A Systematic Control Parameter Tuning Framework Based on a Dual-Loop Model Predictive Control Architecture

Abstract Systematic and effective control parameter tuning is critical to achieving the desired control performance. However, existing automated tuning methods, such as evolutionary algorithms and machine learning techniques, are often time-consuming, data-dependent, and difficult to generalize across different control methods. This paper introduces a novel approach that applies model predictive control (MPC) for control parameter tuning for various systems. Unlike traditional MPC practices that are designed to directly generate control inputs, the proposed framework uses MPC to optimize control parameters, leveraging the inherent stability and feasibility of predesigned controllers. Additionally, an event-triggered MPC strategy is proposed, where an activation trigger improves computational efficiency by activating the MPC optimizer only under critical conditions, such as significant state deviations or large time intervals. Two examples are given to validate the proposed tuning approach: the lookahead distance optimization in pure pursuit control for individual dynamic systems and flocking control parameter calibrations for multi-agent dynamic systems. Simulation results demonstrate that the proposed method can optimize control parameters to achieve better control performance, such as faster convergence and more robust, while maintaining the intrinsic properties of the predesigned controllers.