Yunlong Song - 宋运龙
I am a Founding Research Scientist at Genesis AI, building robot foundation models for dexterous manipulation. I was awarded the SNSF Swiss Postdoctoral Fellowship (SPF) for a postdoctoral position at MIT. I completed my PhD under Prof. Davide Scaramuzza at the Robotics and Perception Group, part of the Department of Informatics and the Department of Neuroinformatics (University of Zurich and ETH Zurich). During my PhD, I had a research stay at the MIT Biomimetic Robotics Lab working with Prof. Sangbae Kim. Earlier, I earned my Master's at TU Darmstadt under Prof. Jan Peters, focusing on reinforcement learning. The road has taken me from a beautiful farming village in southern China through Shenzhen, Shanghai, Darmstadt, Frankfurt, Berlin, Zurich, Pittsburgh, Boston, and now the San Francisco Bay Area.
Research Interests
My research focuses on developing robot learning systems capable of understanding, reasoning, learning, and acting in complex real-world scenarios. I specialize in reinforcement learning and robot foundation models, with broader experience spanning optimal control, numerical optimization, differentiable simulation, and representation learning. I have hands-on experience with a range of robotic platforms, including the Furuta Pendulum, Barrett WAM Arm, high-performance FPV drone, and MIT Mini Cheetah. I led the development of the first autonomous racing drone to beat human champions using reinforcement learning.
My current focuses are:
- Robot Foundation Models and Systems
- Large-scale Reinforcement Learning
- Multimodal Policy Learning
- Dexterous Manipulation
Selected Publications
-
GENE-26.5: Advancing Robotic Manipulation to Human LevelGenesis AI Blog, 2026 Blog
-
Learning Quadruped Locomotion Using Differentiable SimulationCoRL (oral presentation), 2024 PDF Video
Ph.D. Thesis
-
Learning Robot Control: From Reinforcement Learning to Differentiable SimulationLinks: PDF PhD DefenseCommittee: Davide Scaramuzza Marco Hutter Martin Riedmiller Yuke Zhu