Contact Pattern Recognition of a Flexible Tactile Sensor Based on the CNN-LSTM Fusion Algorithm

Area of Science:

• Materials Science
• Robotics
• Artificial Intelligence

Background:

• Tactile sensing is crucial for human-machine interaction.
• Flexible sensors offer enhanced dynamic response capabilities.
• Distinguishing contact patterns is essential for advanced robotic applications.

Purpose of the Study:

• To design and manufacture a flexible tactile sensor using PVDF.
• To develop and evaluate a CNN-LSTM model for classifying tactile contact patterns.
• To compare the performance of the CNN-LSTM model against CNN and Random Forest algorithms.

Main Methods:

• Fabrication of a flexible tactile sensor utilizing PVDF material.
• Collection of time-series data for four distinct contact patterns: stroking, patting, kneading, and scratching.
• Implementation and training of a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model (CNN-LSTM) for pattern recognition.

Main Results:

• The CNN-LSTM model achieved high recognition accuracies for individual patterns: 99.60% (stroking), 99.67% (patting), 99.07% (kneading), and 99.40% (scratching).
• The average accuracy for the CNN-LSTM model was 99.43%, outperforming the CNN model (96.67%) and the Random Forest algorithm (91.39%).
• Experimental results confirm the efficiency of the developed CNN-LSTM model.

Conclusions:

• The developed flexible tactile sensor exhibits excellent dynamic response characteristics.
• The CNN-LSTM model demonstrates highly effective classification and recognition of diverse tactile contact patterns.
• This approach significantly advances the capabilities of tactile sensing in human-machine interaction systems.