Object Detection and Mapping During European Robotic Competitions - Lesson Learned

Abstract

This paper describes the approach to three European robotic competitions ERL 2017 Major Tournament, ERL 2018 Local Tournament and ELROB 2018. In all of the competitions, GPU enabled SLAM was used to deliver the 3D map of the environment during the mission. In both ERL competitions, deep neural networks were used to identify objects of potential interest. Datasets used to train models and architectures of neural networks are described. All of the object detection models used during the competitions are published in a publicly available repository11https://github.com/karolmajek/ERL2017-ERL2018-Emergency-Object-Detection.