This project aims to bridge the gap between machine learning and control engineering. These research fields have traditionally evolved more or less separately, but in recent years the intersections in terms of applications as well theoretical challenges have been growing. This project is concerned with sequential decision making in systems whose dynamics are initially unknown, i.e., with adaptive control or Reinforcement Learning (RL) when using the control engineering and machine learning terminologies, respectively.
We will work on problems where disturbances are assumed to be of worst-case nature. In control theory, this assumption is the basis for H•infinity optimal control, which was introduced in the 1980s to counteract the fact that optimization in a statistical setting often gives poor robustness to unmodeled dynamics.
Inspired by the theory for robust control, based on worst-case assumptions, we would like to develop a theory to make RL or adaptive control algorithms robust to unmodeled dynamics.