Event-Driven Tactile Sensing With Dense Spiking Graph Neural Networks

Abstract

Tactile sensing is a fundamental basis for plenty of robot tasks such as object recognition, manipulation, and grasping. The recently-developed event-driven tactile sensors, which feature higher temporal resolution and lower energy consumption, are becoming attractive to endow robots with touch perception capabilities. However, existing methods are often designed for processing frame-like data (e.g., images), and hence cannot be directly applied to event-driven tactile sensing. In this article, we introduce DeepTactile, a novel approach based on a spiking graph neural network (GNN) tailored for event-driven tactile data. By leveraging the local connectivity of taxels, we structure tactile data as graphs. Building on these graphs, we design a spiking graph convolutional network (GCN) to extract meaningful features. The event-driven characteristic of spiking neural networks (SNNs) makes them well-suited for handling event-based learning. Results from experiments conducted on three event-based tactile datasets show that our method surpasses existing state-of-the-art techniques. The source code of DeepTactile is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/cqu-uisc/deepTactile</uri>.