An EMG-Based GRU Model for Estimating Foot Pressure to Support Active Ankle Orthosis Development

Area of Science:

• Biomechanics
• Rehabilitation Engineering
• Machine Learning in Healthcare

Background:

• Aging populations face mobility challenges, particularly ankle dysfunction like foot drop.
• Current active ankle-foot orthoses (AAFO) lack adaptability due to rule-based controls.
• Need for personalized AAFO control synchronized with individual gait dynamics.

Purpose of the Study:

• To develop a neuromuscular activation-driven approach for predicting plantar pressure distributions.
• To enhance real-time control of AAFO using deep learning and surface electromyography (EMG).
• To improve gait adaptability and support for individuals with ankle impairments.

Main Methods:

• Collected surface EMG from four ankle muscles and plantar pressure data using a force-sensitive resistor (FSR) system.
• Utilized a Gated Recurrent Unit (GRU) deep learning model with Root Mean Square (RMS) features for prediction.
• Employed a sliding window method for data preprocessing and segmentation.

Main Results:

• The GRU model accurately predicted plantar pressure distributions in real-time.
• Successfully inferred critical gait events like heel strike, mid-stance, and toe-off across subjects.
• Demonstrated strong EMG signal compatibility and identified individual electromechanical delay (EMD) variations.

Conclusions:

• The proposed predictive framework offers a scalable and interpretable method for AAFO control.
• Enables synchronization of AAFO assistance with user-specific gait dynamics.
• Paves the way for more adaptive and effective mobility assistance devices.