Gaussian Processes for Model Reference Adaptive Control

The aim of the project is to implement a model reference adaptive control (MRAC) scheme in which Gaussian Processes (GPs) are employed to estimate the adaptive elements that MRAC features. The MRAC algorithm relies on a reference model with an adaptive term to mitigate errors caused by model uncertainty. For this project, it is assumed that the uncertainty term can be described by a time-varying prior mean and covariance and that GPs can be used to learn continuous functions that best describe the mean and the covariance. Therefore the GP-MRAC scheme aims to model uncertainty using a GP-based adaptive element. The GP-MRAC algorithm is tested by tracking a trajectory in the presence of the wing-rock dynamics.

This project was developed as part of the final project for a Parameter Inference and State Estimation graduate class and is based on the paper Bayesian Nonparametric Adaptive Control Using Gaussian Processes, which can be obtained using the following links: Preprint, Full Paper

The project report, results, and code can be found at the link below.

Code

• GP_MRAC_Code (Link)