Papers
55
Total Citations
2,563
H-Index
22
About
Lerrel Pinto is a robotics and machine learning researcher whose work sits at the intersection of robot learning, visual representation, and autonomous manipulation. He is perhaps best known for his landmark 2016 study, "Supersizing Self-Supervision," which demonstrated that robots could learn effective grasping behaviors through large-scale autonomous data collection — accumulating 50,000 attempts over 700 robot hours — bypassing the bottleneck of costly human labeling and earning over 1,000 citations. This foundational contribution helped establish self-supervised learning as a viable paradigm for physical robot training. Pinto's subsequent research has consistently pushed the boundaries of what robots can learn without extensive human guidance. His work addresses deformable object manipulation, dexterous imitation, policy generalization across environments, and semantically rich scene representations like CLIP-Fields. Notably, he has tackled the sim-to-real gap through innovations such as asymmetric actor-critic architectures and self-supervised policy adaptation at deployment time. His 2022 study on representation learning for visual imitation revealed surprisingly strong generalization from compact pre-trained representations, influencing how the field approaches data-efficient robot learning. Across his portfolio, Pinto has shaped modern robotics research by demonstrating that scalable, self-supervised, and representation-driven approaches can make autonomous manipulation both practical and broadly deployable.
Research Focus
Key Achievements
Top Papers
- 1Supersizing self-supervision: Learning to grasp from 50K tries and 700 robot hours1,099 citations · 2016
- 2Learning to Manipulate Deformable Objects without Demonstrations164 citations · 2020
- 3Asymmetric Actor Critic for Image-Based Robot Learning105 citations · 2018
- 4The Surprising Effectiveness of Representation Learning for Visual Imitation82 citations · 2022
- 5The Curious Robot: Learning Visual Representations via Physical Interactions77 citations · 2016
- 6
- 7
- 8CLIP-Fields: Weakly Supervised Semantic Fields for Robotic Memory68 citations · 2023
- 9Robot Learning in Homes: Improving Generalization and Reducing Dataset Bias61 citations · 2018
- 10Self-Supervised Policy Adaptation during Deployment58 citations · 2020