Subject-Independent Continuous Locomotion Mode Classification for Robotic Hip Exoskeleton Applications

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Autonomous lower-limb exoskeletons require locomotion mode classification for adaptive assistance.
• Current methods often need user-specific tuning, have slow updates, and require external sensors.

Purpose of the Study:

• To develop a subject-independent, continuous locomotion mode classifier for hip exoskeletons.
• To utilize a deep convolutional neural network (CNN) with minimal, integrated sensors.

Main Methods:

• A deep CNN was trained on an open-source gait biomechanics dataset.
• Model optimization included transition label timing, architecture, and sensor placement.
• Performance was evaluated against machine learning benchmarks.

Main Results:

• The optimized DL model achieved a 3.13% classification error, significantly outperforming benchmarks (p<0.05).
• Steady-state and transitional classification errors were 0.80% and 6.49%, respectively.
• The model demonstrated robust performance on unseen slopes and varied settings.

Conclusions:

• A novel, subject-independent locomotion mode classification framework was presented.
• The deep learning approach enables continuous, seamless mode transitions for hip exoskeletons.
• This technology advances robotic exoskeleton applications for community ambulation.