Master's student in Computer Science at Shenzhen University, based in Shenzhen, China.
My research focuses on world models, vision-language navigation, and reinforcement learning for embodied agents. I also build tools for running and inspecting experiments. Earlier, I worked on sleep staging, hyperspectral reconstruction, and OpenHarmony applications.
Academic Homepage | Publications | Blog | Google Scholar | Email
-
FP2: Foundation Pareto Flow Policy for Multi-Objective Reinforcement Learning
Accepted at NeurIPS 2026. -
MAPF-World: Action World Model for Multi-Agent Path Finding
IEEE Robotics and Automation Letters, 2026.
Paper | arXiv | Code -
Sparse Latent Action World Model for Vision-Language Navigation
ICANN, 2026.
Paper -
MCGA: Mixture of Codebooks Hyperspectral Reconstruction via Grayscale-Aware Attention
ICIC, 2026.
Paper | arXiv | Project Materials -
cVAN: A Novel Sleep Staging Method via Cross-View Alignment Network
IEEE Journal of Biomedical and Health Informatics, 2025 (online 2024).
Paper | Code
- Jev-MAPF: A workbench for testing LLM-based multi-agent path finding, with per-agent decisions, cooperative planning, and replay.
- dimOS Go2 / macOS: My experimental fork of dimOS, adding macOS and Go2 adaptations. Simulation has been tested; real-Go2 work is currently limited to sensing, mapping, and path preview, not closed-loop navigation.
I keep experiment records and implementation notes on my blog.
Agent OS Day 1: Moving a robot arm with natural-language instructions
A real-arm test connecting web instructions, live localization, planning, and execution. Includes the recording, failures, and remaining limitations. This was a hover test, not grasping.
Tools I use: Python, C++, TypeScript, PyTorch, Transformers, Linux, and Docker.