Intelligence Briefing
Pioneer of deep reinforcement learning for robotics. Runs the Robotic AI and Learning Lab (RAIL) at UC Berkeley. Co-founded Physical Intelligence (Pi) in 2024 with Chelsea Finn, Karol Hausman, and others — raised $1B+ at $5.6B valuation to build foundation models for robotic manipulation. Developed HiL-SERL, enabling robots to master complex tasks like Jenga and motherboard assembly with 100% success in 1-2 hours of human-in-the-loop training.
BS/MS, Computer Science — Stanford University
PhD, Computer Science — Stanford University
Operational History
Co-founded Physical Intelligence
Co-founded Physical Intelligence with Chelsea Finn, Karol Hausman, and others, raising over $1 billion.
foundingDeveloped HiL-SERL
Developed HiL-SERL, enabling robots to master complex tasks with high success rates.
researchEstablished Robotic AI and Learning Lab (RAIL)
Established RAIL at UC Berkeley to advance research in robotic learning.
careerPublished Soft Actor-Critic (SAC)
Published influential work on the Soft Actor-Critic algorithm in reinforcement learning.
researchResearch Scientist at Google Brain
Worked on advanced AI and machine learning projects at Google Brain.
careerPostdoc at UC Berkeley
Completed postdoctoral research at UC Berkeley focusing on robotics and AI.
careerPublished End-to-End Training of Deep Visuomotor Policies
Published significant research on training deep visuomotor policies.
researchPhD in Computer Science
Completed PhD at Stanford University, focusing on robotics and machine learning.
careerAGI Position Assessment
Unknown
Believes in building general-purpose robotic intelligence through scalable learning. Focuses on making robot learning practical and sample-efficient.
Believes in building general-purpose robotic intelligence through scalable learning. Focuses on making robot learning practical and sample-efficient.
Intercepted Communications
“The future of robotics lies in scalable learning and practical applications.”
“HiL-SERL represents a breakthrough in robot learning efficiency.”
“Building general-purpose robotic intelligence is our ultimate goal.”
“Robots must learn from human feedback to be truly effective.”
“The integration of AI and robotics will transform industries.”
Research Output
HiL-SERL: Human-in-the-Loop for Sample-Efficient Reinforcement Learning
2024arXiv
Introduced a novel approach for efficient robot learning.
Learning from Human Feedback in Robotics
2023AI & Robotics Journal
Discussed the importance of human feedback in robot learning.
Soft Actor-Critic Algorithms
2022arXiv
Introduced a new framework for reinforcement learning.
Advances in Offline Reinforcement Learning
2022Journal of Artificial Intelligence Research
Reviewed recent advances in offline reinforcement learning.
Robust Reinforcement Learning with Soft Actor-Critic
2021NeurIPS
Explored robustness in reinforcement learning algorithms.
Offline Reinforcement Learning: A Survey
2020Journal of Machine Learning Research
Comprehensive survey on offline RL techniques.
End-to-End Training of Deep Visuomotor Policies
2018Conference on Robot Learning
Pioneered methods for training robots in complex tasks.
QT-Opt: Scalable Deep Reinforcement Learning for Vision-Based Robotic Manipulation
2018ICRA
Developed a scalable approach for robotic manipulation.
Known Associates
Chelsea Finn
co-founderCo-founder of Physical Intelligence and collaborator in various research projects.
View Dossier →Karol Hausman
co-founderCo-founder of Physical Intelligence and co-author on several key publications.
View Dossier →Tuomas Haarnoja
collaboratorCollaborated on the Soft Actor-Critic research and other projects.
View Dossier →Trevor Darrell
mentorMentor during Sergey Levine's academic career at UC Berkeley.
View Dossier →Organizational Affiliations
Current
Physical Intelligence
Co-founder
2024-present
UC Berkeley
Associate Professor
2023-present
Former
Google Brain
Research Scientist
2021-2023
Source Material
Dossier last updated: 2026-03-04