Risk in Stochastic and Robust Model Predictive Path-Following Control for Vehicular Motion Planning

Abstract: In automated driving, risk describes potential harm to passengers of an autonomous vehicle (AV) and other road users. Recent studies suggest that human-like driving behavior emerges from embedding risk in AV motion planning algorithms. Additionally, providing evidence that risk is minimized during the AV operation is essential to vehicle safety certification. However, there has yet to be a consensus on how to define and operationalize risk in motion planning or how to bound or minimize it during operation. In this paper, we define a stochastic risk measure and introduce it as a constraint into both robust and stochastic nonlinear model predictive path-following controllers (RMPC and SMPC respectively). We compare the vehicle's behavior arising from employing SMPC and RMPC with respect to safety and path-following performance. Further, the implementation of an automated driving example is provided, showcasing the effects of different risk tolerances and uncertainty growths in predictions of other road users for both cases. We find that the RMPC is significantly more conservative than the SMPC, while also displaying greater following errors towards references. Further, the RMPCs behavior cannot be considered as human-like. Moreover, unlike SMPC, the RMPC cannot account for different risk tolerances. The RMPC generates undesired driving behavior for even moderate uncertainties, which are handled better by the SMPC.