How can technical systems decide what actions to take based on sensor information? How can a robot learn to move, a vehicle adapt to changing conditions, or a machine improve its performance over time? And how can methods ranging from reinforcement learning to model-predictive control be combined to solve challenging control problems?
This MOOC is your gateway to learning-based control (LBC), an active area of research at the intersection of control systems and machine learning. Control is a key principle behind many technical systems: it uses feedback from the system and its environment to decide what actions to take. LBC extends this idea by using machine learning in the design or operation of control systems, making it an important basis for autonomous systems such as robots, vehicles, and other intelligent machines.
The course starts with the core ideas of feedback control and dynamical systems. Building on this foundation, we introduce central learning-based control approaches: dynamics learning, controller tuning with Bayesian optimization, reinforcement learning for control, and learning-based model-predictive control. Along the way, programming examples on a standard control benchmark help connect the concepts to practical implementation.
By following this course, you will learn when and where machine learning can be useful in control, how to reason about the strengths and limitations of different learning-based control methods, and how to start studying new methods from current research. Overall, the course equips you with skills to address real-world control challenges, to judge when and where learning-based control can be useful, and to study new methods.
