Yunhao Fang方云浩Let’s chat
Portrait of Yunhao Fang
Hangzhou · 2022

Yunhao Fang 方云浩

Research Scientist at Physical Intelligence

I build multimodal agents that perceive, memorize, interact, and learn from experience. My research spans multimodal understanding and generation, long-context processing, and real-time virtual and physical interaction.

Previously, I worked at ByteDance Seed and NVIDIA, and studied at UC San Diego and Zhejiang University.

Google ScholarGitHubXLinkedIn

Highlights

View publication index ↓
The VILA model family connects image, video, and language understanding.
01 · Multimodal perception

Understanding & generation

VILA² · NVILA · VILA-U · Seed1.5-VL · WorldModelBench
Artificial Hippocampus Networks transform a growing context into compact long-term memory.
02 · Memory

Long-context processing

AHN · LongVILA
03 · Real-time interaction (virtual)

Virtual-world agents

Lumine
03 · Real-time interaction (physical)

Physical-world agents

π₀.₇ · π*₀.₆ · EgoVLA
Equation 4, the Experience Policy Distillation loss for learning from agent experience.
04 · Learn from experience

Experience distillation

Sample-efficient agent learning
Research arc01 — 04

The long view

From multimodal perception to agents that memorize, interact, and learn.

  1. 01Perceive

    Unify multimodal understanding and generation.

  2. 02Memorize

    Process long contexts without losing what matters.

  3. 03Interact

    Operate in virtual and physical worlds in real time.

  4. 04Learn

    Distill agent experience into model weights.

Publication indexSelected
01

Multimodal perception

Understanding & generation

  1. 2025Seed1.5-VL Technical ReportTechnical report
  2. 2025WorldModelBench: Judging Video Generation Models As World Models*NeurIPS Datasets & Benchmarks
  3. 2025NVILA: Efficient Frontier Visual Language ModelsCVPR
  4. 2025VILA-U: A Unified Foundation Model Integrating Visual Understanding and GenerationICLR
  5. 2024VILA²: VILA Augmented VILA*ICCVW Oral
  6. 2023Distilling Large Vision-Language Model with Out-of-Distribution Generalizability*ICCV
02

Memory

Long-context processing

  1. 2026Artificial Hippocampus Networks for Efficient Long-Context Modeling*ICML
  2. 2025LongVILA: Scaling Long-Context Visual Language Models for Long VideosICLR
03

Real-time interaction

Virtual- and physical-world agents

  1. 2026π₀.₇: A Steerable Generalist Robotic Foundation Model with Emergent CapabilitiesTechnical report
  2. 2026EgoVLA: Learning Vision-Language-Action Models from Egocentric Human VideosIROS
  3. 2025π*₀.₆: A VLA That Learns From ExperienceTechnical report
  4. 2025Lumine: An Open Recipe for Building Generalist Agents in 3D Open Worlds*Technical report
04

Learn from experience

Experience distillation

  1. 2026Sample-Efficient Learning from Agent ExperienceTechnical report
  2. 2023Deductive Verification of Chain-of-Thought Reasoning*NeurIPS
Complete list on Google Scholar

Have a hard problem at the edge of models and the physical world?

Let’s chat.