Object Recognition Using Mechanical Impact, Viscoelasticity, and Surface Friction During Interaction.

Area of Science:

• Robotics
• Mechanical Engineering
• Artificial Intelligence

Background:

• Current robotic haptic object recognition often uses statistical measures from interaction signals.
• Mechanical properties offer a more robust basis for object representation.

Purpose of the Study:

• To propose a novel framework for robotic haptic object recognition using key mechanical properties.
• To evaluate the effectiveness of stiffness, viscosity, friction coefficient, and coefficient of restitution for object classification and clustering.

Main Methods:

• Estimation of mechanical properties (stiffness, viscosity, friction, coefficient of restitution) in real-time using a dual Kalman filter.
• Utilizing these properties for object classification and clustering in a robotic system.
• Testing the framework on a robot identifying 20 distinct objects through haptic exploration.

Main Results:

• The proposed framework achieved a high recognition rate of 98.18 ± 0.424%.
• All four mechanical properties were found to be essential for optimal recognition performance.
• The method demonstrated superior performance in object clustering compared to statistical parameter-based approaches.

Conclusions:

• Robotic haptic object recognition can be significantly improved by incorporating intrinsic mechanical properties.
• The dual Kalman filter provides an effective real-time estimation of these properties.
• This framework offers a more robust and efficient approach to object recognition and clustering for robots.