As the quality of perception systems available for automated driving (AD) increases, we investigate the development of an AD agent based on Reinforcement Learning which exploits underlying systems for longitudinal and lateral control. The goal is addressed by designing high-level actions, trying to imitate the commands of a real driver. The proposed agent is trained in a simulated motorway environment and compared to an agent which outputs low-level actions. Our preliminary results show similar performance results, a more pronounced human-like behaviour and a huge reduction in needed training time because of the higher-level of the available actions.
Investigating High-Level Decision Making for Automated Driving
Capello A.;Forneris L.;Pighetti A.;Bellotti F.;Lazzaroni L.;Cossu M.;De Gloria A.;Berta R.
2023-01-01
Abstract
As the quality of perception systems available for automated driving (AD) increases, we investigate the development of an AD agent based on Reinforcement Learning which exploits underlying systems for longitudinal and lateral control. The goal is addressed by designing high-level actions, trying to imitate the commands of a real driver. The proposed agent is trained in a simulated motorway environment and compared to an agent which outputs low-level actions. Our preliminary results show similar performance results, a more pronounced human-like behaviour and a huge reduction in needed training time because of the higher-level of the available actions.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.