Papers
331
Total Citations
34,754
H-Index
83
About
Sergey Levine is a leading researcher at the intersection of deep reinforcement learning, robot learning, and decision-making, whose work has fundamentally shaped how autonomous systems acquire complex behaviors. He is perhaps best known for co-developing Trust Region Policy Optimization (TRPO), a landmark algorithm for reliable policy learning that has accumulated over 3,100 citations and remains a cornerstone of modern reinforcement learning. His contributions to foundational RL methodology extend further through Generalized Advantage Estimation and Soft Actor-Critic, together drawing nearly 3,700 citations, advancing the stability and sample efficiency of continuous control algorithms. Levine has been equally influential in bridging deep learning with physical robotics. His pioneering end-to-end visuomotor policy work demonstrated that robots could learn directly from raw pixel inputs without hand-engineered perception pipelines, earning over 3,100 combined citations. Projects like QT-Opt and Deep Visual Foresight pushed scalable, vision-based robotic manipulation into practical territory. More recently, his comprehensive treatment of offline reinforcement learning has helped define an emerging subfield critical for real-world deployment. Across more than 12,000 citations in his most-cited works alone, Levine's research continues to drive autonomous robots closer to genuine generalist intelligence.
Research Focus
Key Achievements
Top Papers
- 1Trust Region Policy Optimization3,141 citations · 2015
- 2Soft Actor-Critic Algorithms and Applications1,952 citations · 2018
- 3High-Dimensional Continuous Control Using Generalized Advantage Estimation1,750 citations · 2015
- 4End-to-end training of deep visuomotor policies1,715 citations · 2016
- 5
- 6End-to-End Training of Deep Visuomotor Policies1,399 citations · 2015
- 7DeepMimic802 citations · 2018
- 8
- 9Deep visual foresight for planning robot motion627 citations · 2017
- 10