Juht-järgijasüsteem mehitamata maismaasõidukile
Files
Date
2017
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Mehitamata sõidukeid saab kasutada efektiivselt, kui neil on autonoomse sõitmise võimekus.Autonoomse sõitmise üks funktsionaalsustest on juht-järgijasüsteem, mis tähendab, et robot peab järgi minema etteantud objektile. Tavaliselt on defineeritud objektiks inimene, mis võimaldab hoida inimeste käed vabana, võrreldes situatsiooniga, kus masina liigutamiseks tuleb kasutada pulti, mis muudab sõdurid militaarolukorras haavatavaks sihtmärgiks. Olemasolevad juht-järgija süsteemid kasutavad funktsioneerimiseks kombinatsiooni erinevatest sensoritest, kasutades nii LIDAReid, GPS-i, infrapunamarkereid kui ka kaameraid. Selle töö eesmärgiks on arendada ja testida juht-järgijasüsteem, mis kasutab ainult kaamerasisendit ning mis oma kontseptsioonilt peaks olema võimeline järgnema igale objektile, st objekti tüüp ja välimus ei pea olema eeldefineeritud.Täpsemalt kasutab süsteem käitumuslikku kloonimist sügavate konvolutsiooniliste tehisnärvivõrkude treenimiseks, et ennustada kaamera sisendi põhjal roboti liigutamiseks vajalikke kiiruse ja pööramise käske. Tulemused näitavad, et pakutud süsteem töötab, kuid vajab edasiarendust, et teha see reaalse olukorra jaoks piisavalt robustseks ning turvaliseks.
Unmanned ground vehicles can be utilized effectively, if they have autonomous capabilities. One of those capabilities is leader-follower functionality, which means that the robot has to follow a predefined object. Usually the predefined object is a person, which is still more effective than controlling the robot with teleoperation using a remote control. It is worthwhile, because teleoperation requires the full attention of the operator while leader-follower system allows the person to keep their situational awareness, which is essential in a military environment. Leader-follower systems often exploit multipletechnologies: LIDARs, GPS, infrared markers, radar transponder tags and cameras. The aim of this thesis is to develop and test the proof of concept leader-follower system, which relies on camera vision only and is able to follow any object which means that the class of the object does not have to be predefined. Specifically, the approach uses behavioral cloning to train deep siamese network with convolutional layers to determine the velocity and turning commands of the vehicle based on input from camera. The results show that the proof of concept system works, but requires further development in order to make it robust and safe.
Unmanned ground vehicles can be utilized effectively, if they have autonomous capabilities. One of those capabilities is leader-follower functionality, which means that the robot has to follow a predefined object. Usually the predefined object is a person, which is still more effective than controlling the robot with teleoperation using a remote control. It is worthwhile, because teleoperation requires the full attention of the operator while leader-follower system allows the person to keep their situational awareness, which is essential in a military environment. Leader-follower systems often exploit multipletechnologies: LIDARs, GPS, infrared markers, radar transponder tags and cameras. The aim of this thesis is to develop and test the proof of concept leader-follower system, which relies on camera vision only and is able to follow any object which means that the class of the object does not have to be predefined. Specifically, the approach uses behavioral cloning to train deep siamese network with convolutional layers to determine the velocity and turning commands of the vehicle based on input from camera. The results show that the proof of concept system works, but requires further development in order to make it robust and safe.