A Review on Applications of Gaussian Processes for Complex Control Systems

As data-driven machine learning models, Gaussian processes (GPs) have been gradually applied to control challenges in stochastic-uncertain systems in recent years, leveraging their strengths in modeling small-sample noisy data and their inherent capability to provide their own uncertainty estimates. This paper systematically reviews the advantages of GPs and their applications in critical control components, including system dynamics identification, control policy/cost optimization, stability/safety guarantees, and computational acceleration. Additionally, it identifies current limitations of GPs in control applications and proposes several potential open research directions for future investigation.