Continuous locomotion mode classification using a robotic hip exoskeleton.

Area of Science:

• Robotics
• Biomedical Engineering
• Machine Learning

Background:

• Robotic exoskeletons enhance human mobility for various ambulation tasks.
• Current machine learning for prostheses is not extensively applied to exoskeletons.
• Conventional mode identification methods cause delayed assistance and reduced benefits.

Purpose of the Study:

• To develop a continuous gait phase-based Bayesian classifier for robotic exoskeletons.
• To improve the accuracy and responsiveness of exoskeleton assistance across different locomotion modes.
• To overcome limitations of discrete-time mode identification in current systems.

Main Methods:

• Developed a gait phase-based Bayesian classifier.
• Utilized mechanical sensors on a robotic hip exoskeleton.
• Conducted experiments with five able-bodied subjects.

Main Results:

• The classifier accurately identified five ambulation modes continuously throughout the gait cycle.
• Implementing multiple models within the gait cycle reduced classification error by 35% (p < 0.05).
• Bilateral sensor information reduced error by 43% compared to unilateral information (p < 0.05).

Conclusions:

• The developed classifier enhances mode classification for faster controller updates.
• Continuous mode classification provides more natural and seamless exoskeleton assistance.
• Findings offer valuable insights for future exoskeleton development using mechanical sensors.