Sergey Levine is a professor in the Department of Electrical Engineering and Computer Sciences at UC Berkeley, where he is affiliated with the Berkeley Artificial Intelligence Research (BAIR) Lab. He received his B.S., M.S., and Ph.D. in computer science from Stanford, completing his doctorate in 2014, and joined the Berkeley faculty in 2016.
Levine’s research centers on machine learning for decision making and control, with an emphasis on deep reinforcement learning for robotics. His group is known for algorithms that train deep neural network policies to combine perception and control in a single end-to-end system, rather than treating sensing and acting as separate engineered stages, as well as for advances in inverse reinforcement learning and offline reinforcement learning that lets robots learn from previously collected data. He co-authored the TRPO algorithm with John Schulman and Pieter Abbeel and the model-agnostic meta-learning (MAML) method with Chelsea Finn.
In 2024 Levine co-founded Physical Intelligence, a startup building foundation models to control robots and other physically actuated devices, part of a broader push to bring the scaling recipe of large language models to embodied machines.