Virtual environment for assistant mobile robot

Abstract

This paper shows the development of a virtual environment for a mobile robotic system with the ability to recognize basic voice commands, which are oriented to the recognition of a valid command of bring or take an object from a specific destination in residential spaces. The recognition of the voice command and the objects with which the robot will assist the user, is performed by a machine vision system based on the capture of the scene, where the robot is located. In relation to each captured image, a convolutional network based on regions is used with transfer learning, to identify the objects of interest. For human-robot interaction through voice, a convolutional neural network (CNN) of 6 convolution layers is used, oriented to recognize the commands to carry and bring specific objects inside the residential virtual environment. The use of convolutional networks allowed the adequate recognition of words and objects, which by means of the associated robot kinematics give rise to the execution of carry/bring commands, obtaining a navigation algorithm that operates successfully, where the manipulation of the objects exceeded 90%. Allowing the robot to move in the virtual environment even with the obstruction of objects in the navigation path.