Project/ X-Garment — Hongyu Yue's works

About

X-GarmentA Smart Garment for Muscle Compensation Recognition

Overview
Compensation is the unconscious use of other muscle groups to assist or substitute for fatigued or impaired muscles during a movement. Although the motion may appear correct, the joint is loaded improperly, reducing the effectiveness of rehabilitation and increasing the risk of secondary injuries. These compensations are hard to detect in traditional training and even more difficult to correct once they become habitual.
This led me to ask: Could a lightweight garment, similar to a regular sports top, continuously detect muscle compensatory patterns in real time? This would facilitate effective home-based rehabilitation training.

ContributionIndividual (Capstone Project)

DurationSep 2025 – Present (4 months)

InstructorQi Wang (Head of Center for Digital Innovation, Tongji University)

KeywordsSmart Textiles, Rehabilitation, Deep Learning, Wearable Device

StatusOngoing

Please view in a full-screen browser window for the best experience.

Problem & Project DirectionCompensation is the unconscious use of other muscles to assist or substitute for a fatigued or impaired muscle during a movement. Although the motion may appear correct, the joint is loaded improperly, reducing the effectiveness of rehabilitation and increasing the risk of secondary injuries. These compensations are hard to detect in traditional training and even more difficult to correct once they become habitual.

This led me to ask: Could a lightweight garment, similar to a regular sports top, continuously detect muscle compensatory patterns in real time? This would facilitate effective home-based rehabilitation training.

Gaps in Existing Research  Current compensation monitoring relies heavily on optical motion capture, cameras, or IMU sensors. These approaches are costly, prone to occlusion or drift, and challenging to deploy in home-based rehabilitation scenarios. A fully textile-based monitoring garment offers far greater potential for daily use. However, accurately recognizing subtle compensatory patterns in complex multi-DoF movements using soft textile structures remains a significant technical challenge.

ApproachMy approach embeds textile stretch sensors within the garment, enabling them to deform with body movement and produce continuous strain signals. Combined with deep learning models, the system identifies subtle compensatory patterns and predicts multi-DoF joint angles. This architecture offers clear advantages in wearability, comfort, and data richness.

System OverviewThe system uses a one-time optical motion capture session solely for initial calibration to establish ground-truth joint angles. During actual training, textile sensors deform passively with movement, converting strain into electrical signals. 

The analysis framework contains two parallel processing paths: 
1. Compensation Recognition Path: A BiLSTM with an attention mechanism processes the 14-channel sensor sequences to classify movement patterns and determine whether the current repetition is performed correctly. 
2. Angle Regression Path: An Attention-LSTM maps the same time window into Euler angles, quantifying how much each joint actually rotates.

Together, these two paths drive a real-time interface that displays both whether compensation occurs and the precise joint angles.

Garment DesignTo obtain stable and reliable sensing data, the textile sensors and conductive fibers must be securely integrated into the fabric while remaining free from unwanted stretching or mechanical interference. This imposes strict requirements on garment structural design. We therefore performed multiple iterations and optimizations across garment patterning and routing, sensor placement and fixation, and integration of the driver module.

Ongoing WorkWhen deploying textile-based sensing systems for long-term rehabilitation, several challenges emerged: repeated donning and doffing alter the initial pre-stretch state of the sensors; extended movement introduces relative displacement between garment and skin; and the generic sensor layout still offers room for improved information efficiency. Accordingly, my next phase focuses on systematic refinements across sensor structure, sensor layout, and garment architecture, with the goal of improving signal reliability during prolonged use. 

All Works

Hongyu Yue 岳洪宇