Adaptive Compliance Shaping with Human Impedance Estimation

TL;DR

This paper introduces an adaptive exoskeleton control method that uses real-time human impedance estimation from surface EMG and stretch sensors to enhance performance and stability.

Contribution

It presents a novel compliance shaping controller that incorporates an online human stiffness estimation via a machine learning model trained on sensor data.

Findings

Improved bandwidth and amplification of the exoskeleton control.

Enhanced robustness and stability of the system.

Accurate real-time human stiffness prediction from sensor data.

Abstract

Human impedance parameters play an integral role in the dynamics of strength amplification exoskeletons. Many methods are used to estimate the stiffness of human muscles, but few are used to improve the performance of strength amplification controllers for these devices. We propose a compliance shaping amplification controller incorporating an accurate online human stiffness estimation from surface electromyography (sEMG) sensors and stretch sensors connected to the forearm and upper arm of the human. These sensor values along with exoskeleton position and velocity are used to train a random forest regression model that accurately predicts a person's stiffness despite varying movement, relaxation, and muscle co-contraction. Our model's accuracy is verified using experimental test data and the model is implemented into the compliance shaping controller. Ultimately we show that the online estimation of stiffness can improve the bandwidth and amplification of the controller while remaining robustly stable.