Real-world point clouds from the AD Assurance Lab's roof-mounted Ouster OS1-64 at Western Michigan University, recorded in and around Kalamazoo, Michigan from 2020 to 2025. The 2020 and 2021 drives also carry the car's Aptiv ESR 2.5 long-range radar tracks, so vehicles can be located independently of the LiDAR, out to about 175 m. That makes these drives useful for studying detection at long range, where a LiDAR has only a few points on a car.

None of the scans has 3D labels.

Collection	When	Recordings	Duration	LiDAR scans	Size	Also included
`highway_2020`	2020-03-05	1 highway drive	2.2 min	1,342	0.8 GB	Aptiv ESR radar tracks, Ouster IMU
`niro_2021`	Jan-Mar 2021	11 drives	46.5 min	17,505	14.0 GB	Aptiv ESR radar tracks (10 drives), GPS (6 drives), Ouster IMU
`urban_2023-12`	2023-12-04	1 urban drive	26.9 min	16,125	13.3 GB	Ouster IMU
`radar_tests_2023-24`	2023-12-11 to 2024-02-15	7 short test recordings	3.6 min	2,127	1.7 GB	second-radar target lists (2 recordings), Ouster IMU and metadata
`calibration_2023-24`	2023-12-21, 2024-05-10	2 camera-lidar calibration recordings (chessboard target)	3.8 min	2,293	1.9 GB	Ouster IMU (2024-05-10)
`shuttle_2025-01`	2025-01-29	1 campus shuttle drive	3.0 min	1,793	1.6 GB	registered LiDAR map (14.8 M points), Ouster IMU and metadata

Sensors

Sensor	Make and model	In this dataset	Source
LiDAR	Ouster OS1-64, Gen1 hardware (part 840-101855-02, serial 991951000513), 64 channels, 33.6° vertical field of view, 1024 × 10 Hz	every collection	sensor metadata (2023-2025); the beam angles recovered from every other recording match it to 0.01°, so one unit made them all
Radar	Aptiv (Delphi) ESR 2.5, 24 V, long range ±10° to 174 m, mid range ±45° to 60 m	radar tracks, `highway_2020` and `niro_2021`	Brown et al. 2022
Second radar	most likely a smartmicro UMRR-96 Type 153 (inferred)	target lists in two `radar_tests_2023-24` recordings	Brown et al. 2023 sensor list
GNSS	Swift Navigation Duro (Piksi Multi), SBAS or SPP fixes, none RTK	`gnss.parquet` in six `niro_2021` drives	Brown et al. 2022
LiDAR IMU	Ouster built-in IMU	`ouster_imu.parquet`

The vehicles also carried sensors whose data are not in this dataset. The Kia Niro's Mobileye 630, Stereolabs ZED 2 stereo camera and FLIR thermal camera in 2021 will be released with the lab's MI-Snow1000 dataset. The cameras, GNSS, road weather and vehicle CAN data of the 2023 urban drive, the radar tests and the 2025 shuttle are in the companion dataset AD-Assurance-Lab/lab-drives, and so are the cameras of the two calibration recordings. The lab's other Ouster recordings, the NCAR weather runs (2020-21) and a 2022 HD-map run, are public as raw ROS bags in AD-Assurance-Lab/infrastructure-sensing-doe-eems116.

Files

highway_2020/radar_lidar_highway/       lidar/, ouster_imu.parquet, radar_tracks.parquet, drive.json
niro_2021/<drive>/                      lidar/, ouster_imu.parquet, radar_tracks.parquet, gnss.parquet, drive.json
urban_2023-12/<recording>/              lidar/, ouster_imu.parquet, recording.json
radar_tests_2023-24/<recording>/        the same, plus radar_targets.parquet where a second radar was logged
shuttle_2025-01/<recording>/            lidar/, ouster_imu.parquet, ouster_metadata.json, recording.json
shuttle_2025-01/lidar_map/              scans.pcd (registered map), cropped map, single scans, 4x4 registration transform
calibration_2023-24/<recording>/        lidar/, recording.json, and ouster_imu.parquet for 2024-05-10

Each lidar/scans_*.parquet row is one full 360° scan of 65,536 points:

Column	Type	Meaning
`timestamp_ns`	int64	ROS receive time, ns since the Unix epoch
`header_stamp_ns`	int64	the driver's scan time stamp
`xyz`	float32 × 196,608	x, y, z in metres; (0, 0, 0) means no return
`range_mm` (2020-21) or `range` (2023-25)	uint32 × 65,536	range in mm
`intensity`	float32 × 65,536	signal strength
`reflectivity`	uint16 × 65,536	calibrated reflectivity
`noise` (2020-21) or `ambient` (2023-25)	uint16 × 65,536	ambient light

Point order differs by driver. The 2020-2021 drives were logged with the older ouster_ros driver, which orders points by azimuth column, then beam: reshape to (1024, 64). The 2023-2025 recordings use the newer driver, which orders them by beam, then column: reshape to (64, 1024). The per-beam elevation angles are in ouster_metadata.json (radar tests and shuttle); the same values apply to every recording, since one sensor made them all.

