Stability Analysis of 3C Electronic Industry Robot Grasping Based on Visual-Tactile Sensing

Abstract

This study addresses the stability of robotic grasping and assembly in the 3C electronics industry, an area often plagued by issues such as slippage and twisting of target objects during grasping, which negatively impacts production efficiency and product quality. Previously, solutions primarily relied on conventional visual detection techniques, which suffered from low recognition rates in complex environments, strong dependence on light and background, and suboptimal performance with transparent or reflective objects. Although advances in sensor technology allowed traditional tactile detection methods to effectively gauge grasp strength and prevent objects from being damaged, these approaches were constrained by the objects' material, shape, and friction, affecting accuracy. Unfortunately, both of these require considerable datasets to train. We propose a new grasping solution with small datasets. It integrates Visual-Tactile sensing to counter these limitations via a Dual-Stream neural network-based model to synthesize data. The merged data are then fed into a Vision Transformer (ViT) network to determine the grasping stability, with an achieved accuracy of 98.86%. This approach enhances grasping stability and exhibits superior adaptability in complex environments.