About Me

I am an incoming Ph.D. student at Harbin Institute of Technology (HIT). I received my Master of Science (M.Sc.) degree in Railway Engineering (ITS & Autonomous Systems track) from Technical University of Munich Asia (TUM Asia), and my Bachelor of Engineering (B.Eng.) degree in Civil Engineering with Transport from the Joint Program of the University of Leeds and Southwest Jiaotong University (SWJTU-Leeds).

Currently, I also work as an AI Agent Developer at Sihe Information Technology Co., Ltd., focusing on Long-horizon Task Agents, model distillation, and Agent Skill / SOP design. Previously, I worked on L4 autonomous driving and intelligent transportation systems R&D at the Mobility Innovation Campus (MIC) with IABG mbH in Munich, Germany, and on traffic operations simulation at ST Engineering in Singapore.

My research lies at the intersection of Artificial Intelligence, Autonomous Driving, Driving World Models, and Intelligent Transportation Systems (ITS). I aim to bridge physical-world mechanisms and AI algorithm-based decision-making to build responsive, closed-loop simulators and intelligent vehicle-infrastructure cooperative control systems.

🌟 Featured Research

From Foresight to Reactive Traffic Worlds: A Survey of Driving World Models for Closed-Loop Simulation

Abstract: A comprehensive survey formalizing the Simulation-Closure Spectrum (Replay-Constrained → Ego-Responsive → Traffic-Reactive) and establishing a 5-layer claim-specific validity stack. Accompanied by full preprint, LaTeX package, and editable figures.

Survey Details SSRN GitHub PDF

Toward Continually Growing World Models: An OaK-Inspired Architecture for Learned Abstraction and Planning

Abstract: Formulates Structural Continual World Modeling (SCWM), establishing a dual-axis lifelong adaptation framework combining parametric weight updates with structural lifecycle growth (Feature Construction → SubTask → Option → Model → Planning) based on Richard Sutton's OaK architecture.

Paper Details Zenodo DOI GitHub PDF

🔬 Research Interests

Continual World Models & Reinforcement Learning: Structural Continual World Modeling (SCWM), lifelong model learning, OaK architecture (Feature $\to$ SubTask $\to$ Option $\to$ Model $\to$ Planning), temporal abstraction, and plasticity retention.
Driving World Models: Generative world models for closed-loop traffic simulation, ego-response fidelity, and traffic reactivity under counterfactual / off-log actions.
AI Agent Engineering & LLMs: Long-horizon task agents, agent skill & SOP design, tool-use workflows, tokenization/BPE mechanisms, and model distillation (Teacher–Student).
Autonomous Driving & V2X: Vehicle-to-Infrastructure (V2I/V2X) cooperative perception, BEV scene representations, and real-world connected vehicle control.
Reinforcement Learning in Transportation: Deep Q-Networks (DQN) and adaptive traffic signal control for mixed autonomous and human-driven traffic.

📢 Recent News

[Aug 2026] Released comprehensive research survey on Driving World Models: “From Foresight to Reactive Traffic Worlds: A Survey of Driving World Models for Closed-Loop Simulation” with full open-source resources (Survey Details

SSRN

GitHub

PDF).

[Aug 2026] Released research paper on Continual World Models: “Toward Continually Growing World Models: An OaK-Inspired Architecture for Learned Abstraction and Planning” (Zenodo DOI GitHub).

[Jun 2026] Completed research paper: “Conditional Effects of Connected-Vehicle Requests in Reinforcement-Learning Traffic Signal Control”, conducting 2,730 multi-seed runs and 2D crossed-bootstrap evaluation on a site-derived Munich testbed (Paper Details