The human nervous system is remarkably adept at interpreting sensory information, despite relying on signals that might bewilder many in computing: a cacophony of activity spikes transmitted across numerous neurons, where they merge with other similar signals.
Now, researchers have harnessed this spiking circuitry to create an artificial robotic skin, emulating some key principles of sensory signal transmission and integration found in human neurons. Although the system incorporates some non-neural elements, it benefits from the availability of chips capable of running neural networks using these spiking signals. This compatibility facilitates seamless integration with energy-efficient hardware for AI-based control software.
Spiking Signals for Localization
The nervous system under our skin is intricate and sophisticated, equipped with specialized sensors to detect sensations such as heat, cold, pressure, and pain. Generally, these sensors channel information to the spinal column for preliminary processing, enabling reflex actions without the need for brain involvement. However, the signals do travel via specialized neurons to the brain for further analysis and potential conscious awareness.
The team of researchers based in China developed a comparable system for a robotic hand's artificial skin. They focused on pressure detection, incorporating features of the nervous system such as identifying the location of stimuli and injuries, and employing multiple processing layers.
The project began with the creation of a flexible polymer skin containing embedded pressure sensors, which connected to the rest of the system through conductive polymers. Subsequent system layers converted sensor inputs into a sequence of activity spikes—brief electrical current pulses.
These spike trains convey information in four ways: through pulse shape, magnitude, spike duration, and spike frequency. In biological systems, spike frequency is the most prevalent method of information transmission, which the researchers utilized to represent pressure levels sensed by the device. The other informational forms helped develop a mechanism akin to a bar code, identifying the originating sensor for each reading.