Neuromimetic encoding/decoding of spatiotemporal spiking signals from an artificial touch sensor

A framework to discriminate tactile stimuli delivered to an artificial touch sensor is presented. Following a neuromimetic approach, we encode the signals from a 24-capacitive sensor fingertip into spiking activity through a network of leaky integrate-and-fire neurons. The activity resulting from the stimulation of the touch sensor through Braille-like dot patterns is then analysed by means of a newly defined Information measure which explicitly takes into consideration the metrics of the spike train space. Results show that an optimal discrimination of the entire set of 26 stimuli (i.e. 100% correct classification) is reached early after the stimulus onset. Interestingly, the method proves to be effective with both statically and dynamically delivered stimulation which are hard to decode because of the similarity between encoded firing activity given to the proximity of the patterns presented. The decoding analysis allowed us to corroborate the working hypothesis that human tactile discrimination relies on optimal encoding/decoding processes already at the level of the primary stage neurons in the somatosensory pathway.

I. Introduction

Fast and reliable tactile discrimination plays a paramount role in human behaviour in order to guarantee rapid response and adaptation to stimuli delivered to the fingertips [5]. Even simple object manipulation requires the ability to identify the object's properties and perform optimal action selection based on closed-loop control policies. The same holds for humanoid robotics applications in which human-like haptic tasks must rely on high stability, precision and adaptability of the system. More specifically, at the early stages of the ascending pathway, there must be a faithful encoding of the tactile stimulations into populations of spike trains, so that the central nervous system can actually decode the signals and discriminate the stimulations.