Optimal Control and Decision-Making

Master Wahlmodul, Advanced Control Course with Machine Learning Elements

This course covers a set of topics in optimal control, optimization and learning-based control, with a focus on theoretical foundations and practical algorithm design. Through a combination of theoretical insights and algorithmic development, this course provides students with the knowledge and tools needed to design effective control strategies for dynamic systems, with attention to real-world challenges such as disturbances, model uncertainties, and constraint handling.

Note that this course is now jointly offered with Dr. Marion Leibold as a core course. It was previously offered as "Control for Robotics: From Optimal Control to Reinforcement Learning." Advanced topics with further emphasis on robotics are covered in a follow-up course, "Advanced Robot Learning and Decision-Making."

Learning Objectives

At the end of the course, students are able to:

• derive optimal controller equations for mobile robots,
• implement optimal controllers in practice and analyse their properties,
• understand state-of-the-art approaches to learning-based control and reinforcement learning,
• design learning-based controllers to cope with non-idealities such as model errors.

Prerequisites

Motivation and interest in robot learning and control, as well as knowledge in dynamic systems, control theory (including state-space approaches), calculus, probability theory, linear algebra, and programming (Matlab and Python). Additional knowledge and courses on the following topics will be beneficial: basics of optimization, mobile robotics, machine learning, and optimal control.

Teaching and Learning Approach 

The core components of this course are:

• Lectures (introducing the core concepts)
• Tutorials (working through examples)
• Jupyter Notebook Examples (practising the application of the concepts to practical problems and gaining hands-on implementation experience)

Evaluation

The students will demonstrate their knowledge in a two-hour written final exam at the end of the course. The exam will assess the students’ understanding of optimal control and reinforcement learning. Exam questions will cover the lecture content as well as the practical content from the tutorials and Jupyter notebook examples.