Bioinspired spiking architecture enables energy constrained touch encoding

Area of Science:

• Robotics and Artificial Intelligence
• Biomimetic Sensing
• Neuroscience

Background:

• The human sense of touch is vital for interaction and object manipulation, relying on distributed mechanoreceptors and efficient neural processing.
• Current robotic tactile sensing faces challenges in wiring, energy consumption, scalability, and parallel processing.
• Replicating biological touch capabilities is essential for advancing autonomous systems.

Purpose of the Study:

• To develop a modular artificial tactile system that overcomes limitations of current robotic sensing solutions.
• To mimic the early stages of the human somatosensory system for enhanced tactile information processing.
• To demonstrate a scalable and energetically sustainable solution for touch perception in autonomous systems.

Main Methods:

• Integration of a Fiber Bragg Grating (FBG)-based electronic skin (e-skin) with a spiking neural network (SNN).
• Implementation of the SNN on a neuromorphic chip for efficient, low-power computation.
• Bioinspired connectivity was used to investigate its influence on tactile processing.

Main Results:

• Achieved up to 10x localization super-resolution and improved localization accuracy by 32% compared to deep learning methods.
• Demonstrated generalization to multitouch and dynamic tactile conditions.
• The neuromorphic SNN showed robustness to analog neuron constraints, enabling highly parallel, sub-mWatt computation.

Conclusions:

• The developed artificial tactile system offers a scalable, energy-efficient solution for touch perception.
• The system provides mechanistic insights into bioinspired tactile processing.
• This approach has immediate applications in autonomous systems requiring safe human interaction and operation in dynamic environments.