Autonomous control of scaled racing cars is an increasingly popular approach for testing and presenting new control algorithms while also reducing the operational costs required to bring such a model along with a proper racing track to life. In most cases where autonomous vehicles come into play, the environment map is not known a priori, in which case the algorithm should sufficiently drive the car in real time, without having planned an optimal trajectory. This study presents a reactive method based solely on LiDAR readings that successfully navigates the car through unknown dynamic environments with obstacles, taking into account the principle of safety: not to crash into obstacles or other vehicles. We show that the algorithm significantly outperforms two other existing reactive algorithms and renders a cost-effective solution to the problem of autonomous driving on unknown maps, as well as it renders itself as a possibly good “specialist" model in Expert Intervention Learning (EIL) approach-based models for autonomous vehicles, especially the F1TENTH cars.
Reactive Control Algorithm for F1TENTH Autonomous Vehicles in Unknown Dynamic Environments
Lect. Notes in Networks, Syst.
Proceedings of the XXI Polish Control Conference ; 2023 ; Gliwice, Poland June 26, 2023 - June 29, 2023
2023-06-14
10 pages
Article/Chapter (Book)
Electronic Resource
English
Navigation of Autonomous Vehicles in Unknown Environments Using Reinforcement Learning
British Library Conference Proceedings | 2007
|