Soft-Sensor System for Grasp Type Recognition in Underactuated Hand Prostheses.

Area of Science:

• Robotics and Biomedical Engineering
• Sensor Technology
• Machine Learning Applications

Background:

• Underactuated hand prostheses lack sophisticated haptic feedback, limiting user dexterity and control.
• Integrating advanced sensing capabilities is crucial for restoring natural hand function.

Purpose of the Study:

• To develop and validate an intelligent soft-sensor system for the PrHand prosthesis.
• To incorporate haptic perception through optical fiber-based angle and force sensors.
• To enable accurate grasp type recognition using machine learning algorithms.

Main Methods:

• Fabrication and testing of optical fiber sensors for finger joint angles and fingertip contact forces.
• Evaluation of sensor response linearity and reliability across different configurations.
• Implementation and comparison of six machine learning algorithms for grasp classification.
• Validation using k-fold cross-validation with k=10.

Main Results:

• Selected angle sensors demonstrated a polynomial response with R2 > 92%.
• Tactile force sensors accurately tracked applied forces with R2 > 94%.
• K-nearest neighbor achieved 98.5% accuracy, and decision tree achieved 93.3% accuracy in classifying eight grip types.

Conclusions:

• The developed intelligent soft-sensor system effectively adds haptic perception to the PrHand prosthesis.
• Optical fiber sensors provide reliable angle and force measurements.
• Machine learning algorithms, particularly k-nearest neighbor, enable robust grasp type classification, enhancing prosthesis usability.