Radar-LiDAR Dataset
NEWS
- 02/2020: Our extended paper (MulRan) is accepted at ICRA 20!
- 09/2019: The extended version of the dataset (link) is released!
This Dataset is submitted to ICRA2019 Full Day Workshop : Dataset Generation and Benchmarking of SLAM Algorithms for Robotics and VR/AR.
Abstract
Radar - LiDAR dataset provides radar data with the baseline trajectory of a vehicle for evaluation together with a 3D LiDAR point cloud as prior information.
Unlike existing radar datasets, the provided radar data support both raw-level and image format data, including 360º cumulated 1D intensity arrays with time stamps and 360º polar images.
Dataset information
1. Radar data
We provide two types of radar data.
One is 1D intensity arrays (raw data) at each angle. In this dataset, data are accumulated by 1 sweep (400 angle bins), and timestamps are added to reduce the communication load. Another is 360º polar images.
All of the radar data are provided to using robot operating system (ROS) bag files.
2. Baseline trajectory
In this dataset, partially measured VRS-GPS data, fiber optic gyro (FOG) data, and graph SLAM were used to estimate the baseline of the vehicle position.
If a loop occurs in the vehicle path, the relative poses are calculated by an iterated closest point (ICP) algorithm using a local point cloud and the poses are used as a constraint in the graph SLAM.
The baseline of the vehicle location in 6-DOF is provided to CSV format file.
3. 3D pointcloud
The 3D point cloud maps were constructed using the estimated baseline and LiDAR measurements. The maps are provided to LASer (LAS) format file.
Sequence 01
Sequence 02
Sequence 03
Sequence 04