A Mobile Robot Visual Perception System based on Deep Learning Approach

Abstract

In this paper, we present the novel mobile robot perception system based on a deep learning framework. The hardware subsystem consists of an Nvidia Jetson Nano development board integrated with two parallelly positioned Basler daA1600-60uc cameras, while the software subsystem is based on the convolutional neural networks utilized for semantic segmentation of the environment scene. A Fully Convolutional neural Network (FCN) based on the ResNet18 backbone architecture is utilized to provide accurate information about machine tool models and background position in the image. FCN model is trained on our custom-developed dataset of a laboratory model of manufacturing environment and implemented on mobile robot RAICO (Robot with Artificial Intelligence based COgnition).