M.S. Student in Control Science & Engineering @ Harbin Institute of Technology, Shenzhen
Focusing on Generative AI for Robotics, Reinforcement Learning, and Efficient Visuomotor Policies.
I am currently a Master's student at HIT-Shenzhen, advised by [Insert Supervisor Name if applicable]. My research interest lies at the intersection of Robotics and Generative AI.
I am particularly enthusiastic about:
- 🤖 Embodied AI: Building robust policies that work in the wild (Sim-to-Real).
- 🎨 Generative Models: Applying Diffusion Models & Flow Matching to robot control.
- ⚡ Efficient Inference: Making SOTA models run on edge devices (Jetson/NUC).
| Project | Description | Tech Stack |
|---|---|---|
| [PocketDP3] |
Robust Pocket-Scale 3D Visuomotor Policy • Proposed Diffusion Mixer (DiM) decoder with <1% params of SOTA. • Achieved 2-step inference for real-time control on edge devices. |
PyTorch Diffusion RoboTwin |
| [Efficient-Flow] |
Lightweight MLP-based Flow Matching • Designed a pure MLP trajectory predictor using Optimal Transport. • 10x faster inference compared to traditional diffusion policies. |
Flow Matching Optimal Transport |
| Project | Description | Tech Stack |
|---|---|---|
| [Agile-Delay] (ICRA 2026 Accepted) |
Asynchronous End-to-End Learning • Solved high-latency perception issues with Temporal Encoding Module (TEM). • 100Hz control on NUC; Zero-shot Sim-to-Real in dense forests. |
RL Sim-to-Real NUC |
| [STORM] (IROS 2025 Accepted) |
Spatial-Temporal Iterative Optimization for UAV • B-spline based optimization with Guidance Gradient. • Guaranteed geometric safety in cluttered environments. |
C++ Optimization ROS |
Python |
C++ |
PyTorch |
OpenCV |
ROS |
Isaac Gym |
Linux |
Docker |
Git |
CMake |
Bash |
LaTeX |
