Qize Yu 于骐泽

Embodied AI · Foundation Models · LLMs · CV

I am currently a graduate student at Peking University. My research vision is to solve embodied intelligence through vision and language. For System 1, I believe visual perception is the foundation of physical intelligence, “If we want machines to think, we need to teach them to see.” For System 2, I believe embodied intelligence can ultimately be solved by unlocking the full potential of LLM intelligence, bringing reasoning and generalization into the physical world.

Before embodied AI, I explored quantitative trading, and earlier, number theory, optimization theory, and low-level computer vision. These experiences shaped my passion for using mathematical theories and rigorous formulations to model complex problems.

My goal is to create the GPT-4 moment for Physical AI.

News

Publications

I'm open to collaborations on related projects.

Education

Experience

XpengApr 2026 – Present
Physical AI Talent Program Intern @ GIC Foundation Model Team

Conducted large-scale VLA and VLM pre-training, focusing on spatial intelligence and Omni-model fusion training with full-duplex speech and Action Expert, and served as project lead for X-locate, a Location Foundation Model (VLA 3.0 & IRON subproject), produced papers: GroundingPI and GroundAnything.

Knowin AINov 2025 – Mar 2026
Research Intern & Project Leader @ Embodied AI Algorithm Team

Proposed and led the development of a novel VLA model framework.

Industrial and Commercial Bank of ChinaNov 2024 – Mar 2026
Research Intern & Project Leader @ Head Office

Based at the Head Office of ICBC, led the development of two LLM-based systems, both of which were incorporated into the bank’s “ICBC Zhiyong” LLM ecosystem.

J.P. MorganApr 2024 – Aug 2024
Research Intern @ Algo Trading Team

Conducted research on High-Frequency Trading (HFT), focusing on the use of Alternative Data for factor discovery and quantitative trading in cryptocurrency markets.

MicrosoftOct 2023 – Feb 2024
Research Intern @ MSRA & STCA BingAds

Focused on growth analytics for Bing and Edge. Developed ads optimization and LTV prediction models utilizing hundreds of millions of user data points, enhancing ad relevance and long-term value. Integrated models into the production pipeline, establishing the first AI-powered growth analytics system.

Honors

Second Place, RoboDojo Safe World Models Challenge · ECCV 2026Certificate Sep 2026
Merit Student & Scholarship of Peking UniversityDec 2025
Outstanding GraduatesJun 2024
First Prize, Chinese Mathematical Olympiad (Provincial)Nov 2018

Photographer

Outside of research, I find inspiration in travel and landscape photography, always chasing extraordinary landscapes and fleeting moments of light. I've had the privilege of serving as an official photographer for Peking University, China's Ministry of Foreign Affairs, and National Geographic, with my work reaching over 100 million views worldwide and receiving over 1 million likes on personal platforms.

Mathematician

Mathematics is another lifelong passion of mine. Here, I've shared some of my handwritten proofs and mathematical explorations. Beyond these glimpses, a much more ambitious and systematic body of work, which I've been developing since late 2022, is on its way. Stay tuned!

Quant Researcher

I'm also deeply fascinated by quantitative trading, particularly in cryptocurrency markets, where I enjoy exploring market dynamics, quantitative strategies, and the mathematics behind trading. Here, I share some of my trading recipes, research notes, and personal insights. I'm always happy to exchange ideas with fellow enthusiasts, especially about this book!