Locomotion Joint Angle and Moment Estimation With Soft Wearable Sensors for Personalized Exosuit Control

Area of Science:

• Robotics and Wearable Technology
• Biomedical Engineering
• Machine Learning Applications

Background:

• Soft sensors offer a promising alternative to traditional methods for human posture detection.
• Current research focuses on calibration, with limited practical application due to user diversity and complex scenarios like human-robot interaction.

Purpose of the Study:

• To develop and validate a flexible sensing system for accurate prediction of joint angles and moments.
• To enhance the generalization and accuracy of soft sensor models for practical applications.
• To demonstrate the system's utility in a flexible exosuit for personalized assistance and improved human-robot interaction.

Main Methods:

• Collected gait data from eight participants with varying walking patterns.
• Utilized Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) models for joint angle and moment estimation.
• Pre-trained models on an open-source dataset and fine-tuned with collected data, incorporating both static and dynamic features.

Main Results:

• Achieved a Mean Absolute Error (MAE) of 4.43° for knee joint angle estimation.
• Achieved a MAE of 0.12 Nm/kg for hip joint moment estimation.
• Demonstrated reduced muscle activation and improved user satisfaction (mobility, comfort) in a flexible exosuit prototype.

Conclusions:

• The developed flexible sensing system accurately predicts joint angles and moments, overcoming limitations of traditional methods.
• The system shows significant potential for practical applications, particularly in advancing human-robot interaction through personalized exosuit control.
• This research highlights the enhanced applicability of soft sensors in real-world scenarios, improving user mobility and comfort.