Driving with guardian: Blending user inputs with safety ensuring barriers

Abstract

This paper considers the practical aspects of using a control barrier function induced by a robust controlled invariant set for supervising user inputs. Our focus application is path following/lane keeping functionality in advanced driver assist systems. We define specific normalized barrier functions for ellipsoidal and polytopic invariant sets. Then, we use the magnitude of these barriers to determine an optimum safe control input as well to determine when to blend this input with the user's input. This method allows for a smooth overriding of the user input and is well defined in the case that the barrier is violated due, for instance, to model mismatch. The method is demonstrated in a realistic simulation environment. These simulations show the proposed method to be a computationally light way to safeguard user inputs in a smooth manner.