About me#

I am Yufan Wei, a second-year Ph.D. student at University of California, San Diego, advised by Prof. Biwei Huang and working closely with Dr. Kun Zhou. My research focuses on causal intelligence and Embodied AI, with a particular interest in representation-level modeling across policy models and world models. Previously, I obtained my B.S. in Computer Science from the University of Minnesota.

Interests World Model & Representation Learning Embodied AI / VLA Causal AI

What’s New#

Research Experience#

Work Experience#

  • Aether AI — AI Research Intern, Summer 2026
  • Apple Inc. @ Beijing — Data Analyst Intern, 2023
Figure for CausalWM: Causal Chain-of-Thought Reasoning for Embodied World Model
CausalWM: Causal Chain-of-Thought Reasoning for Embodied World Model

Ziming Xu, Shuang Liang, Ruobing Han, Ziqiao Xi, Mingxing Rao, Kun Zhou, Zijun Zhang, Yuchen Yan, Yufan Wei, Junbo Huang, Yifei Shao, Fang Nan, Biwei Huang

A 16B embodied world model that makes physical reasoning explicit — predicting optical flow, then 3D pointmaps, then future video, with each step reused as context for the next. Top-1 on the TriWorldBench leaderboard.
391 words / 2 minutes
Figure for CD-LAM: Causally Debiased Latent Action Model for Embodied Action Conditioned World Models
CD-LAM: Causally Debiased Latent Action Model for Embodied Action Conditioned World Models

Yufan Wei, Kun Zhou, Lingjun Mao, Zijun Zhang, Ziming Xu, Ziqiao Xi, Shuang Liang, Ruobing Han, Yuchen Yan, Xinyue Wang, Fan Feng, Biwei Huang

A LAM-side causal debiasing method for continuous latent actions — removing action-irrelevant confounders before world-model training. −42%/−26% action-following error and >12× fewer training updates vs. DreamDojo.
295 words / 1 minutes
1