CAS ETH in Applied Automation Technology

The course will include the application of case studies, hands-on exercises, moderated group discussions, and guest talks, as well as an excursion and public networking events.

This first module serves as an introduction to the topics of systems theory and automatic control. We start with a brief overview of the role of control theory throughout history and up until current times. The participants will realize the fundamental importance of such hidden science that enables the vast majority of modern technologies and infrastructures. We focus the attention on the fundamental control scheme, the role of feedback, and we discuss how mathematics can be used to model physical systems via differential equations.

Dynamical systems are essentially mathematical models that are used to describe reality, from robotic arms to infectious diseases to human decision making. In this module, we explore the most important aspects of dynamical systems, such as solutions, equilibria, stability, controllability, and observability. From an practitioner/engineering perspective, all of these properties constitute the fundamental bulding blocks that are used to design automatic controllers, i.e., that automatically regulate the behaviour of the system.

In this module we leave the domain of analysis and modelling, and enter the domain of control. The objective of a controller is to gather information on the state of the system (from the sensors) and turn it into corrective actions (via the actuators). By designing smart controllers, we can guarantee the system automatically behaves in a desired manner and performs required tasks. In this first control-oriented module, we discuss the difference between open-loop and closed-loop control, and explore the intuitions behind the Proportional, Integral, and Derivative action of the PID controller. 

Real world control problems are often complex, saftety-critical, and performance oriented. That's why PID control alone might not be enough to guarantee performance and satisfy requirementes, and needs to be complemented by more sophisticated algorithms. In this module we explore modern control techniques that are the industry gold standard in several domains. We first introduce the general framework of optimal control, where optimization tools are used to maximize performance and minimizing costs. Model Predictive Control (MPC) is a powerful formulation to design fast on-line controllers which can handle constraint satisfaction for safetey guarantees. Finally, to break down the complex interactions and reduce the computatonal power needed to tackle large-scale problems, we discuss multiagent control systems.

Besides complexity, another aspect is crucially important and needs to be addressed with the right tools: uncertainty. The mathematical models we build might not be sufficiently accurate to represent real world behaviour, the information gathered by the sensors might be affected by noise, the corrective actions we plan via the actuators might be delayed more than we expect. To cope with these practical considerations and many more, we introduce widely-used methods such as system identification, state estimation, and robust control.

Perhaps the largest revolution in recent history of systems and control has been the development of data-driven and reinforcement learning methods. Data pervade each and every corner of most industries and represent an extremely valuable asset. We show how to make the most use of data to design smart controllers that optimize performance and satisfy safety requirements. We introduce the most relevant reinforcement learning techniques and introduce the recent theory of data-enabled predictive control.