GEGLU-Transformer for IMU-to-EMG Estimation with Few-Shot Adaptation

2026-04-28

AI Summary

This study addresses the challenge of robustly deploying surface electromyography (EMG) in real-world scenarios, where signal quality is compromised by electrode sensitivity, non-stationarity, and inter-subject variability. To overcome these limitations, the authors propose an adaptive learning framework that reconstructs continuous muscle activation signals from inertial measurement unit (IMU) data. The approach uniquely integrates Gaussian Error Gated Linear Units (GEGLU) with a Transformer encoder to map IMU inputs to EMG outputs, enhanced by an attention mechanism and a few-shot fine-tuning strategy. Under leave-one-subject-out cross-validation, the model achieves a correlation coefficient of r = 0.706 and R² = 0.474 without subject-specific adaptation. With fine-tuning on merely 0.5% of target-user data, performance improves markedly to r = 0.761 and R² = 0.559, demonstrating strong cross-user generalization and efficient personalization capability.

Abstract

Reliable estimation of neuromuscular activation is a key enabler for adaptive and personalized control in wearable robotics. However, surface electromyography (EMG) remains difficult to deploy robustly outside laboratory settings due to electrode sensitivity, signal non-stationarity, and strong subject dependence. In this work, we propose an adaptive IMU-to-EMG learning framework that reconstructs continuous muscle activation envelopes from wearable inertial measurements across heterogeneous movement conditions. The approach combines a Transformer encoder with Gaussian Error Gated Linear Units (GEGLU-Transformer) to enhance cross-subject generalization and enable rapid subject-specific personalization. Under a strict leave-one-subject-out (LOSO) protocol on a multi-condition lower-limb biomechanics dataset, the proposed architecture achieves r = 0.706 +/- 0.139 and R^2 = 0.474 +/- 0.208 without subject-specific adaptation. With only 0.5% adaptation data, performance increases to r = 0.761 +/- 0.030 and R^2 = 0.559 +/- 0.047, demonstrating rapid adaptation and early performance saturation. These results support attention-based architectures combined with lightweight adaptation as a practical and scalable alternative to direct EMG sensing for real-world wearable robotic applications.

Problem

• EMG estimation
• wearable robotics
• subject dependence
• signal non-stationarity
• neuromuscular activation

Innovation

• GEGLU-Transformer
• IMU-to-EMG estimation
• few-shot adaptation
• cross-subject generalization
• wearable robotics