Papers
16
Total Citations
2,645
H-Index
11
About
Suraj Nair is a researcher at the forefront of robot learning, foundation models, and safe reinforcement learning, whose work spans some of the most consequential challenges in modern AI and robotics. He gained widespread recognition as a co-author of "On the Opportunities and Risks of Foundation Models" (2021), a landmark report that introduced the term "foundation model" to describe large-scale, adaptable systems like GPT-3 and DALL-E — a paper that has since accumulated over 2,100 citations and helped shape an entire field's vocabulary. His contributions to safe reinforcement learning through Recovery RL offer principled solutions to the exploration-safety tradeoff, enabling RL agents to operate more reliably in uncertain real-world environments. In robotics, Nair has pioneered scalable learning approaches, including work on large-scale multi-robot datasets (RoboNet), video prediction for planning, and multimodal imitation learning incorporating audio and vision. His recent research on vision-language-action models — OpenVLA and the FAST tokenization framework — reflects a commitment to making generalist robot policies practical and openly accessible. Across his career, Nair has demonstrated a rare ability to bridge theoretical rigor with real-world applicability, making him a distinctive and impactful voice in embodied AI research.
Research Focus
Key Achievements
Top Papers
- 1On the Opportunities and Risks of Foundation Models2,177 citations · 2021
- 2Recovery RL: Safe Reinforcement Learning With Learned Recovery Zones193 citations · 2021
- 3
- 4OpenVLA: An Open-Source Vision-Language-Action Model39 citations · 2024
- 5FAST: Efficient Action Tokenization for Vision-Language-Action Models28 citations · 2025
- 6
- 7Recovery RL: Safe Reinforcement Learning with Learned Recovery Zones20 citations · 2020
- 8
- 9Model-Based Visual Planning with Self-Supervised Functional Distances17 citations · 2020
- 10RoboNet: Large-Scale Multi-Robot Learning16 citations · 2019