Paul Wohlhart
Papers
8
Total Citations
388
H-Index
5
About
Paul Wohlhart is a leading researcher at the intersection of computer vision, robotics, and machine learning, whose work has fundamentally advanced how robots perceive, reason, and act in the real world. His primary contributions lie in three key areas: bridging the sim-to-real gap for robotic grasping, scaling deep reinforcement learning for real-world deployment, and pioneering vision-language-action models for open-world manipulation. Wohlhart’s most impactful work, the RT-2 model (267 citations), demonstrates how Internet-scale vision-language models can be directly incorporated into end-to-end robotic control, enabling emergent semantic reasoning and unprecedented generalization. He has also developed innovative simulation-to-real techniques, such as randomized-to-canonical adaptation networks, that dramatically reduce the need for costly real-world data collection. Notably, his system for sorting waste in office buildings using a fleet of mobile manipulators showcases the practical deployment of deep RL at scale. Through his research at Google Robotics, Wohlhart has consistently pushed the boundaries of what robots can achieve, from grasping novel objects to following complex human instructions, making him a pivotal figure in the quest for general-purpose robotic intelligence.
Research Focus
Key Achievements
Top Papers
- 1RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control267 citations · 2023
- 2
- 3
- 4Open-World Object Manipulation using Pre-trained Vision-Language Models24 citations · 2023
- 5
- 6
- 7
- 8