Interactive Quadruped Felid Class Robot with a Neural Processing Unit

Abstract

The authors consider the urgent task of developing bionic robots, in particular robots on four legs. Their advantages are the ability to move on uneven terrain, to perform reconnaissance, rescue and other dangerous work, where they could replace humans. A review of the existing best-known and most functional bionic four-legged robots is given, with descriptions of their strengths and weaknesses, as well as peculiarities of their movement and use. The main problems in the development of such devices and their control systems are highlighted. The article provides information on the research and development of an interactive bionic robot of the felid class, whose skeletal structure control implementation is deeply explored. The features of hardware and software implementation of the robot are considered, and schematic and real images of the construction are presented. The application of a microcomputer device with a neural processing unit to solve the problem of machine vision is highlighted. The results of testing machine vision using the Yolo3 neural network in streaming video mode are presented. The average accuracy of the open face recognition as a result of the tests was 95 %. For different degrees of occlusion, the average score was 80 %, and occlusion variants in which the neural network was unable to recognize faces were also identified. The article concludes with a discussion of the advantages and disadvantages of the proposed robot and the possibility of its application in human life, including the solution of various practical tasks.