Bhavya Sukhija Receives Award for Outstanding Master Thesis Conducted at DSME

Left, an overview of the Master thesis and on the right Bhavya Sukhija Copyright: © Bhavya Sukhija  

Bhavya Sukhija has been awarded the ETH Medal for his excellent Master thesis from the Department of Mechanical and Process Engineering at ETH Zurich. Bhavya conducted his Master thesis as an exchange student at the Institute for Data Science in Mechanical Engineering, short DSME, at the RWTH Aachen University, supported by the IDEA League program. He worked on the topic of "Globally Optimal Safe Robot Learning in Higher Dimensions" under the supervision of Dominik Baumann (now at Uppsala University) and Sebastian Trimpe. Bhavya was co-supervised by Matteo Turchetta, David Lindner, and Andreas Krause from ETH Zurich.

In his thesis, Bhavya developed algorithms for learning globally optimal controllers for high-dimensional dynamical systems while providing safety guarantees. When learning controllers from data, we need to allow the learning agent to explore to ideally end up with an optimal policy. However, without prior knowledge, random exploration may lead to violation of safety constraints that damage the system. Bhavya showed theoretically that under certain regularity conditions, we can develop exploration strategies that are guaranteed to be safe with high probability and can find globally optimal policies. He further demonstrated the effectiveness of his developed algorithm on a robot arm (videos are available at the projects website). The results of Bhavya's thesis are currently under review at the Artificial Intelligence Journal, a preprint is available on arXiv.