Contingent Nonlinear Model Predictive Control for Collision Imminent Steering in Uncertain Environments

Abstract

A novel uncertainty based contingent model predictive control algorithm is presented for autonomous vehicles operating in uncertain environments. Nominal model predictive control relies on a model to predict future states over a horizon and hence requires accurate models and parameterization. In application, environmental conditions and parameters may be unknown or varying, posing robustness issues for model predictive control. This work presents a new selectively robust adaptive model predictive control algorithm that is applied to collision imminent steering controllers for automotive safety. In this context, uncertainties in the road coefficient of friction are estimated using unscented Kalman filtering and the controller is updated based upon the estimated uncertainties. The utility of the uncertainty based controller is demonstrated in a collision imminent steering scenario and compared to nominal deterministic model predictive control, as well as a baseline adaptive scheme. The results suggest the uncertainty based controller can improve the robustness of model predictive control by nearly 50% for deterministic model predictive control and over 10% for the baseline adaptive scheme.