Active Prior Tactile Knowledge Transfer for Learning Tactual Properties of New Objects

Area of Science:

• Robotics
• Artificial Intelligence
• Materials Science

Background:

• Humans leverage prior tactile experiences to efficiently recognize new objects.
• Robotic systems often lack the ability to transfer learned tactile knowledge effectively.
• Multi-modal artificial skin offers rich sensory data for tactile exploration.

Purpose of the Study:

• To develop a method for robots to actively transfer prior tactile knowledge to new objects.
• To enhance the learning efficiency and accuracy of robotic tactile property recognition.
• To investigate the robustness of knowledge transfer against irrelevant data.

Main Methods:

• Implementing Active Prior Tactile Knowledge Transfer (APTKT) in a robotic arm with multi-modal artificial skin.
• Utilizing pressing, sliding, and static contact movements with varied action parameters for data acquisition.
• Building prior tactile knowledge from feature observations across multiple sensory modalities.
• Incorporating predictions from prior object observation models as auxiliary features.

Main Results:

• Improved discrimination accuracy by approximately 10% with a single training sample and prior object features.
• Achieved over 20% improvement in discrimination accuracy by using prior object observation model predictions as auxiliary features.
• Demonstrated robustness against negative knowledge transfer (irrelevant prior tactile knowledge).

Conclusions:

• Active Prior Tactile Knowledge Transfer (APTKT) enables robots to learn object properties more efficiently.
• Multi-modal sensory data and strategic action parameters are key to effective tactile knowledge transfer.
• The proposed method shows promise for real-world robotic applications requiring tactile object recognition.