AI Research · HKUST(GZ)

Tianshuo Xu.

Modeling a world
that keeps moving.

I build generative models that maintain world state and follow physical dynamics, with applications in autonomous driving and embodied intelligence.

I’m a Ph.D. student in Artificial Intelligence at HKUST(GZ), advised by Prof. Yingcong Chen. Previously, I received my bachelor’s and master’s degrees from Xiamen University, advised by Prof. Rongrong Ji.

World modelsAutonomous drivingEmbodied intelligenceVideo generation

01 / Research

Highlighted research.

Persistent memory.
Physically consistent motion.

01NeurIPS 2026

ReMind

Teaching video generators to remember.

A causal video generator with dynamic memory: objects and events keep evolving through camera motion, occlusion, and changes in visibility.

The view disappears.
The world keeps moving.

Pouring, rising liquid levels, growing granular piles, and spreading batter. Eight complete examples of state evolving while hidden.

Explore all examples ↗
02NeurIPS 2026

Motion Forcing

Controllable motion, consistent physics.

A decoupled framework for robust video generation that separates camera and object control, with applications in driving, rigid-body dynamics, and robotic grasping.

Same scene. Different controls.
Motion follows your intent.

Compare left cut-in + braking with right cut-in. Control inputs stay visible above synchronized results.

Explore all controls ↗

02 / Experience

Research in practice.

Research internships
across industry.

Jan – Sep 2026

Sunnyvale, California

Applied Intuition

Research Intern · Research Team

Developed ReMind to elicit dynamic memory in video diffusion transformers and enable coherent out-of-sight state evolution.

World models

Jan 2024 – Nov 2025

Shenzhen, China

Huawei

Research Intern · Noah’s Ark Lab

Developed Occ-LLM for occupancy forecasting, planning, and scene question answering. Built generative world models for autonomous driving.

Autonomous driving

Nov 2021 – Jun 2023

Beijing, China

MEGVII

Research Intern · Foundation Model Group

Optimized large diffusion models for image synthesis and explored fine-tuning and prompt-tuning for perceptual image super-resolution.

Generative models

03 / Publications

Selected publications.

Full publication list ↗

My name is highlighted. * denotes equal contribution.

8 publications shown

04 / Contact

Let’s talk research.

World models, generative video, and physical intelligence.