Papers
13
Total Citations
1,142
H-Index
10
About
Steven Bohez is a prominent robotics and machine learning researcher whose work sits at the intersection of deep reinforcement learning, sim-to-real transfer, and legged robot locomotion. He is perhaps best known for his landmark 2018 paper "Sim-to-Real: Learning Agile Locomotion For Quadruped Robots," which has accumulated nearly 800 citations across versions and demonstrated that complex quadruped gaits could be learned entirely from scratch using deep reinforcement learning — eliminating the need for laborious manual engineering. This work helped establish sim-to-real transfer as a central paradigm in modern robot learning. Bohez has also made significant infrastructure contributions, co-developing **dm_control** (186 citations), a widely adopted software suite that has become a standard benchmark environment for continuous control research. His subsequent work has pushed the field forward through innovations in safe reinforcement learning, multi-modal sensor fusion, motion capture imitation, and visually realistic sim-to-real transfer using Neural Radiance Fields. His 2023 Barkour benchmark introduced a meaningful framework for evaluating animal-level agility in quadruped robots. Across his career, Bohez has consistently bridged the gap between simulation and physical deployment, making robust, agile robot locomotion increasingly accessible and reproducible for the broader research community.
Research Focus
Key Achievements
Top Papers
- 1Sim-to-Real: Learning Agile Locomotion For Quadruped Robots673 citations · 2018
- 2dm_control: Software and tasks for continuous control186 citations · 2020
- 3Sim-to-Real: Learning Agile Locomotion For Quadruped Robots114 citations · 2018
- 4
- 5Value constrained model-free continuous control30 citations · 2019
- 6Sensor fusion for robot control through deep reinforcement learning26 citations · 2017
- 7
- 8
- 9
- 10Barkour: Benchmarking Animal-level Agility with Quadruped Robots13 citations · 2023