
Zeshi Yang (楊 澤世)
I am an independent researcher focused on advancing video generation and its applications in visual reasoning and planning. My current work explores how to integrate generative video models with reinforcement learning to enable intelligent agents that can perceive, reason, and act within complex visual environments. I am also experienced in 3D motion synthesis and control, with a deep background in physics-based character animation and deep reinforcement learning.
I received my Ph.D. in 2023 from Simon Fraser University, where I worked on controllable and physically realistic character animation. Prior to that, I earned a bachelor's degree in applied physics from USTC in 2018. My long-term goal is to bridge generative AI with decision-making systems to create grounded, interpretable, and task-driven agents.
Education & Career
B.S. in Applied Physics, University of Science and Technology of China, 2014 - 2018
Ph.D in Computer Science, Simon Fraser University, 2018 - 2023
Visiting Scholar, Peking Univerisity, China, 2021 Aug - 2022 Aug
Research Intern, LightSpeed Studios, Tencent America, 2022 Sep - 2023 May
Research Scientist, miHoYo, 2023 June - 2024 Sep
Research Scientist, start-up, 2024 Oct - Now
Publications
Real-time Diverse Motion In-betweening with Space-time Control
ACM SIGGRAPH Conference on Motion, Interaction and Games (MIG 2024)
Learning-based 2D Irregular Shape Packing
ACM Transactions on Graphics, Volume 42, Article 6 (Proc. ACM SIGGRAPH Aisa 2023)
Acquiring Stylized Motor Skills for Physics-based Characters
PhD Thesis, Simon Fraser University, 2023
Learning to Use Chopsticks in Diverse Gripping Styles
ACM Transactions on Graphics, Volume 41, Issue 4, Article 95 (Proc. ACM SIGGRAPH 2022)
Efficient Hyperparameter Optimization for Physics-based Character Animation
ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games 2021 (I3D 2021)
Discovering Diverse Athletic Jumping Strategies
ACM Transactions on Graphics, Volume 40, Issue 4, Article 91 (Proc. ACM SIGGRAPH 2021)
Neural fidelity warping for efficient robot morphology design
2021 International Conference on Robotics and Automation (ICRA 2021)
Redirected Smooth Mappings for Multiuser Real Walking in Virtual Reality
ACM Transactions on Graphics, 38(5), 2019.(Presented at Siggraph Asia)