Safety Critical NMPC in Obstacle-Existing Scenes for Car-Like Mobile Robots With Limited Detection

Abstract

This paper proposes a safety critical nonlinear model predictive control (NMPC) to track a target in an obstacle-existing scene for a car-like robot with a limited detection ability. A temporary artificial reference within the detection region is employed as a temporary target to reduce the meaningless trajectory prediction beyond the detection region for the vehicle. In order to satisfy the safety requirements for obstacle avoidance, the non-convex admissible set generated by obstacles is considered and a compression-expansion-inflation (CEI) algorithm is presented to map a subspace of the nonconvex admissible set into a convex one. Then, a modification NMPC with the artificial reference is proposed, where the iteration of the NMPC and the selection of the artificial reference are both achieved in the convex mapping set. It results in the appropriate selection of the artificial reference based on the Euler distance and guarantees the feasibility and stability of the NMPC. In this paper, the CEI algorithm does not introduce any conservatism but abandons the partial admissible set by an additional decision-making process. Furthermore, the proposed algorithm is applied in both simulation and real scenes with single and multiple obstacles.