Doctoral (PhD) Student Positions in control and learning theory

The Automatic Control Laboratory (IfA) in the Department of Information Technology and Electrical Engineering of ETH Zurich is a community of approximately 50 researchers from more than 20 countries working on the development of methods and computational tools for automation, exploring their potential for promoting our social well-being in areas such as energy systems, transportation, and industrial processes. We are looking for two doctoral students to join our international team and contribute to our research efforts in the area of control and automation and their connections to learning methods. The doctoral students will be supervised by Professor John Lygeros in collaboration with Dr. Efe Balta from Inspire AG and Professors Giancarlo Ferrari-Trecate and Daniel Kuhn from EPFL, under the auspices of the NCCR Automation.

The NCCR Automation is a National Centre of Competence in Research that fosters the collaboration of numerous researchers from across Switzerland with interests in automatic control, related fields such as optimisation and machine learning, and their applications. The extensive activities of the NCCR Automation in education (ranging from primary to continuing education), technology transfer, entrepreneurship, outreach, equal opportunities, and open source/access/data, support our researchers in expanding their interests and skills beyond research.

Control theory is arguably the technological foundation of the unprecedented drive towards automation we have been experiencing over the past decades. As the systems we are interested in automating become larger and more complex, control methods that can deal with uncertainty and partial information have become increasingly important. This has led to a rapprochement between control theory and methods traditionally associated with machine learning. The two doctoral student positions we are looking to fill aim to explore the interplay between these two areas. They will be integrated in the wider team of the Automatic Control Laboratory and the NCCR Automation ecosystem.

We are looking for two motivated doctoral students to contribute to this effort. The envisioned research will address:

1. Distributionally robust Markov Decision Processes (MDP). Distributional robustness has been studied for finite state and action MDP, where uncertainty can be encoded by constraints on the entries on the stochastic transition matrix. Compared to a standard MDP, robustification typically gives rise to additional regularization terms. The aim is to extend this approach to different types of uncertainty descriptions, structural properties of the underlying chains, and infinite state-action MDP. (In collaboration with Giancarlo Ferrari-Trecate and Daniel Kuhn at EPFL.)
2. Policy gradient for control parametrisations: Policy gradient methods are often associated with deep reinforcement learning and policies parametrised by neural networks. The aim is to extend this approach to the design of policies based on control architectures. (In collaboration with Efe Balta from Inspire AG.)

In addition to methodology development, in both cases we envision testing the methods on benchmark problems, robotic testbeds available at the Automatic Control Laboratory, and real-world applications we cater to in our lab, including energy systems, industrial processes, and mobility.

You are highly motivated and dedicated with a Master’s degree in a field addressing control theory, including electrical or mechanical engineering, or applied mathematics. Programming, modelling, and data analysis skills in python and machine learning/optimization libraries/toolboxes support you in contributing to our ongoing software development efforts. Your spoken and written English skills help you navigate our international environment.

We are offering a multifaceted and challenging position in a modern research environment with excellent infrastructure. The ideal starting date is July 2025 with a planned duration of 4 years.