A hybrid quasi-digital/neuromorphic architecture for tactile sensing in humanoid robots

Abstract

Neuromorphic engineering aims to study and develop, among the others, a new class of systems mimicking key aspects of biological systems, such as spiking (event-driven) information processing and communication, adaptive and learning behavior. Common application domains are sensory and cognitive systems, with a strong relation with the (humanoid) robotic world. Indeed, several neuromorphic sensors, inspired by human senses, have been developed, except tactile ones. Here we fill the gap by presenting a hybrid quasi-digital/neuromorphic architecture for robotic tactile sensors. Quasi-digital sensors share many features with neuromorphic systems, first of all the information encoding in the time domain. They naturally fit into an asynchronous event-driven fully-digital system, but to be fully integrated into neuromorphic applications the right framework and architecture has to be defined. Thanks to this approach, it is possible to seamlessly combine a precise continuously measuring system and a reactive sensory systems, retaining at the same time the advantages of low complexity, low area, low power consumption, distributed, and robust data acquisition systems.