Proximate model predictive control strategy for autonomous vehicle lateral control

Abstract

A proximate model predictive control strategy is proposed for autonomous vehicle lateral control, which substantially reduces iteration in on-line optimization. Nodal state vectors are generated in the feasible state space, for which the quadratic optimization problem is solved off-line. Vertices are determined to represent a given state as an interpolation between them. An approximate optimal solution is computed from the interpolation between the optimal solutions at each vertex, and is used for warm-start on-line optimization to produce a proximate optimal solution. The proposed proximate model prediction control is shown to exhibit proximate optimality in very few on-line iterations.