ROVR Open Dataset

ROVR-Open-Dataset: A Large-Scale Depth Dataset for Autonomous Driving

Xianda Guo^1* Ruijun Zhang^2* Yiqun Duan^3* Ruilin Wang² Matteo Poggi⁴ Keyuan Zhou⁵ Wenzhao Zheng⁶ Wenke Huang¹ Gangwei Xu⁷
Yanlun Peng⁸ Yuan Si⁹ Qin Zou^1,†

¹Wuhan University ²CASIA ³University of Technology Sydney ⁴University of Bologna ⁵Zhejiang University ⁶University of California, Berkeley ⁷Huazhong University of Science and Technology ⁸Great Wall Motor ⁹ROVR Labs, Inc.

^*Equal contribution. ^†Corresponding author.

ROVR OpenDataset is a large-scale open 3D dataset designed for autonomous driving, robotics, and 4D perception tasks. It features:

One of the largest multi-modal datasets combining LiDAR and camera
Over 10,000 hours of high-quality driving data collected across North America, Europe, and Asia
Multi-sensor inputs: automotive-grade solid-state LiDAR, HD camera, RTK GNSS, and high-precision IMU
Synchronized point cloud and image data for multi-task learning
Semantic segmentation and 3D bounding box annotations coming soon
Centimeter-level positioning accuracy enabled by GEODNET RTK corrections
Free for academic research
For commercial licensing, contact us at support@rovr.network

About Sensors

ROVR uses an automotive-grade sensor suite for reliable data capture under real-world conditions. The setup includes:

LiDAR: 126-beam automotive-grade solid-state LiDAR
Camera: Automotive-grade HD camera
GNSS: Triple-frequency RTK receiver with global centimeter-level accuracy
IMU: Automotive-grade IMU with low bias instability
Processor: Embedded high-performance edge AI unit with 6 TOPS
Storage: 1 TB onboard storage
Protection: IP67 waterproof and dustproof rating

ROVR-Open-Dataset: A Large-Scale Depth Dataset for Autonomous Driving

About Sensors

Multiple Scenes

Multiple Annotations