Embedded Files
The Seonsor Configuration
Two Spinning LiDAR (OS2-128 & VLP-16)
Two Solid state LiDAR (Livox Avia & Aeva Aeries II)
SPAN-CPT7 with VEXXIS GNSS-501 Dual Antenna
Xsens MTi-300 IMU
This setting does not not represent the peak performance, but rather for optimizing data acquisition.
Aeva offers a range of configurations for data acquisition, with one such configuration capable of detecting objects at a distance of up to 500 meters.
The sensor coordinate information
The orientation (axis) and distance between the sensors have been ascertained using a CAD model. This information is subsequently utilized as the initial estimation in the extrinsic calibration process.
Inter-Sensor Calibration in the HeLiPR Dataset
In the HeLiPR dataset, we offer three distinct calibrations to enhance user convenience:
Multiple LiDAR Calibration: We employed the mlcc repository for calibrating multiple LiDARs. The OS2-128 served as the primary LiDAR during this process. A circular trajectory (intended for loop closure) was maneuvered, ensuring the minimum distance for trajectory precision throughout the batch optimization. The trajectory estimation was derived using DLO. The bag file dedicated to multiple LiDAR calibration is accessible in the Download section.
LiDAR-IMU Extrinsic Calibration: This calibration was executed using FAST-LIO2. Due to the vehicle's constraints in providing 6-DoF motion and the relative inaccuracy of the z-axis in SLAM when compared to other axes, we utilized a CAD model for the z-axis. We chose the Ouster and IMU for calibration because of Ouster's extensive FOV and high point number. Given that both Ouster and IMU are collinearly installed and share an axis, the CAD model's accuracy is comparatively high.
INS-IMU Extrinsic Calibration: This was conducted with MA-LIO. By leveraging the extrinsic calibration between multiple LiDARs and IMU, the IMU trajectory is produced as an output from MA-LIO. When this data is amalgamated with the positional information sourced from the INS, we achieved the extrinsic calibration between IMU and INS using a hand-eye calibration technique.
All the extrinsic calibrations are presented in IMU coordinates.
Data Format
Multiple LiDARs
Binary file (structures are shown on the right)
Individual LiDAR Groundtruth
TXT format with [time, x, y, z, qx, qy, qz, qw]
Calibration parameter
TXT format with rotation matrix and translation vector
Inertial data
CSV format for raw measurement from IMU, INS
(2023 /09) In LiDAR_GT, the global_sensor_gt.txt is also updated to find correspondence between MulRan)
LiDAR Data
Four LiDARs were asynchronously acquired at a frequency of 10Hz and saved as a bin file in the format depicted in the above image. Given that each LiDAR has unique channels, we have documented the parsing order for each LiDAR. The filename of each binary file corresponds to the acquisition timestamp, and these timestamps are also archived in stamp.csv. The file player releases the topics grounded on this timestamp file.
IMU Data
The data structure for the IMU is organized in the following order: [timestamp, qx, qy, qz, qw, eul x, eul y, eul z, gyr x, gyr y, gyr z, acc x, acc y, acc z, mag x, mag y, mag z]. This information is stored in the 'inertial_data/xsens_imu.csv' file. Additionally, it is broadcasted as a ROS topic under the name '/imu/data_raw', which publishes data in the format [timestamp, quaternion, angular_velocity, linear_acceleration] at a frequency of 100Hz.
INS Data
The INS data is stored in the 'inertial_data/inspva.csv' file, organized in the following sequence: [timestamp, latitude, longitude, height, north velocity, east velocity, up velocity, roll, pitch, azimuth, status]. This data serves as the foundation for creating the ground truth and was acquired at a frequency of 50Hz.
Individual LiDAR Ground Truth in the HeLiPR Dataset
A notable contribution of the HeLiPR dataset is its provision of individual LiDAR ground truth. Given that every LiDAR sensor has distinct acquisition timestamp and installation points, the scanning positions differ due to both spatial and temporal variations.
Since place recognition evaluations are predominantly decided by range, the precise positioning of these data points holds paramount importance.
Recognizing this, the HeLiPR Dataset offers individual LiDAR groundtruth. These ground truths are formulated based on the extrinsic calibration of each LiDAR and the B-Spline Interpolation, which is tailored to each LiDAR's unique acquisition timestamp.
Illustrative Example: Ground Truth for Roundabout01
When employing the individual ground truth and adopting the Velodyne as a reference point, valid place recognition candidates (or true positives) are observable for all LiDARs within a 3m radius of alternate trajectories (as marked by the black box).
However, when focusing on the exact location and timing of the INS (with a permissible discrepancy of 5ms), only Aeva displays a true positive.
This problem leads to complications in validating place recognition candidates if relying solely on the INS data. Such discrepancies arise primarily due to the spatial distance and temporal delays encountered during Inter-LiDAR Place recognition.
Page updated
Google Sites
Report abuse