Research on blindsight technology for object recognition and attitude determination based on tactile pressure...

Area of Science:

• Robotics
• Mechanical Engineering
• Artificial Intelligence

Background:

• Robotic grasping often relies on visual guidance, which is limited by lighting and occlusion.
• Existing methods struggle with accurate object recognition and posture determination in complex environments.

Purpose of the Study:

• To propose a non-visual method for object recognition and posture determination using contact force feedback.
• To develop a system that enhances robotic hand grasping capabilities in challenging conditions.

Main Methods:

• Constructed a contact surface deformation model based on Hertz contact and elastoplastic unloading theories.
• Developed a dimensionality reduction analysis for vector force arrays, converting 3D force data into a 2D scalar array.
• Utilized convolutional neural networks (CNNs) trained on mechanical features of contact surfaces.

Main Results:

• Achieved millimeter-level precision in grasping position determination.
• Determined grasped object posture with a two-dimensional angle accuracy of less than 1 degree.
• Demonstrated the method's effectiveness in meeting power system business scenario requirements.

Conclusions:

• The contact force feedback method offers a robust alternative to visual guidance for robotic grasping.
• This blind-sense technology has broad applications in humanoid robot non-visual perception.
• The approach provides significant engineering value for advanced robotic manipulation.