Towards a Software Component to Perform Situation-Aware Dynamic Risk Assessment for Autonomous Vehicles

Assuring an adequate level of safety is the key challenge for the approval of autonomous vehicles (AV). Dynamic Risk Assessment (DRA) enables AVs to assess the risk of the current situation instead of behaving according to worst-case expectations regarding all possible situations. While current DRA techniques typically predict the behavior of others based on observing kinematic states, Situation-Aware Dynamic Risk Assessment (SINADRA) uses probabilistic environmental knowledge about causal factors that indicate behavior changes before they occur. In this paper, we expand upon previous conceptual ideas and introduce an open-source Python software component that realizes the SINADRA pipeline including situation class detection, Bayesian network-based behavior intent prediction, trajectory distribution generation, and the final risk assessment. We exemplify the component's usage by estimating front vehicle braking risks in the CARLA AV simulator.