This content is not included in your SAE MOBILUS subscription, or you are not logged in.
A Combined LiDAR-Camera Localization for Autonomous Race Cars
- Florian Sauerbeck - Technical University of Munich, Institute of Automotive Technology, School of Engineering and Design, Germany ,
- Lucas Baierlein - Technical University of Munich, Institute of Automotive Technology, School of Engineering and Design, Germany ,
- Johannes Betz - University of Pennsylvania, Real-Time and Embedded Systems Lab (mLab), Department of Electrical & Systems Engineering, USA ,
- Markus Lienkamp - Technical University of Munich, Institute of Automotive Technology, School of Engineering and Design, Germany
ISSN: 2574-0741, e-ISSN: 2574-075X
Published January 06, 2022 by SAE International in United States
Citation: Sauerbeck, F., Baierlein, L., Betz, J., and Lienkamp, M., "A Combined LiDAR-Camera Localization for Autonomous Race Cars," SAE Intl. J CAV 5(1):61-71, 2022, https://doi.org/10.4271/12-05-01-0006.
Autonomous Racing is gaining increasing publicity as an attractive showcase of state-of-the-art technologies and the enhanced algorithms used for autonomous driving. The Indy Autonomous Challenge (IAC) tackled autonomous high-speed wheel-to-wheel racing at the famous Indianapolis Motor Speedway (IMS) in October 2021. To solve this problem, advanced autonomous driving algorithms were developed by each team. In this article, we display a multi-sensor localization approach developed for usage in the IAC. To decouple the lateral and longitudinal position of the ego vehicle, a trackbound coordinate system is used that can be transformed to conventional Cartesian coordinates. The longitudinal motion of the car is tracked via a modified version of the OpenVSLAM that outputs the progress of the already driven distance. The Steel and Foam Energy Reduction (SAFER) barrier, which encloses the whole oval, is detected by a three-dimensional (3D)-LiDAR, and the transformation of the barrier to the ego vehicle is estimated. We have validated the new approach via different simulation methods. Despite the challenging high-speed racing scenario, we achieved accuracies of less than 2 m root mean square error (RMSE) for longitudinal localization and less than 40 cm RMSE for lateral localization, depending on velocity and opponent vehicles.