Flexible coding scheme for robotic arm control driven by motor imagery decoding.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interface (BCI) technology enables machine control via physiological signals, with motor imagery (MI) electroencephalography (EEG) being a key source.
• Current MI-based BCI systems face limitations in classification model generalization and real-world prototype adoption, hindering widespread practical use, especially in rehabilitation.

Purpose of the Study:

• To address the limitations of current MI-EEG classification models by proposing an optimized neural network architecture.
• To enhance the generalization ability and accuracy of MI-EEG classification for improved BCI performance.
• To enable multi-degree-of-freedom control of robotic arms using enhanced BCI signal processing.

Main Methods:

• Artifact removal from MI-EEG signals using thresholding techniques based on artifact evaluation indices.
• Data augmentation of MI-EEG signals utilizing the empirical mode decomposition (EMD) algorithm.
• Optimization of the classification model through ensemble learning (EL) and transfer learning (TL) fine-tuning strategies.
• Mapping EEG recognition results to robotic arm control commands via a flexible binary encoding strategy for multi-DOF control.

Main Results:

• Empirical mode decomposition (EMD) demonstrated significant data augmentation for small datasets.
• Ensemble learning (EL) improved intra-subject model performance, while transfer learning (TL) enhanced inter-subject performance.
• A binary coding strategy successfully expanded control instructions, enabling 7 degrees of freedom robotic arm control using four MI-EEG signal types.

Conclusions:

• The proposed optimized neural network architecture significantly improves MI-EEG classification accuracy and model generalization.
• The integration of EL, TL, and EMD offers a robust approach for enhancing BCI system performance.
• This work advances BCI capabilities by extending the control instruction set and enabling complex robotic arm manipulation.