research
My research focuses on leveraging a synergistic combination of AI and Hardware to enable user-independent, data-efficient recognition of diverse time-series signals, including Force, IMU, strain, and EMG from body joints and muscles.
The goal is to develop, energy-efficient wearable devices that enhance human-computer interaction, leveraging high-quality sensor datasets and adaptable AI models for user-agnostic, data-efficient performance
A simplified low‑channel EMG gesture interface
This research propose a self-supervised learning framework for wearable sensing that enables low-channel EMG devices to capture rich body kinematics traditionally requiring high-density sensor arrays.
Using a compact, wearable system, our approach achieves performance comparable to high-density EMG for human–computer interaction tasks such as sign language translation and gait force prediction. Nature Sensors, 2026 [PDF]
Task/User-Agnostic Wearable Human-Computer Interface
This research demonstrates advanced gestural interface capabilities enabled by high-quality datasets collected from newly developed wearable sensors. It features co-developed AI models that adapt to multiple users and tasks with limited training data.
The model leverages Transformer-based Few-shot learning for multi-task interaction, showcasing keyboard-less virtual typing and object/gesture recognition. Nature Electronics, 2023 [PDF]
AI-Augmented Wristband for Gesture Recognition
This research introduces a wrist-mounted single sensor that captures subtle skin deformations caused by finger movements, inspired by visible ligament shifts on the wrist.
The analog signal output enables low-latency processing, with an LSTM-based model predicting both finger identity and bending angle. The system was designed for adaptability across users using transfer learning and fine-tuning with minimal new data.
Nature Communications, 2021 [PDF]
Wearable embedded design for mutimodal sensing
This research demonstrates embedded circuit design for multimodal physiological sensing, integrating EMG, IMU, and temperature sensors on an ultra-thin PCB.
It further integrates a 915 MHz RF rectifier to enable battery-less operation. The system is implemented in a TI-RTOS environment, supporting compact, wearable, and real-time gesture and physiological signal acquisition. Adv.Func.Mat, 2022 [PDF]
