Papers
13
Total Citations
757
H-Index
9
About
Josh Merel is a leading researcher in reinforcement learning, imitation learning, and robotics, whose work bridges the gap between simulation and real-world motor control. He is best known for developing scalable methods that combine deep reinforcement learning with small amounts of demonstration data to train complex visuomotor policies. His seminal paper "Reinforcement and Imitation Learning for Diverse Visuomotor Skills" (217+ citations) introduced a model-free approach enabling robots to learn manipulation tasks directly from RGB camera inputs. Merel also co-created dm_control (186+ citations), the widely-used software package that provides standardized tasks and MuJoCo-based infrastructure for continuous control research. His work on "Catch & Carry" (98+ citations) addressed the long-standing challenge of producing realistic, vision-guided whole-body controllers for humanoid characters interacting with objects. More recently, he has explored transferring movement skills from human and animal motion capture data to real legged robots, and investigating flow-based navigation strategies inspired by aquatic animals. Across his highly-cited publications, Merel has advanced the frontiers of imitation learning, reusable skill modules, and physically-grounded AI.
Research Focus
Key Achievements
Top Papers
- 1Reinforcement and Imitation Learning for Diverse Visuomotor Skills217 citations · 2018
- 2dm_control: Software and tasks for continuous control186 citations · 2020
- 3Reinforcement and Imitation Learning for Diverse Visuomotor Skills116 citations · 2018
- 4Catch & Carry98 citations · 2020
- 5Robust Imitation of Diverse Behaviors64 citations · 2017
- 6
- 7
- 8
- 9Neuroprosthetic Decoder Training as Imitation Learning12 citations · 2016
- 10Flow Currents Support Simple and Versatile Trail-Tracking Strategies8 citations · 2023