Experimental validation of data enabled predictive control

Developments in machine learning have increased interest in data-driven methods, particularly in predictive control. Data-enabled predictive control has been claimed to provide a data-driven end-to-end solution. It includes robustness guarantees obtained through regularization, borrowed from popular machine learning methods. These novel methods use data in a new framework, where analysis methods used in adaptive control and prior data-driven methods cannot easily be applied. Full understanding of the apparent robustness is lacking. This project aims to analyze the properties of data-driven predictive control and relate the method to prior results.

The student will implement data enabled predictive control for a quadcopter. Based on the theoretical results from this research project, the student will perform experiments to show when the method does and does not work. This experimental work is an integral part of this project and results from this work are expected to be included in a publication.

Faculty Supervisor:

Klaske van Heusden

Student:

Partner:

Eindhoven University of Technology

Discipline:

Engineering

Sector:

Education

University:

The University of British Columbia - Okanagan

Program: