Neuromorphic Artificial Sense of Touch: Bridging Robotics and Neuroscience

Abstract

The development of a neuromorphic artificial sense of touch is presented. The system allows to code tactile information by means of a sequence of spikes, mimicking the neural dynamics of SA and FA human mechanoreceptors. The developed neuromorphic fingertip was able to encode naturalistic textures with a very high rate of disambiguation, up to 97% over a 10% chance level, by means of Victor-Purpura spike metrics and kNN decoding. A neurocomputational architecture inspired to the Cuneate Nucleus was also developed in order to achieve categorization of tactile stimuli in real-time while gathering the data stream. The implemented architecture was assessed by experimenting stimuli differing in the orientation of the tactile edges. The presented results are intended to contribute towards the restoration of a quasi-natural sense of touch in limb neuroprostheses, to develop effective and computationally lean artificial touch systems for robotic applications and to contribute to the open neuroscientific debate about the human somatosensory system.