Yatian Pang 庞雅天

I am a Research Scientist at NVIDIA Cosmos Lab. My long-term goal is to build general-purpose AI agents that can reason about the physical world, simulate possible futures, and choose effective actions using world models that unify multimodal understanding and generation.

I obtained my Ph.D. at the National University of Singapore. I feel incredibly lucky to be supervised by Prof. E. H. Francis Tay, collaborate closely with Prof. Li Yuan from PKU, and gain valuable experience working with Alibaba's Qwen, Everlyn AI, and A*STAR.

Email Scholar GitHub LinkedIn

Latest

Jun 2026 Cosmos 3 released ↗
Jan 2026 Officially Dr. P!
Dec 2025 Join NVIDIA as Research Scientist!

Selected projects

Research in focus

Understanding · Generation · Action

World models · 2026

Cosmos 3

NVIDIA Cosmos Team, Yatian Pang (2026)

Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture.

Project ↗ Code ↗

Multimodal understanding · 2025

Qwen3-VL

Qwen Team, Yatian Pang (2025)

One of the most popular open-sourced VLMs. Key contributions to long and streaming video understanding within the Qwen3-VL framework, achieving SOTA results across multiple video benchmarks.

Project ↗ Code ↗

Video generation · 2024

Open-Sora-Plan

Open-Sora-Plan Team, Yatian Pang (2024)

Pioneering open-source effort to reproduce Sora. Responsible for the high-resolution, long-duration video generation architecture and curating open-source high-quality video datasets.

Paper ↗ Code ↗