k-Means-Based Tactile Data Recognition Method for Sharp and Dangerous Objects for Robots

Abstract

A robot senses its surroundings through its “skin.” Through this tactile sensing method, a robot can obtain tactile data of various shapes and sharp objects. Then, a robot can analyze whether objects are sharp and dangerous through these tactile data. This study proposes a k-means algorithm-based tactile data recognition method for sharp and dangerous objects for robots. It develops a distributed pressure tactile sensing device that aims to simulate how the human skin layer works. By using this device, multiple sets of data are obtained after collecting the data of seven types of objects with different degrees of sharpness, namely, a hobby knife, sharp pliers, a diamond-shaped block, a square charger, a pencil-shaped block, a ping-pong ball, and a cylindrical block. The data are classified and stored using the k-means algorithm. By sensing the data of the seven types of sharp objects studied in this work in real time on the developed device for analysis, human brain's judgment on the sharpness of the objects is simulated. Through the analysis of a large amount of experimental data and algorithm optimization, experimental results show that the device can recognize relatively regular square objects, objects with curved surfaces with small changes in the radius of curvature, and objects with tips of less complexity. Recognition accuracy can reach 95%.