Contract-based Hierarchical Model Predictive Control and Planning for Autonomous Vehicle

Abstract

Planning and control of autonomous vehicles are becoming increasingly important for many applications. However, autonomous vehicles are often subject to disturbances and uncertainties, which become critical especially in cluttered and dynamic environments. To provide guaranteed constraints satisfaction, e.g. for collision avoidance, we propose a hierarchical model predictive control and planning approach. The moving horizon planning layer and the low-level model predictive controller agree on a “contract” (precision conditions). The high-level moving horizon planner is based on a mixed-integer programming formulation using a simplified model on a slow time scale, and constraint tightening. The autonomous vehicle itself is controlled by a lower-level tube-based model predictive controller. The decomposition of the control problem reduces the computational cost, enables real-time implementation while it allows to provide guarantees. To ensure compatibility between the levels and guarantee safety, we do explicitly consider the problem of recursive feasibility of the hierarchical controller ensuring constraint satisfaction and obstacle avoidance, despite the action of (unknown) disturbances. Simulation results illustrate the efficiency and applicability of the proposed hierarchical strategy.