Simultaneous recognition of multiple similar objects with tactile information and their 3D coordinates captured locally

Abstract

Recognition of the characteristics of objects and the recognition of the objects themselves are necessary. The objective of this research is the simultaneous recognition of multiple similar objects based on tactile and coordinate information of objects obtained by grasping movements of a multi-fingered robot hand equipped with a 3-axis tactile sensor. We propose PointNet++, which can capture information locally and globally, as a proposed method. The research objective was achieved by learning with the model of the proposed method. In addition, as a comparative method, we used FNN and PointNet++. In addition, the results of multiple classification tasks with different test data showed that the proposed method was able to recognize multiple similar objects at the same time with high accuracy and stability.