Generation of tactile maps for artificial skin

Area of Science:

• Computational neuroscience
• Robotics
• Sensory processing

Background:

• Previous studies successfully learned surface representations from visual and auditory data.
• Advancements in robotic artificial skin offer a new sensory modality for research.
• Tactile sensing presents unique sensorimotor challenges compared to vision and audition.

Purpose of the Study:

• To investigate ordered sensory map formation in tactile systems using computational tools.
• To develop a method for learning representations of complex tactile surfaces.
• To address the challenges of processing tactile data for 3D embeddings.

Main Methods:

• Development of a novel multi-dimensional scaling algorithm named ANISOMAP.
• Utilizing intrinsic properties of tactile data to construct somatotopic maps.
• Testing the algorithm on both simulated and hardware tactile data.

Main Results:

• ANISOMAP successfully learns representations of non-trivial tactile surfaces.
• The algorithm produces high-quality, topologically and geometrically accurate 3D embeddings.
• Reconstructions demonstrated robustness against hardware failures.

Conclusions:

• Computational tools can effectively learn tactile surface representations.
• ANISOMAP advances multi-dimensional scaling for complex sensory data.
• The developed method has implications for robotics and artificial sensory systems.