A Reinforcement Learning Approach to Optimize the longitudinal Behavior of a Partial Autonomous Driving Assistance System

Best Paper AwardInternational audienceThe Partially Autonomous Driving Assistance System (PADAS) is an artificial intelligent co-driver, able to act in critical situations, whose objective is to assist people in driving safely, by providing pertinent and accurate information in real-time about the external situation. Such a system intervenes continuously from warnings to automatic intervention in the whole longitudinal control of the vehicle. This paper illustrates the optimization process of the PADAS, following a statistical machine learning methods - Reinforcement Learning - where the action selection is derived from a set of recorded interactions with human drivers. Experimental results on a driving simulator prove this method achieves a significant reduction in the risk of collision