M.S. in Control Science & Engineering @ Harbin Institute of Technology, Shenzhen
Generative AI for Robotics | Diffusion Policy | Flow Matching | Sim-to-Real
I am a Master's student at HIT-Shenzhen. My research lies at the intersection of Generative AI and Robotics.
- 🤖 Embodied Manipulation: Diffusion Policy, Flow Matching, VLA post-training (DPO), Sim-to-Real
- 🚁 Aerial Robotics: End-to-end RL, asynchronous perception-control, trajectory optimization
- ⚡ Efficient Deployment: 2-step diffusion inference, TensorRT acceleration, edge devices (Jetson/NUC)
| Venue | Paper | Role |
|---|---|---|
| CoRL 2026 (Under Review) | LDP: Shaping Latent Spaces for Diffusion-Based Robotic Manipulation | 1st Author |
| NeurIPS 2026 (Under Review) | Hydra-DP3: Frequency-Aware 3D Diffusion Policies | 1st Author |
| NeurIPS 2026 (Under Review) | Information Filtering via Variational Regularization | Co-author |
| ECCV 2026 (Under Review) | EfficientFlow: Lightweight MLP-based Flow Matching for Trajectory Prediction | 1st Author |
| ECCV 2026 (Under Review) | LAP: Latent Diffusion Planner for Autonomous Driving | Co-author |
| ICRA 2026 ✅ | Agile in the Face of Delay: Asynchronous End-to-End Learning for UAV | 1st Author |
| IROS 2025 ✅ | STORM: Spatial-Temporal Iterative Optimization for UAV | 1st Author |
| IEEE RA-L ✅ | Decentralized End-to-End Multi-AAV Pursuit via DRL | 2nd Author |
| Project | Highlights | Tech |
|---|---|---|
| LDP | VAE-based latent space shaping + Diffusion Forcing + Staircase Sampling; SOTA on RoboTwin/Adroit | DiT Flow Matching CVAE |
| Hydra-DP3 | Diffusion Mixer (<1% params of SOTA); 2-step inference without distillation; frequency-domain analysis | Diffusion DCT MLP-Mixer |
| Project | Highlights | Tech |
|---|---|---|
| Agile-Delay (ICRA'26) | Asynchronous RL with Temporal Encoding; 100Hz control; Zero-shot Sim-to-Real in forests | RL Sim-to-Real NUC |
| STORM (IROS'25) | B-spline spatial-temporal optimization; Guidance Gradient; geometric safety guarantee | C++ Optimization ROS |
| Multi-AAV Pursuit (RA-L) | Decentralized end-to-end MARL; Predictive Spatial-Temporal Observation; zero-shot cross-formation generalization | MARL Sim-to-Real Isaac Gym |
| Project | Highlights | Tech |
|---|---|---|
| EfficientFlow | Pure MLP + Optimal Transport; 10x faster than diffusion baselines | Flow Matching MLP |
| LAP | Latent diffusion planner; 10x speedup; fine-grained feature distillation | VAE Diffusion BEV |
- 🥇 National 1st Prize — National Intelligent Car Competition (ROS Outdoor, Team Lead)
- 🥈 National 2nd Prize — RoboMaster Super Battle (Sentry Perception)
- 🥈 National 2nd Prize — RoboMaster UAV Intelligent Perception (NMPC Control)
- 🥈 National 2nd Prize — National Intelligent Car Competition (Baidu Smart Traffic, Team Lead)
Python |
C++ |
PyTorch |
OpenCV |
ROS |
Isaac Sim |
Linux |
Docker |
Git |
CMake |
Bash |
LaTeX |
