Position Uncertainty-Integrated Potential Function for Collision Avoidance Systems Based on Model Predictive Control

Abstract

In this article, an avoidance potential function with an uncertainty-integrated isopotential contour is proposed to perform vehicle collision avoidance with model predictive control (MPC) under the presence of uncertainty. A superquadric-based potential function describing an object is modified to reflect uncertainty, caused by detection sensor noise, using the suggested uncertainty area and surface-morphing function. In particular, the uncertainty area is designed such that sensor uncertainty expands the isopotential contour at the object boundary. Thus, the proposed potential function is robust against uncertainty in the object position. The surface-morphing function is designed to modify the isopotential contour at the object boundary by comparing the size of the uncertainty area and the object. Then, the proposed model is directly integrated into the cost function of MPC to generate the optimal steering angle and engine torque as control inputs. Specifically, the iterative linear quadratic regulator (iLQR) among MPCs is adopted to handle nonlinearity in the system and potential function. The proposed model is validated in the virtual test, which shows that the vehicle avoids both static and moving objects using the proposed model even when uncertainty exists without parameter adjustment.