MARS-LVIG dataset: A multi-sensor aerial robots SLAM dataset for LiDAR-visual-inertial-GNSS fusion
Haotian Li,Yuying Zou,Nan Chen,Jiarong Lin,Xiyuan Liu,Wei Xu,Chunran Zheng,Rundong Li,Dongjiao He,Fanze Kong,Yixi Cai,Zheng Liu,Shunbo Zhou,Kaiwen Xue,Fu Zhang
DOI: https://doi.org/10.1177/02783649241227968
2024-01-27
The International Journal of Robotics Research
Abstract:The International Journal of Robotics Research, Ahead of Print. In recent years, advancements in Light Detection and Ranging (LiDAR) technology have made 3D LiDAR sensors more compact, lightweight, and affordable. This progress has spurred interest in integrating LiDAR with sensors such as Inertial Measurement Units (IMUs) and cameras for Simultaneous Localization and Mapping (SLAM) research. Public datasets covering different scenarios, platforms, and viewpoints are crucial for multi-sensor fusion SLAM studies, yet most focus on handheld or vehicle-mounted devices with front or 360-degree views. Data from aerial vehicles with downward-looking views is scarce, existing relevant datasets usually feature low altitudes and are mostly limited to small campus environments. To fill this gap, we introduce the Multi-sensor Aerial Robots SLAM dataset (MARS-LVIG dataset), providing unique aerial downward-looking LiDAR-Visual-Inertial-GNSS data with viewpoints from altitudes between 80 m and 130 m. The dataset not only offers new aspects to test and evaluate existing SLAM algorithms, but also brings new challenges which can facilitate researches and developments of more advanced SLAM algorithms. The MARS-LVIG dataset contains 21 sequences, acquired across diversified large-area environments including an aero-model airfield, an island, a rural town, and a valley. Within these sequences, the UAV has speeds varying from 3 m/s to 12 m/s, a scanning area reaching up to 577,000 m2, and the max path length of 7.148 km in a single flight. This dataset encapsulates data collected by a lightweight, hardware-synchronized sensor package that includes a solid-state 3D LiDAR, a global-shutter RGB camera, IMUs, and a raw message receiver of the Global Navigation Satellite System (GNSS). For algorithm evaluation, this dataset releases ground truth of both localization and mapping, which are acquired by on-board Real-time Kinematic (RTK) and DJI L1 (post-processed by its supporting software DJI Terra), respectively. The dataset can be downloaded from: https://mars.hku.hk/dataset.html.
robotics