Fabian Schramm
Université Paris Sciences et Lettres, École Normale Supérieure
Papers
4
Total Citations
28
H-Index
2
About
Fabian Schramm is a leading robotics researcher whose work sits at the intersection of optimal control, numerical optimization, and reinforcement learning for humanoid and exoskeleton systems. His research focuses on bridging the critical gap between simulation and real-world hardware, particularly in the domains of bipedal locomotion and push recovery. Schramm’s most impactful contribution is his pioneering work on reactive stepping for humanoid robots, where he demonstrated that reinforcement learning can successfully learn robust balancing and recovery behaviors in simulation, and then transfer those policies to the real-world Atalante exoskeleton—a breakthrough that has garnered 15 citations. He is also the lead developer of ProxQP, a state-of-the-art convex quadratic programming solver designed for real-time robotics applications, which has already earned 9 citations for its efficiency and versatility in whole-body control and planning. Additionally, Schramm has advanced the field of nonsmooth optimal control by introducing randomized smoothing techniques that enable Differential Dynamic Programming to handle discontinuous dynamics, and he developed QPLayer, a framework for differentiating through convex optimization layers in neural networks. His work is essential for researchers tackling the challenges of real-time, robust control in legged robotics.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4QPLayer: efficient differentiation of convex quadratic optimization2 citations · 2023