Peizhou Cao
Portrait of Peizhou Cao
Beijing ↔ Shanghai 2026

Ph.D. researcher · spatial intelligence

Peizhou Cao.

Building models and worlds that help machines understand space.

I am a second-year Ph.D. student at Beihang University and Shanghai AI Laboratory, co-supervised by Jiangmiao Pang and Dahua Lin. My work sits at the intersection of spatial intelligence, simulation, and embodied AI.

Spatial intelligence Simulation Embodied AI
Explore selected work Start a conversation
Geometry-grounded perception Simulatable worlds Video spatial reasoning Generalist embodied agents

01 / Research landscape

A map of the questions I work on.

Instead of treating perception, data, and simulation as separate problems, I study how they connect. Explore a node to jump into the corresponding work.

Living research system

06 questions · 10 relationships · hover, focus, or tap a node

trains grounds populates hosts plans Core question Spatial intelligence Perceive → reason → act 01 Model · G²TAMGrounded trackingPixels + geometry 02 Benchmark · MMSISpatial reasoningVideo → world model 03 Data · SynthVerseSynthetic worldsScale + diversity 04 Environment · InternScenesSimulatable scenesLayouts → environments 05 System · GRUtopiaEmbodied agentsWorld → action
Research thread Spatial intelligence

The central question: how can a machine perceive geometry, reason about space, and act inside a world?

Model Data World System

02 / Selected work

Research, built to be used.

Recent work spanning models, benchmarks, datasets, and large-scale embodied environments.

G²TAM teaser with six panels comparing static and dynamic scene tracking from text, point, and box prompts
Paper teaser · prompted tracking
01 ICML · 2026

G²TAM: Geometry Grounded Track Anything Model

C. Zhu, P. Cao, J. Lin, W. Hu, Y. Ran, J. Pang, T. Wang, X. Liu

Paper ↗ Code coming soon
MMSI-Video-Bench teaser illustrating spatial construction, spatial reasoning, planning, and benchmark categories
Paper teaser · spatial reasoning
02 arXiv · 2026

MMSI-Video-Bench: A Holistic Benchmark for Video-Based Spatial Intelligence

J. Lin*, R. Xu*, S. Zhu, S. Yang, P. Cao, et al.

Paper ↗ Code ↗
SynthVerse teaser collage of animated, embodied, articulated, and manipulation sequences with colored point trajectories
Paper teaser · synthetic motion
03 SIGGRAPH · 2026

SynthVerse: A Large-Scale Diverse Synthetic Dataset for Point Tracking

W. Zhao, H. Xu, X. Miao, Q. Zhao, R. Zhang, K. Huang, N. Gao, P. Cao, et al.

Paper ↗ Code ↗
InternScenes teaser showing layout diversity, 1.9 million objects across 288 classes, layout generation, and embodied AI applications
Paper teaser · simulatable scenes
04 NeurIPS D&B · 2025

InternScenes: A Large-Scale Simulatable Indoor Scene Dataset with Realistic Layouts

P. Cao*, W. Zhong*, Y. Jin, L. Luo, W. Cai, J. Lin, et al. (*Equal Contribution)

Paper ↗ Code ↗
GRUtopia teaser showing more than 100,000 scenes across 89 categories and robot tasks in navigation, dialogue, and manipulation
Paper teaser · city-scale robotics
05 arXiv · 2024

GRUtopia: Dream General Robots in a City at Scale

H. Wang, J. Chen, W. Huang, Q. Ben, T. Wang, B. Mi, …, P. Cao, et al.

Paper ↗ Code ↗
06 IEEE T-CYB · 2026

A Progressive Semi-Distillation Model for Dual-Source Remote Sensing Image Classification

H. Zhu, P. Cao, L. Jiao, X. Li, B. Hou, X. Yi, W. Zhao, W. Ma

Paper ↗ Code ↗

03 / Experience

Where the work takes shape.

A path through electronic engineering, computer science, and large-scale AI research.

Education

Ph.D. in Computer Science and Technology

Beihang University · Beijing, China

04 / Contact

Hello

Research conversations, open-source collaborations, and ambitious problems are always welcome.

Email GitHub