Beijing → Singapore · 2026
I build intelligent systems that connect knowledge, spatial worlds, and physical action.
M.Sc. student at NTU · B.Eng. in Automation at BIT
Selected research
Five projects trace the question I keep returning to: how can intelligent systems turn representations into reliable action?
2024—present · Selected workDeveloping an embodied-AI framework that decomposes language instructions into navigation, search, approach, grasp, and placement. The project studies VLN/VLM planning, VLA or diffusion-policy control, navigation–manipulation hand-offs, recovery, and safety in simulation with limited real-robot validation.
M.Sc. dissertationCo-developed an LLM-agent framework that learns from evolving literature and generates research hypotheses. I built API orchestration, figure-digitisation skills, and local inference workflows for large-scale ablations.
Paper result · 72% of topics yielded validated hypotheses · 91% lower token cost
Read the paper ↗Built an end-to-end pipeline for simulator-ready environments: generated scenes and assets, estimated support surfaces, checked collisions, and composed 3D Gaussian Splatting worlds for a downstream UAV RL loop.
Undergraduate thesisBuilt a geospatial pipeline that combines illumination and slope constraints to rank candidate landing zones, turning Martian DEM evidence into an inspectable selection workflow.
View code ↗Designed a two-stage pipeline for clustering pulse streams and classifying emitter types. It reached 89% accuracy, a 12-point absolute improvement over the CNN baseline on the same dataset.
Research projectPublications
Tao, J., Wang, Y., Liu, X., Yang, M. · ICML 2026 AI4Research Workshop · arXiv:2604.12243
Wang, Y., Kang, X., Zhang, S. · GRMSD 2024 · First author
Chen, W., Zhou, Y., Wang, Y., Zhang, X., Wang, J. · ArtInHCI 2024
About
I’m an M.Sc. student in Computer Control and Automation at Nanyang Technological University, working on hierarchical mobile manipulation with VLN and VLA models. I received my B.Eng. in Automation from Beijing Institute of Technology.
My work moves between embodied AI, scientific agents, generated 3D environments, and control. I care about the point where a convincing model becomes a system that can make useful decisions under real constraints.
Questions I’m exploring
Research agents, evolving literature, memory, and the gap between retrieval and reasoning.
3D representations, simulation, spatial constraints, and environments for autonomous systems.
Language-conditioned task decomposition, VLN–VLA hand-offs, recovery, and safety under real sensor and physics constraints.
Contact & links
If our interests overlap, I’d be glad to compare notes, explore a research idea, or contribute to an ambitious prototype.
Open to research collaboration and lab opportunities