LAUSR

Automated Driving System (ADS) requires a high fidelity decision-making strategy to palliate to uncertain environment and changing dynamics of other road users. Considering the uniqueness of each traffic situation, the task of modeling every use-case is nearly impossible. One solution is to verify the safety of the decided/planned maneuvers during the vehicle’s navigation. This will give ability to the system to re-plan and evade any dangerous situation. The main focus of this work relies on guaranteeing safety of the ADS in sudden hazardous and risky situation. In this aim, an evasive strategy is proposed as a part of an overall Probabilistic Multi-Controller Architecture (P-MCA) designed for safe automated driving under uncertainties. This P-MCA is composed of several complementary interconnected modules, and addresses thus the full pipeline from risk assessment, path planning to decision-making and control for an ADS. The evasive strategy relies on two identified steps. The first step is performed through the decision-making framework, where a Sequential Decision Networks for Maneuver Selection and Verification (SDN-MSV) calculates a discrete evasive action maneuver based on defined situational criteria. The second step consists in computing the corresponding low-level control. It is based on the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) that allows the ego-vehicle to pursue the advised collision-free evasive maneuver to avert an accident and to guarantee the vehicle’s safety at any time. The reliability and the flexibility of the overall proposed P-MCA and its elementary components have been validated in simulated traffic conditions, with various driving scenarios, and in real-time.