Research

Research Intern · Beijing, China

HACo: Haptic Active Compliance for Dexterous Manipulation

arXiv:2609.36596 ↗

• Built compliant dual-arm teleoperation and collected 500 multimodal demonstrations across five force-sensitive tasks.

• Conditioned a pretrained VLA policy on fingertip tactile and joint-torque feedback using a haptic encoder.

• Trained a flow-matching policy and deployed real-time action chunking for closed-loop control.

• Achieved 83% mean success across five real-robot tasks, versus 35% for the strongest evaluated baseline.

Research Intern · Beijing, China

Fine-Tuning Mixtral 8x7B on Biochemical Molecular Data

• Constructed and preprocessed a PubChem-based dataset for molecular property prediction tasks.

• Implemented KNN retrieval over SMILES representations to provide similar-molecule context for in-context learning.

• Fine-tuned Mixtral 8x7B and refined the evaluation protocol, improving test accuracy from 67% to 77%.

• Tech: Python, Git, VS Code.

Graduate Researcher · Baltimore, MD, USA

Event-Based Visuo-Tactile Sensing for Robot Imitation Learning

• Building a multimodal imitation-learning pipeline that fuses RGB, tactile, and event streams on a UR5 arm with a UMI gripper, targeting contact, slip, and fast-motion cues missed by RGB-only policies.

• Built a VIVE-based teleoperation setup and multi-rate data pipeline, aligning RGB (30 Hz), tactile/event (120 Hz), and 3D pose (75 Hz) into observation–action samples.

• Implemented and tested initial Diffusion Policy and Flow Matching baselines for the UR5 arm and UMI gripper.

Graduate Researcher · Baltimore, MD, USA

OpenTac: Shear-Force Simulation and Robot Experiments

• Developed shear-force sensing simulations in Isaac Sim for robot contact scenarios.

• Contributed to integrating scalable piezoresistive tactile skins onto a UMI handheld gripper for robot data collection.

• Designed and ran sensor characterization and robotic grasping experiments to evaluate normal- and shear-force feedback under varied contact conditions.

Final Year Project — Faculty Dean's Honors List · Macao SAR

Singular Spectrum Analysis for Biosignal Denoising and Feature Extraction

• Designed a two-level Singular Spectrum Analysis framework combining component correlation and waveform reconstruction for ECG and EEG signal analysis.

• Localized ECG PQRST features and removed EEG artifacts while preserving neural rhythms across delta, theta, alpha, and beta bands.

• Validated the method on synthetic and real recordings, recovering physiological patterns such as eyes-open vs. eyes-closed alpha rhythm differences.

• Tech: Python, NumPy, SciPy, Matplotlib, NeuroKit2.

Research Assistant · Hong Kong SAR

Neural Decoding for Implantable Brain–Machine Interfaces with ANN and SNN

• Preprocessed raw primate neural recordings into supervised datasets with movement coordinates, finger positions, and spike trains.

• Designed and benchmarked ANN and SNN decoders for neural decoding, comparing accuracy–efficiency trade-offs for implantable BMI scenarios.

• Tech: Python, MATLAB.

Research Assistant · Macao SAR

FPGA-Based Hardware Time-Triggered System for Biomedical Engineering

• Implemented FPGA-based signal conditioning and filtering pipelines in Quartus for biomedical data acquisition.

• Refined hardware architecture and I/O interfaces to support deterministic time-triggered behavior.

• Tech: Quartus, MATLAB.