This work proposes a methodology for controlling the opening and closing movements of virtual hands based on the decoding of electroencephalography (EEG) signals associated with imagined left and right-hand grasping. The signals are processed using a hybrid approach that combines a feature extractor to enhance the signal-to-noise ratio with a scattering wavelet transform. A logistic regression model is employed to classify motor imagery events (left vs. right-hand). Using data from 20 volunteers obtained from an open-access database, the algorithm achieves an average accuracy of 99±2% when considering the delta, theta, and alpha/mu brain rhythms; 84±11% when using only alpha/mu rhythms; and 82±11% for signals containing information from the delta, theta, alpha/mu, and beta rhythms. To transform decoded intentions into robotic commands, a trajectory planner and control strategy were implemented to ensure movement stability and minimize tracking error during the hand’s motion in a small-diameter grasp. Experimental validation was performed using a hand simulated in Simulink/Matlab. The results demonstrated that the computed torque controller outperformed the MP-ADRC, with errors for the proximal phalanx of <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="40430_2025_5711_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="89" /> </InlineMediaObject> <EquationSource Format="TEX">$1.5332\cdot 10^{-3}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>1.5332</mn> <mo>·</mo> <msup> <mn>10</mn> <mrow> <mo>-</mo> <mn>3</mn> </mrow> </msup> </mrow> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="40430_2025_5711_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="89" /> </InlineMediaObject> <EquationSource Format="TEX">$4.4498\cdot 10^{-3}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>4.4498</mn> <mo>·</mo> <msup> <mn>10</mn> <mrow> <mo>-</mo> <mn>3</mn> </mrow> </msup> </mrow> </math></EquationSource> </InlineEquation>, respectively. Although the computed torque controller achieved a lower error, it depends on the accuracy of the dynamic model, making it more susceptible to unmodeled disturbances. In contrast, the MP-ADRC estimates disturbances and requires less model knowledge, making it more suitable for real-world applications. As a contribution, this work presents an integrated framework for classifying motor imagery hand movements with an accuracy of 99±2% for single-subject signals and 86% for inter-subject signals, outperforming baseline methods.

Motor imagery decoding and control techniques for virtual robotic hand movement

期刊论文
发表日期：

2025年6月20日

查看全文

Angie J. Valencia-Castaneda,
Luciano Santos Constantin Raptopoulos,
Max Suell Dutra

摘要

This work proposes a methodology for controlling the opening and closing movements of virtual hands based on the decoding of electroencephalography (EEG) signals associated with imagined left and right-hand grasping. The signals are processed using a hybrid approach that combines a feature extractor to enhance the signal-to-noise ratio with a scattering wavelet transform. A logistic regression model is employed to classify motor imagery events (left vs. right-hand). Using data from 20 volunteers obtained from an open-access database, the algorithm achieves an average accuracy of 99±2% when considering the delta, theta, and alpha/mu brain rhythms; 84±11% when using only alpha/mu rhythms; and 82±11% for signals containing information from the delta, theta, alpha/mu, and beta rhythms. To transform decoded intentions into robotic commands, a trajectory planner and control strategy were implemented to ensure movement stability and minimize tracking error during the hand’s motion in a small-diameter grasp. Experimental validation was performed using a hand simulated in Simulink/Matlab. The results demonstrated that the computed torque controller outperformed the MP-ADRC, with errors for the proximal phalanx of $1.5332\cdot 10^{-3}$ $1.5332 \cdot 10^{- 3}$ and $4.4498\cdot 10^{-3}$ $4.4498 \cdot 10^{- 3}$ , respectively. Although the computed torque controller achieved a lower error, it depends on the accuracy of the dynamic model, making it more susceptible to unmodeled disturbances. In contrast, the MP-ADRC estimates disturbances and requires less model knowledge, making it more suitable for real-world applications. As a contribution, this work presents an integrated framework for classifying motor imagery hand movements with an accuracy of 99±2% for single-subject signals and 86% for inter-subject signals, outperforming baseline methods.