Adaptive control and reinforcement learning for vehicle suspension control: A review

Abstract

The growing adoption of electric vehicles has drawn a renewed interest in intelligent vehicle subsystems, including active suspension. Control methods for active suspension systems have been a research focus for many years, and with recent advances in machine learning, learning-based active suspension control strategies have emerged. Classically, suspension controllers have been model-based and thus limited by necessarily simplified models of complex suspension dynamics. Learning-based methods address these limitations by leveraging system response measurements to improve the system model or controller itself. Previous surveys have reviewed conventional and preview-based active suspension controllers, but a detailed examination of newer learning-based methods is lacking. This article addresses this gap by presenting the mathematical foundations of these controllers and categorizing existing implementations. The review classifies learning-based suspension control literature into two main categories: adaptive control, which emphasizes stability through online learning, and reinforcement learning, which aims for optimality through extensive system interactions. Within these broader domains, various sub-categories are identified, allowing practitioners and researchers to quickly find relevant work within a specific branch of learning-based suspension control. Furthermore, this article discusses current trends in the field and proposes directions for future investigations. These contributions can serve as a comprehensive guide for the future research and development of learning-based suspension controllers.