Target Classification Method of Tactile Perception Data with Deep Learning.

Area of Science:

• Robotics
• Machine Learning
• Sensor Technology

Background:

• Accurate manipulator operation relies on tactile sensing for target classification.
• Increasing complexity in tactile data challenges traditional machine learning algorithms.
• Existing methods struggle with pure tactile data classification due to data uncertainties.

Purpose of the Study:

• To develop an advanced model for accurate target classification using tactile sensing data.
• To address the limitations of typical machine learning algorithms in handling complex tactile data.
• To improve the precision of manipulator operations through enhanced tactile perception.

Main Methods:

• Proposed a novel deep learning model, ResNet10-v1, integrating convolutional neural networks and residual networks.
• Optimized convolutional kernel, hyperparameters, and loss function for the ResNet10-v1 model.
• Applied K-means clustering to further enhance target classification accuracy.

Main Results:

• The ResNet10-v1 model demonstrated significant improvements in tactile target classification accuracy.
• Feasibility and effectiveness were validated through extensive experimental testing.
• The optimized model successfully handled complex and uncertain tactile sensing data characteristics.

Conclusions:

• The ResNet10-v1 model offers a robust solution for tactile perception classification in robotics.
• The method provides a valuable reference for advancing tactile sensing and manipulator control.
• Further research will focus on enhancing the model's generalization capabilities for broader applications.