Pattern Recognition of EMG Signals by Machine Learning for the Control of a Manipulator Robot.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Customization is a key challenge in developing effective Human Machine Interfaces (HMI) for physiotherapy and rehabilitation.
• Adapting assistance systems to individual user characteristics is crucial for successful therapy outcomes.

Purpose of the Study:

• To develop a customizable HMI for rehabilitation using surface Electromyography (sEMG) signals.
• To investigate the use of neural networks for pattern recognition of muscle contractions.
• To control an anthropomorphic manipulator robot based on user muscle activity.

Main Methods:

• A database of sEMG signals from the biceps brachii in healthy individuals was created.
• Neural networks were employed for pattern recognition of muscle contraction signals.
• One-Hot Encoding was used for movement labeling, controlling a robot via a state machine.
• A LabVIEW test platform was designed for real-time EMG signal classification and trajectory generation.

Main Results:

• Preliminary results indicate a reduced learning curve with interface customization.
• The developed system successfully used muscle contractions to direct a virtual robot's end effector.
• Real-time trajectory generation was achieved through EMG signal classification.

Conclusions:

• The proposed HMI system demonstrates the feasibility of using customized sEMG-based control for rehabilitation.
• Customization significantly enhances the user's learning experience and system adaptability.
• This approach offers a promising direction for developing personalized assistive technologies in physiotherapy.