Perception of Tactile Directionality via Artificial Fingerpad Deformation and Convolutional Neural Networks

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Humans possess natural tactile directionality perception through fingerpad skin displacement.
• Robotic systems require enhanced sensory feedback for sophisticated object manipulation.
• Artificial tactile sensing offers a pathway to replicate human-like perception in robots.

Purpose of the Study:

• To develop and evaluate a robotic system capable of perceiving tactile directionality.
• To utilize artificial tactile sensors and convolutional neural networks (CNNs) for this purpose.
• To improve robotic object manipulation and enhance situational awareness in telerobotic systems.

Main Methods:

• Deformable artificial tactile sensors were employed to capture fingerpad deformation data.
• Two CNNs were trained on tactile images generated during object perturbations.
• A primary CNN provided directionality estimates, while a secondary CNN quantified uncertainty.

Main Results:

• The primary CNN achieved a low error rate of 4.3% for a 20° angular resolution.
• The model demonstrated real-time implementation with various object shapes and widths.
• Performance was benchmarked against an open-source force estimation network.

Conclusions:

• The developed system successfully replicates tactile directionality perception in robots.
• This technology can significantly enhance the situational awareness of human operators in telerobotic applications.
• It provides a foundation for developing advanced decision-making algorithms for semi-autonomous robots.