Automated controller tuning for Weighted Multiple Model Adaptive Control

Abstract

Multiple model adaptive control (MMAC) uses models of different operating modes and optimizes a set of system controllers that can ensure stability over all modes. Traditionally, one controller is tuned for every operating mode but the number of operating modes and all their combinations increase exponentially. We present a novel convex-hull-based optimization algorithm that automatically generates a controller parameter set for a set of operating modes. The main contribution of the presented algorithm is that it can find a smaller controller set than the traditional one-controller per operating mode tuning approach, which we demonstrate empirically on a quadcopter trajectory tracking simulation using five different operating modes. The presented algorithm achieves this by constructing a convex hull in the controller parameter space to ensure the chosen parameters are affinely independent, which results in a set of only three controllers for five investigated operating modes without a significant loss in performance.