Effect of Gait Speed on Trajectory Prediction Using Deep Learning Models for Exoskeleton Applications.

Area of Science:

• Biomechanics
• Robotics
• Machine Learning

Background:

• Gait speed significantly influences biomechanical patterns and joint kinematics.
• Accurate prediction of gait trajectories is crucial for advanced applications like exoskeleton control.

Purpose of the Study:

• To evaluate the effectiveness of fully connected neural networks (FCNNs) in predicting human gait trajectories across a wide range of walking speeds.
• To assess the performance of different FCNN models (generalised, low-speed, high-speed, low-high-speed) for predicting gait kinematics, particularly hip, knee, and ankle angles.
• To determine the predictive capabilities of FCNNs for speeds within and outside the training data range, relevant for real-time exoskeleton adaptation.

Main Methods:

• Collected gait data from 22 healthy adults walking at 28 speeds (0.5-1.85 m/s).
• Developed and evaluated four distinct fully connected neural network models.
• Assessed predictive performance using short-term (one-step-ahead) and long-term (200-time-step) recursive predictions, measuring Mean Absolute Error (MAE).

Main Results:

• Specialized low- and high-speed models showed significant performance degradation (43.7%-90.7% MAE increase) on excluded speeds.
• The low-high-speed model demonstrated improved performance (2.8% short-term, 9.8% long-term) when tested on excluded medium speeds.
• FCNNs showed interpolation capabilities for speeds within the training range but reduced accuracy for speeds outside this range.

Conclusions:

• Fully connected neural networks can interpolate gait predictions within the tested speed range.
• Predictive accuracy of FCNNs diminishes for gait speeds exceeding or falling below the training data boundaries.
• Model selection and training data range are critical factors for reliable FCNN application in dynamic gait prediction for assistive devices.