Research

Primary research focus

Robotics

• World Action Model Policies

  How can we design a compact latent space that dynamically captures all information useful to the action expert, enabling action prediction to make the fullest use of relevant context?

• In-Context Learning

  How can real-world physical properties guide robots toward human-preferred behavior without updating model weights, while enabling robot policies to remember and reuse these preferences?

Research direction

Efficient AI

I am also interested in accelerating video and image generation models.

• Feature Cache and Resume

  Caching and reusing intermediate features to reduce redundant computation in generative models.

• Asynchronous Reconstruction

  Accelerating video and image generation through asynchronous computation.