2024 - 2027
M.Eng. in Mechanical Engineering
Huazhong University of Science and Technology, Wuhan, China
Expected graduation: June 2027
Thesis: High-speed, high-precision vision-based tactile sensing for robotic dexterous hands.
Hi there!
I am an M.Eng. student in Mechanical Engineering at Huazhong University of Science and Technology, advised by Prof. Hua Yang, with expected graduation in June 2027. My research focuses on vision-based tactile sensing and visuotactile learning for dexterous, contact-rich robotic manipulation.
My work spans tactile sensor design, 3D geometry reconstruction, three-axis force estimation, and real-time FPGA deployment. I also work on visuotactile fusion, imitation learning, and simulation-to-real transfer for multi-scale fine manipulation, connecting contact perception with robot action learning.
Education
2020 - 2024
B.Eng. in Robotics Engineering
Huazhong University of Science and Technology, Wuhan, China
Research Interests
- Tactile geometry and force perception
- Visuotactile multimodal fusion
- Imitation learning for fine manipulation
- Simulation-to-real transfer
- Real-time edge perception
Technical Practice
- Sensing
- Optical and mechanical sensor design, elastomer fabrication, multispectral imaging, geometric and force calibration
- Robot Learning
- Visuotactile fusion, imitation learning, simulation-to-real transfer, long-horizon fine manipulation
- Perception
- Photometric stereo, surface reconstruction, force estimation, inverse rendering, synthetic data
- Hardware
- FPGA acceleration, embedded deployment, camera interfaces, PCB prototyping, Vivado, Vitis HLS, Vitis, Zynq MPSoC, model quantization
- Programming
- Python, C/C++, MATLAB, Verilog/SystemVerilog, PyTorch, OpenCV
News
- Released the FasTac project page and arXiv preprint.
- Launched this academic website.
- Published the SMF-PSNN project page and paper in IEEE/ASME Transactions on Mechatronics.
- Published the GelSplitter3D project page and paper at IEEE/RSJ IROS.
Publications
Selected peer-reviewed work. My name is shown in bold.
FasTac: A Curved Multispectral Vision-Based Tactile Sensor for High-Speed High-Precision 3D Shape and Force Perception
A compact RGB-NIR tactile fingertip combining boundary-prior 3D reconstruction, position-aware three-axis force estimation, and a low-latency FPGA perception pipeline.
arXiv:2607.28416; submitted to IEEE/ASME Transactions on Mechatronics, 2026
High-Precision RGB-NIR Vision-Tactile Sensor via Neural Inverse Rendering for Robotic Perception
High-precision tactile reconstruction, self-calibration, and synthetic tactile image generation with RGB-NIR sensing.
IEEE/ASME Transactions on Mechatronics, 2025
3D Vision-Tactile Reconstruction from Infrared and Visible Images for Robotic Fine-Grained Tactile Perception
A curved RGB-NIR tactile fingertip with neural photometric stereo and geometry-prior normal integration.
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025
Projects
Selected research and engineering work across tactile sensing, robot learning, and deployment.
FasTac: High-Speed Tactile Geometry and Force Perception
Led the development of a compact curved RGB-NIR tactile sensing system spanning sensor design, synchronized multispectral imaging, boundary-prior 3D reconstruction, HyperForce three-axis force estimation, and FPGA edge deployment.
Implemented hardware-oriented pipelines for surface-normal inference, boundary-prior depth reconstruction, and normal-force estimation on Xilinx Zynq UltraScale+ platforms.
Depth mean absolute error is 0.0415 mm. The FPGA image-to-normal-force pipeline achieves 1.09 ms/frame latency and 8.41 mJ/frame energy consumption. Computation rates reach 1022 Hz for image-to-depth and 918 Hz for image-to-normal-force.
Visuotactile Imitation Learning for Multi-Scale Fine Manipulation
Worked on visuotactile imitation learning for multi-scale, contact-rich bimanual manipulation. The project combines tactile-informed simulation data generation, memory-efficient multimodal sparse attention, and multi-expert action learning to address limited demonstrations, small-target perception, and coarse-to-fine action coordination.
Reduces memory usage for multi-scale fusion from 24 GB to 7.4 GB. The complete method achieves a 100% success rate in the evaluated 800-step simulated long-horizon embroidery task.
Visuotactile Sensor Development and Dexterous-Hand Integration
Assisted with mechanical design for two-finger grippers, three-finger grippers, and five-finger dexterous hands. Responsible for end-effector vision-based tactile sensor design, 3D reconstruction and force estimation algorithms, calibration, testing, and system integration.
Delivered vision-based tactile sensor prototypes and accompanying technical documentation for two-finger, three-finger, and five-finger platforms.
Industry Impact. My work on vision-based tactile sensor adaptation, calibration, and perception algorithms supported Zhejiang Shiyue Technology (LumiBot) in developing the Lumi-Tac vision-based tactile sensor and Lumi-Dex dexterous-hand product line.
Neural Inverse Rendering and Synthetic Tactile Data
Developed SMF-PSNN for near-field multi-source illumination self-calibration, surface-normal and depth reconstruction, and RGB-NIR tactile image synthesis. The calibrated rendering model generates transferable tactile observations for pose estimation, fingerprint recognition, and slip detection.
Neural inverse rendering reaches 30.22 dB PSNR, improving by 6.33 dB over the single-light-source baseline.
Build
Hardware systems I helped build for vision-based tactile manipulation.
Two-Finger Gripper
Vision-based tactile end-effector for fine manipulation and contact reconstruction. Project page ↗
Three-Finger Gripper
Multifinger tactile platform for object interaction and manipulation demos. Project page ↗
Five-Finger Dexterous Hand
Dexterous hand prototype with integrated tactile modules. Project page ↗