Papers
17
Total Citations
863
H-Index
11
About
Jonathan Scholz is a robotics researcher whose work sits at the intersection of reinforcement learning, motion planning, and data-driven robot manipulation. He is perhaps best known for his 2017 paper "Leveraging Demonstrations for Deep Reinforcement Learning on Robotics Problems with Sparse Rewards," which has accumulated over 510 citations and remains a landmark contribution to the field. By extending the DDPG algorithm to incorporate expert demonstrations alongside real interactions, Scholz addressed one of robotics' most stubborn challenges: learning effective policies when reward signals are rare and costly to obtain. Beyond this foundational work, Scholz has made sustained contributions to robot navigation in cluttered environments, particularly through his research on Navigation Among Movable Obstacles, where he developed decision-theoretic planners capable of handling uncertain real-world dynamics. His later work on reward sketching and batch reinforcement learning demonstrates a commitment to scaling data-driven robotics to practical, multi-task settings without requiring exhaustive hand-engineering. His research on self-supervised semantic keypoints further reflects a broad interest in building richer perceptual representations for manipulation. Across more than a decade of contributions, Scholz has consistently bridged theoretical rigor with real-world applicability, making him a notable figure in modern robot learning research.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4A Physics-Based Model Prior for Object-Oriented MDPs40 citations · 2014
- 5
- 6
- 7Combining motion planning and optimization for flexible robot manipulation34 citations · 2010
- 8
- 9Navigation Among Movable Obstacles with learned dynamic constraints18 citations · 2016
- 10