About

I am Mingyuan Jia (贾明远), an undergraduate in the Department of Automation at Tsinghua University. My research focuses on training design for adaptation in embodied systems.

I will join the joint PhD program between Shanghai Jiao Tong University and Shanghai Qi Zhi Institute, advised by Prof. Mengdi Xu.

My research focuses on how training design can strengthen adaptation in embodied systems. At the model level, I explore world modeling through multitask perception and prediction, and how data and supervision can encourage models to actively use context. At the system level, I study how task and interaction design can put agentic systems under the right environmental pressures to learn to adapt quickly. My long-term goal is continual learning after deployment and recursive self-improvement (RSI).

I have worked with Prof. Yong Li at FIBLAB and Prof. Yueqi Duan at the Tsinghua Spatial Intelligence & Vision Group. I also feel fortunate to learn from Prof. Gao Huang at LEAP Lab. I am currently an R&D intern at Shanghai Coco Matrix Intelligent Technology, working on robotic model pre-training.

News

  • Worldscape-MoE has been accepted to NeurIPS 2026.

  • Joined Shanghai Coco Matrix Intelligent Technology as an R&D intern, working on robotic model pre-training.

  • Started research with Prof. Yueqi Duan at the Tsinghua Spatial Intelligence & Vision Group.

  • Joined Manifold AI as an R&D intern to work on WorldScape.

  • AirScape appeared at ACM Multimedia 2025, BNI track (Oral).

Interests

World modeling

Modeling world states and their transitions as a general principle for prediction, planning, and interaction.

Robotics

Enabling robots to adapt at deployment, learn continually, and improve from physical experience.

Agentic systems

Orchestrating foundation models, tools, memory, and planning to turn individual capabilities into coherent systems.

Education & research

Sep 2023 — Present

Tsinghua University

Department of Automation · Undergraduate

Talent Program for Intelligent Robotics. Top 20%.

Research experience

Oct 2026 — Present

LEAP Lab

Tsinghua University · Advisor: Prof. Gao Huang

Learning and conducting research in the Learning And Perception (LEAP) Lab.

Feb — Aug 2026

Spatial Intelligence & Vision Group

Tsinghua University · Advisor: Prof. Yueqi Duan

Compositional 3D scene reconstruction and evaluation.

May 2025 — Apr 2026

FIBLAB

Tsinghua University · Advisor: Prof. Yong Li

Controllable world models and spatially consistent video generation.

Industry experience

Jul 2026 — Present

Shanghai Coco Matrix Intelligent Technology

R&D Intern · Mentor: Yuxiang Gao

Core member of the robotic model pre-training team.

Nov 2025 — Apr 2026

Manifold AI

R&D Intern · Mentor: Wei Wu

Key contributor to WorldScape, focusing on 3D spatial consistency training and memory-oriented KV cache optimization.

Publications & preprints

* Equal contribution. † Project lead.

Hobbies

Guitar

Photos coming soon.

0 / 0

Images are shared here to illustrate my personal interests. All rights belong to their respective owners. For copyright concerns or removal requests, please contact me.