32.0 GB	ROSBAG, GT
urban-sunset 2	test data for error map-aided sensor fusion at sunset	32.0 GB	ROSBAG, GT
urban-night 1	simulation data to generate error maps at sunset	32.6 GB	ROSBAG, GT
urban-night 2	test data for error map-aided sensor fusion at sunset	32.0 GB	ROSBAG, GT

The topics within the rosbag are listed below:

topic	type	frequency	description
/carla/ego_vehicle/lidar	sensor_msgs/PointCloud2	10Hz	32 line LiDAR
/carla/ego_vehicle/image	sensor_msgs/Image	20Hz	FoV of 90° 960x600 image
/carla/ego_vehicle/gnss	nav_msgs/Odometry	10Hz	GNSS data
/carla/ego_vehicle/gnss_path	nav_msgs/Path	10Hz	GNSS path, for vis use
/carla/ego_vehicle/gt_path	nav_msgs/Path	10Hz	GT path, for vis use
/carla/ego_vehicle/imu	sensor_msgs/Imu	100Hz	IMU data

Run

Error map estimation

You can refer the config folder of the sensor parameter for running LOAM, LIOSAM, VINS and LVISAM with this data

Error-map-aided multi-sensor fusion

We are arranging the code, which will come soon!!

Acknowledge

We acknowledge the CarlaFLCAV for the useful tools to collect the data in CARLA. The authors also thank the valuable comments from ITSC 2023 reviewer, we will include the speed profile in the future.

Citation

If you use this work for your research, you may want to cite

@INPROCEEDINGS{rsalio2023huang,
  author={Huang, Feng and Wen, Weisong and Zhang, Guohao and Su, Dongzhe and Hsu, Li-Ta},
  booktitle={2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC)}, 
  title={Adaptive Multi-Sensor Integrated Navigation System Aided by Continuous Error Map from RSU for Autonomous Vehicles in Urban Areas
}, 
  year={2023},
  volume={},
  number={}
}