Sergey Levine is an American computer scientist, roboticist, and professor at UC Berkeley. He is widely regarded as one of the most prolific and influential researchers in the field of deep reinforcement learning for robotics.
Guided Policy Search and Scaling Robotic Learning
Levine earned his Ph.D. from Stanford University in 2014. At UC Berkeley, he co-directs the Robotic AI & Learning (RAIL) Lab. His research pioneered Guided Policy Search and algorithms like **QT-Opt**, which allowed fleets of robotic arms to share neural network weights to learn to grasp arbitrary household objects in real time.
Offline Reinforcement Learning
Levine has also pioneered the field of **offline reinforcement learning**, developing mathematical methods that allow AI agents to learn optimal policies from static, pre-existing datasets rather than requiring active trial-and-error, a crucial step for deploying AI in high-stakes fields like medicine and autonomous driving. In 2024, he co-founded Physical Intelligence to build foundation models for robots.
