Honglak Lee

About

Honglak Lee is a prominent researcher at the intersection of deep learning, robotics, and reinforcement learning, whose work has fundamentally advanced how machines perceive, learn, and interact with the physical world. He is perhaps best known for his pioneering application of deep learning to robotic grasping, where his landmark paper "Deep Learning for Detecting Robotic Grasps" demonstrated that neural networks could replace laborious hand-engineered features in RGB-D scene understanding — a contribution that has accumulated over 1,600 citations and remains a touchstone in robot manipulation research. Beyond robotics perception, Lee has made significant contributions to hierarchical reinforcement learning, notably through data-efficient HRL frameworks that make complex task-solving more tractable in real-world settings. His research on generalization in deep RL, including network randomization techniques and context-aware dynamics models, addresses critical challenges in deploying learned policies to novel environments. Earlier foundational work on exponential family sparse coding and self-taught learning reflects his longstanding interest in unsupervised representation learning. From quadruped locomotion to collaborative human-robot construction, Lee's portfolio spans theory and application, making him an influential figure for students interested in building intelligent, adaptable robotic systems.

Research Focus

Key Achievements

Top Papers

1. 1
   Deep learning for detecting robotic grasps

   1,646 citations · 2015
2. 2
   Data-Efficient Hierarchical Reinforcement Learning

   265 citations · 2018
3. 3
   Deep Learning for Detecting Robotic Grasps

   125 citations · 2013
4. 4
   Exponential family sparse coding with applications to self-taught learning

   71 citations · 2009
5. 5
   Network Randomization: A Simple Technique for Generalization in Deep Reinforcement Learning

   48 citations · 2020
6. 6
   Network Randomization: A Simple Technique for Generalization in Deep Reinforcement Learning

   47 citations · 2019
7. 7
   Quadruped robot obstacle negotiation via reinforcement learning

   42 citations · 2006
8. 8
   Mutual physical state-aware object handover in full-contact collaborative human-robot construction work

   33 citations · 2023
9. 9
   Data-Efficient Learning for Sim-to-Real Robotic Grasping using Deep Point Cloud Prediction Networks

   29 citations · 2019
10. 10
    Context-aware Dynamics Model for Generalization in Model-Based\n Reinforcement Learning

    28 citations · 2020

Key Collaborators

Contact & Links