LoCAR–Low-Cost Autonomous Robot for Object Detection with Voice Command and MobileNets

Abstract

This work details the design, construction, implementation and testing of a standalone robot, based on a convolutional neural network, which receives a voice command, searches and recognizes the target through its camera and moves to the object or person properly recognized. The success rate for the recognition stage has reached 82% in the median for objects tested, 100% for chairs, bottles and people. The processing was performed on a Raspberry Pi 3 B board integrated with an Arduino UNO to control the actuators.