Uncertainty-adaptive, risk based motion planning in automated driving

This paper investigates a continuous real-time risk assessment approach that considers environment uncertainties in the trajectory planning for automated vehicles. The probabilistic risk criticality measure considers uncertainties in the collision probabilities as well as accident severities through an impact model. A situational independent, once adjusted motion planning module adapts to uncertainties, e.g., induced by the environment perception due to sensor malfunction or disadvantageous sensor conditions such as volatile weather effects or object distances. As result, while driving, the intelligent vehicle gains an equal residual risk independent of the provided environment information quality. The benefits of the proposed approach are shown by real test-drives with changing sensor equipments for collision avoidance by braking and swerving.