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.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to provide a machine - learning - based hand - state estimation method for light - weight exoskeletons for use in rehabilitation treatment. Specifically, the author aims to reconstruct the degree of hand opening and the level of compliance by using information from light - weight exoskeletons and forearm muscle activity (EMG sensors), thereby evaluating treatment progress and developing adaptive control behaviors. This information is crucial for objectively evaluating the rehabilitation effect and adjusting the treatment plan according to the real - time state of the patient. ### Specific description of the problem 1. **Degree of hand opening (Opening Degree)**: - Defined as a continuous variable \( y_o\in[0, \pi/2] \), where \( y_o = 0 \) represents "hand open" and \( y_o=\pi/2 \) represents "hand closed". - Used to estimate the range of motion (RoM) of the patient's joints to evaluate treatment progress. 2. **Hand compliance (Compliance Level)**: - Defined as a dimensionless variable \( y_c\in[-1, 1] \), where \( y_c = -1 \) represents "stiff hand", \( y_c = 1 \) represents "compliant hand", and \( y_c = 0 \) represents "neutral state". - Used to adapt the actions of the exoskeleton to match the current state of the patient. ### Solution To achieve the above goals, the author proposes a supervised learning method that combines data from exoskeleton motion data and EMG sensors: - **Input feature set**: A joint feature vector \( f=\begin{pmatrix} f_{exo}^T, f_{emg}^T \end{pmatrix}^T \) composed of exoskeleton motion data \( f_{exo} \) and EMG sensor data \( f_{emg} \). - **Model training**: Train using multiple regression models (such as linear regression, multi - layer perceptron, support vector regression, and LSTM) to estimate the hand state \( \hat{y}=m(f|D) \), where \( D \) is a training data set containing labeled sequences. ### Experimental verification The author experimentally verified the effectiveness of this method, including the influence of different feature sets on prediction performance and the generalization ability of the model in different users and different sessions. The results show that the LSTM model exhibits the best prediction performance after fusing exoskeleton and EMG data and can maintain good generalization ability in different sessions. ### Application prospects This method is expected to be applied in actual rehabilitation treatment, especially for the hand - function recovery of stroke patients. Through portable light - weight exoskeletons and EMG sensors, patients can perform frequent rehabilitation training at home without complex external devices. In addition, this system can adaptively adjust the treatment plan according to the real - time state of the patient, improving the rehabilitation effect. ### Main contributions - Proposed a machine - learning method that combines exoskeletons and EMG sensors to estimate hand states. - Verified the effectiveness and generalization ability of this method in different users and different sessions. - Potentially improved the effectiveness and convenience of rehabilitation treatment, especially in the home environment. ### Future work Future research will focus on evaluating the usability and acceptance of this tool among stroke patients and further exploring its potential in actual clinical applications.

Learning Hand State Estimation for a Light Exoskeleton

EMG-Driven Machine Learning Control of a Soft Glove for Grasping Assistance and Rehabilitation

Control of Newly-Designed Wearable Robotic Hand Exoskeleton Based on Surface Electromyographic Signals

Machine learning based Development and Evaluation of a Hand Exoskeleton Rehabilitation Robot for Patients with Partial Paralysis

Soft Exoskeleton Glove for Hand Assistance Based on Human-machine Interaction and Machine Learning

Hand Exoskeleton Design and Human–Machine Interaction Strategies for Rehabilitation

Real time motion intention recognition method with limited number of surface electromyography sensors for A 7-DOF hand/wrist rehabilitation exoskeleton

Functional Variable Selection for EMG-based Control of a Robotic Hand Prosthetic

AI-Powered Camera and Sensors for the Rehabilitation Hand Exoskeleton

Real-Time Control of an Exoskeleton Hand Robot with Myoelectric Pattern Recognition

Development and evaluation of a hand exoskeleton for finger rehabilitation*

Design of a Modular Hand Exoskeleton for Rehabilitation and Training

Deep-Learning Control of Lower-Limb Exoskeletons via simplified Therapist Input

Hand exoskeleton for rehabilitation therapies with integrated optical force sensor

A novel application of a surface ElectroMyoGraphy-based control strategy for a hand exoskeleton system: A single-case study

Vision-Based Pose Estimation for Robot-Mediated Hand Telerehabilitation

Detection of movement onset using EMG signals for upper-limb exoskeletons in reaching tasks

Myoelectric Control of a Soft Hand Exoskeleton Using Kinematic Synergies

Design, Development, and Control of a Hand/Wrist Exoskeleton for Rehabilitation and Training

EMG Pattern Classification to Control a Hand Orthosis for Functional Grasp Assistance after Stroke

Adaptive Semi-Supervised Intent Inferral to Control a Powered Hand Orthosis for Stroke