Safe Control Architecture via Model Predictive Control

Abstract

Ensuring the safe operation of autonomous systems is a critical challenge that demands the development of sophisticated control strategies. This article proposes a safe control architecture (SCA) that employs a supervisor model predictive control (MPC) (supervisor) strategy to ensure the persistent satisfaction of state and input constraints. The supervisor continuously monitors the safety of potentially unsafe inputs generated by an operating controller (OC). If an input is predicted to lead the system to a future state where constraint violations are inevitable, it is deemed unsafe and thus blocked from the system. Instead, a backup input, generated by the supervisor in the previous time step, is applied to the system. However, uncertainties in system dynamics are unavoidable and can lead to incorrect decisions by the supervisor, which is based on MPC with a nominal model. This article proposes to enhance the robustness of the SCA by the integration of tube MPC. The resulting robust SCA (RSCA) has the capability to ensure safe operation of autonomous systems under model uncertainties, making it a practical solution for safety-critical autonomous systems, such as vehicles, drones, or medical robots. This article also proves the recursive feasibility and stability of the RSCA. The effectiveness of the approach is validated for an autonomous vehicle in IPG CarMaker, a high-fidelity simulation environment with realistic data on roads, vehicle dynamics, and obstacles.