Yunlong Song - 宋运龙

I am currently a Research Scientist at Skild AI, developing robot learning algorithms for general-purpose robot manipulation. I completed my PhD under the supervision of Prof. Davide Scaramuzza at the Robotics and Perception Group, which is part of the Department of Neuroinformatics, a joint institute of University of Zurich and ETH Zurich. During my PhD, I also conducted research at MIT, working with Prof. Sangbae Kim on quadruped locomotion. Additionally, I had the opportunity to collaborate with SONY AI on building super-human game agent for Gran Turismo and with Vladlen Koltun on autonomous drone racing. Prior to my PhD, I earned my master's degree from TU Darmstadt, where I was supervised by Prof. Jan Peters.

PhD Defense

Research Interests

My research focuses on developing intelligent robotic systems capable of learning, adapting, and interacting in complex real-world environments. I have explored various topics in robotics and machine learning, including optimal control, reinforcement learning, differentiable simulation, representation learning, and vision-based control. Beyond theoretical understanding, I have hands-on experience with a range of robotic platforms, including the Furuta Pendulum, Barrett WAM Arm, high-performance FPV drone, MIT Mini Cheetah, and robotic manipulators. My past research has primarily focused on developing robust skill-level policies for diverse robotic tasks. Currently, I am particularly interested in integrating large vision-language-action models with low-level robotic skills to enable autonomous systems to perform complex, real-world tasks with greater efficiency and adaptability. My current research interests include:

  • Real-world Reinforcement Learning
  • Reinforcement Learning for Mobile/Whole-body Manipulation
  • Direct Policy Optimization for Pixel-to-Action Control
  • Post-training of Robotics Foundation Models
  • Multimodal Policy Learning

Selected Pulications

Some peronal favorites from my publication list:
  • Learning Vision-based Agile Flight via Differentiable Physics
    Y. Zhang*, Y. Hu*, Y. Song*, D. Zou and W. Lin
  • Learning Quadruped Locomotion Using Differentiable Simulation
    Y. Song, S. Kim, and D. Scaramuzza
  • Reaching the Limit in Autonomous Racing: Optimal Control Versus Reinforcement Learning
    Y. Song, A. Romero, M. Müller, V. Koltun, and D. Scaramuzza
  • Policy Search for Model Predictive Control for Agile Drone Flight
    Y. Song and D. Scaramuzza
  • Flightmare: A Flexible Quadrotor Simulator
    Y. Song, S. Naji, E. Kaufmann, A. Loquercio, D. Scaramuzza
  • Learning High-level Policies for Model Predictive Control
    Y. Song and D. Scaramuzza