Ankle torque forecasting using time-delayed neural networks.

Area of Science:

• Biomechanics
• Robotics
• Machine Learning

Background:

• Predicting joint torque is crucial for developing intuitive human-robot interactions.
• Assistive robots require accurate real-time biomechanical data for effective assistance, especially for patients with neurological disorders.

Purpose of the Study:

• To propose a novel method for predicting ankle torque ahead of time.
• To evaluate the accuracy of a time-delayed artificial neural network for ankle torque prediction.

Main Methods:

• Utilized mean average EMG signals from four muscles, joint angle, and angular velocity of the right ankle as input parameters.
• Trained a time-delayed artificial neural network using data from five healthy subjects.
• Compared model predictions with torque calculated from inverse dynamics.

Main Results:

• The model accurately predicted ankle torque up to 1 second ahead of time.
• Achieved a normalized root mean squared error of less than 15% and a coefficient of determination over 0.85 for predictions up to 1 second in the future.

Conclusions:

• The developed model demonstrates significant potential for predicting joint torque in advance.
• This predictive capability can enhance intuitive interactions between humans and assistive robots.
• The model has direct applications in assisting patients with neurological disorders.