Stochastic Model Predictive Control Approach to Autonomous Vehicle Lane Keeping

Abstract

In real-world scenarios, the uncertainty of measurements cannot be handled efficiently by traditional model predictive control (MPC). A stochastic MPC (SMPC) method for handling the uncertainty of states in autonomous driving lane-keeping scenarios is presented in this paper. A probabilistic system is constructed by considering the variance of states. The probabilistic problem is then transformed into a solvable deterministic optimization problem in two steps. First, the cost function is separated into mean and variance components. The mean component is calculated online, whereas the variance component can be calculated offline. Second, Cantelli’s inequality is adopted for the deterministic reformulation of constraints. Consequently, the original probabilistic problem is transformed into a quadratic programming problem. To validate the feasibility and effectiveness of the proposed control method, we compared the SMPC controller with a traditional MPC controller in a lane-keeping scenario. The results demonstrate that the SMPC controller is more effective overall and produces smaller steady-state distance errors.