Nonlinear stochastic predictive control with unscented transformation for semi-autonomous vehicles

This paper presents a novel predictive control approach based on the unscented transformation with recursive feasibility analysis and an experimental validation for lane keeping of semi-autonomous vehicles. The optimization problem to be solved is nonlinear with stochastic disturbances and probability constraints on states. The unscented transformation is utilized to calculate the propagation of disturbed states over the prediction horizon, and the probability constraints are transformed into constraint functions with Chebyshev´s inequality. A sufficient condition for recursive feasibility is proved by considering the worst case of the disturbance realization. Experiments on the lane keeping system with an uncertain driver model validate the effectiveness of the proposed approach.