I am a third-year undergraduate student at Zhejiang University (Chu Kochen Honors College), majoring in Computer Science and Technology (expected B.Eng. 2027). I am currently a research intern at Tsinghua University, where I work with Prof. Luping Shi and Prof. Rong Zhao on model-based reinforcement learning and world models. Previously, I worked on continuous inference for large language models at Zhejiang University.

I am obsessed with one question: "What is the realizable path toward physical intelligence?"

My answer is to build Physically Consistent and 3D-Aware World Models as a foundation for physical intelligence.

Through my previous work on LM interpretability and continuous inference, I came to realize that today’s models achieve abstraction over text and 2D visual tokens, but these foundations are not enough for true spatial intelligence.

I believe the next generation of physical AI will require stronger 3D inductive biases, better spatial representations, and ultimately the ability to predict how the world changes after actions.

My long-term goal is to turn AI into a physically grounded assistant beyond the screen. An overview of my research path:

News

• [2025-2026] One paper submitted (under review with strong feedback).
• [2025. 10] Started research at Tsinghua University on physical world models and RL exploration.
• [2025. 10] Awarded First-Class Scholarship, Zhejiang University.
• [2024. 10] Awarded National Scholarship (Top 0.5%) & First-Class Scholarship, Zhejiang University.

Publications

• Mind Dreamer: Untethering Imagination via Active Counterfactual Reasoning on Latent Manifolds [PDF]
  Shaojun Xu, Xiaoling Zhou, Yihan Lin, Yapeng Meng, Xinglong Ji, Luping Shi, Rong Zhao

Competitions

• 2026 NVIDIA DGX Spark Hackathon
  [Project] ClawKeeper: Security Management & Work Monitoring System Based on NemoClaw
  Xiaoling Zhou, Lei Hong, Yifeng Guan, Chengbo Sun, Haoyuan Chen

Details

Field: Agent Operations

Built an operator-facing runtime stack that bridges agent capability and operability via security controls, runtime watchdog monitoring, and structured operator feedback.

What we built: Security Control Plane (risk-aware tool interception + auditing) · Runtime Watchdog (timeout/stall/loop & repeated-output detection + health probes) · Operator Feedback (structured events for notifications/dashboards).

Project repo

Awards