Develops generative and vision-language models for unified understanding and synthesis across text, images, video, and 3D. Emphasis on efficient diffusion, 4D scene reconstruction, video depth and tracking, and aligning embodied agents to plan and act from multimodal experience.