Vision-Based Tactile Sensor Mechanism for the Estimation of Contact Position and Force Distribution Using Deep Learning

Area of Science:

• Robotics and Machine Vision
• Sensor Technology
• Artificial Intelligence

Background:

• Tactile sensing is crucial for robotic manipulation and human-computer interaction.
• Accurate estimation of contact position, area, and force distribution remains a challenge.
• Vision-based approaches offer a non-contact method for tactile data acquisition.

Purpose of the Study:

• To develop and validate a vision-based tactile sensor system.
• To train a neural network for estimating tactile contact properties.
• To investigate the influence of sensor design on performance.

Main Methods:

• Utilized a vision-based tactile sensor system with a stereo camera setup.
• Collected image data and input loads under various motions (X, Y, Z, RxRy).
• Modified a VGG-16 convolutional neural network (CNN) for regression tasks using transfer learning.

Main Results:

• Successfully trained a deep learning model to estimate contact position, area, and force distribution.
• Demonstrated the system's ability to handle diverse input loads and motions.
• Validated the model's performance across different tactile sensor thicknesses and shapes.

Conclusions:

• The developed vision-based tactile sensor system effectively estimates contact properties using deep learning.
• The system shows promise for applications requiring precise tactile feedback.
• Further research can optimize sensor design and explore advanced neural network architectures.