Learning Hand State Estimation for a Light Exoskeleton

TL;DR

A supervised approach is built using information from the muscular activity of the forearm and the motion of the exoskeleton to reconstruct the hand's opening degree and compliance level, which is promising for practical use in real rehabilitation.

Abstract

We propose a machine learning-based estimator of the hand state for rehabilitation purposes, using light exoskeletons. These devices are easy to use and useful for delivering domestic and frequent therapies. We build a supervised approach using information from the muscular activity of the forearm and the motion of the exoskeleton to reconstruct the hand's opening degree and compliance level. Such information can be used to evaluate the therapy progress and develop adaptive control behaviors. Our approach is validated with a real light exoskeleton. The experiments demonstrate good predictive performance of our approach when trained on data coming from a single user and tested on the same user, even across different sessions. This generalization capability makes our system promising for practical use in real rehabilitation.

Figures (8)

• Figure 1: Light exoskeletons are promising and powerful tools for effective rehabilitation therapies. The challenge addressed in this work is to provide this kind of device, having little sensory equipment, with advanced perception systems to online measure the patient's state and the therapy progress.
• Figure 2: Data collection setup: a user wears the exoskeleton and the emg sensor at the left hand and forearm, respectively, and an augmented reality headset (top left); detailed view on the exoskeleton and the emg sensor (right); view of the augmented reality scene (bottom left).
• Figure 3: Prediction performance of the different regression architectures ($y$-axis) trained with different feature sets (colors): $R^2$ (left, higher is better) and rmse (right, lower is better) averaged among users and reported for $y_{o}$ (top) and $y_{c}$ (bottom). Error bars denote $80\%$ confidence interval.
• Figure 4: A sequence of data acquired by a single user for the helping acquisition modality and the corresponding target prediction.
• Figure 5: A sequence of data acquired by a single user for the passive acquisition modality and the corresponding target prediction.