Musculoskeletal Neural Network path generator for a virtual upper-limb active controlled orthosis

Area of Science:

• Biomedical Engineering
• Neuroscience
• Robotics

Background:

• Neuromusculoskeletal system modeling is crucial for developing advanced assistive devices.
• Electromyographic (EMG) signals offer a direct interface for controlling prosthetic and orthotic systems.
• Virtual active orthoses require sophisticated control strategies for natural and intuitive user interaction.

Purpose of the Study:

• To present a non-parametric model of the biceps brachii neuromusculoskeletal system.
• To develop a path generator for controlling a virtual active orthosis using EMG signals.
• To validate the model's performance with experimental data and closed-loop implementation.

Main Methods:

• A differential neural network (DNN) identifier was employed as the core of the path generator.
• Raw electromyographic (EMG) signals were used as input to derive reference angular positions and velocities.
• The model was trained and implemented in a closed-loop system for validation.

Main Results:

• The proposed non-parametric model successfully generated reference angular paths for active orthosis control.
• The control strategy ensured users reached set-points within position constraints.
• The device accurately followed natural reference paths derived from raw EMG signals.

Conclusions:

• The developed non-parametric neuromusculoskeletal model provides an effective method for controlling virtual active orthoses.
• The use of DNNs with EMG signals enables intuitive and natural device control.
• Experimental validation confirms the model's potential for real-world applications in assistive robotics.