Iteratively Calibratable Network for Reliable EEG-Based Robotic Arm Control.

Area of Science:

• Neuroscience
• Robotics
• Human-Computer Interaction

Background:

• Robotic arms in shared workspaces require accurate human intention interpretation for safety and efficiency.
• Electroencephalogram (EEG) signals offer a direct human-robotic arm communication channel but face challenges in data collection and calibration due to signal ambiguity.
• Existing EEG-based applications suffer from reduced reliability stemming from calibration difficulties with new users.

Purpose of the Study:

• To develop an iteratively calibratable network to improve the reliability and efficiency of EEG-based robotic arm control.
• To enable effective adaptation of a pre-trained network to new users with minimal new data.
• To enhance the intuitiveness and command class diversity of EEG control systems.

Main Methods:

• Proposed an iteratively calibratable network integrating feature inputs and network expansion techniques.
• Combined motor imagery and speech imagery datasets to increase intuitiveness and command classes.
• Conducted pseudo-online evaluations with real-time robotic arm operation and offline data analysis.

Main Results:

• The proposed method outperformed comparison groups across 10 sessions.
• Achieved competitive results when combining motor and speech imagery paradigms.
• Demonstrated successful network calibration and personalization using only new user data.

Conclusions:

• The iteratively calibratable network significantly enhances EEG-based robotic arm control reliability and efficiency.
• The integration of feature inputs and network expansion allows for effective adaptation to new users.
• Combining diverse datasets like motor and speech imagery broadens the applicability and intuitiveness of EEG control systems.