A Haptic Glove with Flexible Piezoresistive Sensors Made by Graphene and Polyurethane Sponge for Object Recognition Based on Machine Learning Methods

Abstract

The rapid advancement of artificial intelligence technology has propelled flexible tactile sensors into a wide range of application prospects across multiple domains. Flexible tactile sensors can convert the active dynamic tactile sensing signals into digital signals, which provide real-time insight and prediction capabilities by using machine learning methods to analyze the digital signals. This paper reports a low-cost and efficient strategy to fabricate flexible piezoresistive sensors with porous sponge structures. The prepared flexible piezoresistive sensors based on polyurethane (PU) sponge and graphene exhibit excellent properties such as excellent sensitivity (1.7356 kPa^–1 at 0–55 kPa pressure), fast response/recovery time (147 ms/59 ms), small hysteresis error (6.51%), and stable repeatability (under 2000 cyclic pressure tests). The sensor is well suited for wearable devices due to its sensitivity over a wide range and its fast, cost-effective design process. Therefore, a haptic glove is designed with the flexible piezoresistive sensors for object recognition. By wearing the haptic glove, 1500 sets of time series signals during the grasp process for 15 different objects are detected and collected precisely. Then, the Residual Network (ResNet) with great feature extraction and generalization ability is constructed to recognize the 15 objects by the tactile time serial signals detected from the haptic glove, and the corresponding recognition accuracy is 95.67%. This work combines flexible tactile sensors with machine learning methods, providing an effective approach for flexible tactile sensors in more innovative applications.