Stochastic MPC with Situation-Aware Dynamic Risk Assessment for Autonomous Driving

In this paper, we propose a novel interplay between stochastic model predictive control and situation-aware dynamic risk assessment (SINADRA) for autonomous driving. With the help of SINADRA a Bayesian network is engineered which infers a probabilistic behavior classification for a surrounding actor. This classification is considered in the chance constraints of the model predictive controller (MPC). The MPC can guarantee no constraint violations up to a chosen probability. If a constraint violation occurs, the system is handled by an emergency braking assistant (EBA) ensuring the safety of the overall system. The approach of combining MPC with SINADRA results in a less conservative driving behavior while not introducing additional safety violations. The probability of interference by the EBA can be used as a design parameter, that can be tuned by the manufacturer or possibly by the end-consumer through a safe interface.