import numpy as np, pyarrow.parquet as pq
t = pq.read_table('niro_2021/2021-01-18_fev_13-03-56/lidar/scans_000.parquet')
scan = np.asarray(t.column('xyz')[0].values)
xyz = scan.reshape(1024, 64, 3)          # 2020-21 drives; use (64, 1024, 3) for 2023-25
pts = scan.reshape(-1, 3)
pts = pts[np.abs(pts).sum(1) > 0]        # drop empty returns

Frames and alignment

2021 drives (Kia Niro). Points are in the os1_lidar frame, whose x axis points to the rear of the car (Ouster's lidar frame faces the connector). Forward is −x and left is −y. No LiDAR-to-radar calibration was recorded; matching every radar track to the point clouds puts the radar about 2 to 2.5 m ahead of the LiDAR on the same centre line.
highway_2020. Also the os1_lidar frame, but the sensor was mounted the other way round: forward is +x (a highway overpass approaches along +x, and the radar tracks match only there). No radar offset could be fitted.
Radar tracks (2020-21). In the radar frame (x forward, y left, m): x, y are the centre of each track's outline and vx_rel, vy_rel are relative to the car.
2023-2025. Points are in the os_sensor frame. The mounting on each vehicle was not recorded.

Known limits

niro_2021/2021-02-04_01 logged LiDAR at about 3.3 Hz (scans were dropped while recording).
niro_2021/2021-01-17_02 has no radar. GPS is present in six Niro drives (2021-01-16, 2021-01-18 and 2021-01-31).
niro_2021/2021-03-17_fev_16-08-45 is a truncated recording, salvaged to its first 154.5 s.
Some of the 2020-2021 bags mislabel a few message connections; messages that did not match their topic were skipped.
Two 2021 Niro drives in the lab's archive (2021-01-31_05, 2021-02-18_01) have no LiDAR and are not included.
Two short "fusion" bags in the lab's archive, stamped 2024-07-01, are not included. They are replays of two excerpts of the urban_2023-12 drive (about 299 to 329 s and 375 to 392 s into it): every scan and camera frame in them is identical to the urban drive's.

Citation

CC BY 4.0. Please cite the paper for the data you use:

highway_2020, niro_2021 (the lab's 2019 Kia Niro research vehicle and its sensors): N. E. Brown, J. F. Rojas, N. A. Goberville, H. Alzubi, Q. AlRousan, C. Wang, S. Huff, J. Rios-Torres, A. R. Ekti, T. J. LaClair, R. Meyer and Z. D. Asher, "Development of an Energy Efficient and Cost Effective Autonomous Vehicle Research Platform," Sensors 22(16):5999, 2022, doi:10.3390/s22165999. Related work on the same car's radar: N. E. Brown, J. Fanas Rojas, H. Alzu'bi, Q. Alrousan, R. Meyer and Z. D. Asher, "Higher Accuracy and Lower Computational Perception Environment Based Upon a Real-time Dynamic Region of Interest," SAE Technical Paper 2022-01-0078, doi:10.4271/2022-01-0078.
radar_tests_2023-24: related to N. E. Brown, P. Patil, S. Sharma, P. Kadav, J. Fanas Rojas, G. Y. Hong, D. Liao, A. Ekti, C. Wang, R. Meyer and Z. D. Asher, "Real World Use Case Evaluation of Radar Retro-reflectors for Autonomous Vehicle Lane Detection Applications," SAE Technical Paper 2024-01-2042, doi:10.4271/2024-01-2042. The raw bags of the four 2024-01-31 recordings are also in AD-Assurance-Lab/infrastructure-sensing-doe-eems116.
urban_2023-12, shuttle_2025-01, calibration_2023-24: no paper yet; please cite this dataset.

@misc{wmu_ouster_lidar_drives_2026,
  title        = {Ouster LiDAR drives from the WMU AD Assurance Lab},
  author       = {{AD Assurance Lab, Western Michigan University} and Asher, Zachary D.},
  year         = {2026},
  howpublished = {Hugging Face dataset},
  url          = {https://huggingface.co/datasets/AD-Assurance-Lab/ouster-lidar-drives}
}

Provenance

The 2020 highway drive and the 2021 drives were recorded on the lab's Kia Niro. Five of the 2021 drives come from the Revision Autonomy dataset and six short ones were recorded for an FEV industry project on the same car. The 2023 urban drive, the 2023-2024 radar tests and the 2023-2024 calibration recordings were made with the lab's vehicles and sensors at Western Michigan University. The 2025 drive was recorded on a campus shuttle; its camera is in lab-drives.

The four 2024-01-31 radar tests were part of U.S. Department of Energy award DE-EE0009657 (EEMS116).

Contact: Zachary D. Asher, Western Michigan University, zach.asher@wmich.edu.

Downloads last month: 142

Total file size:

33.3 GB