Automated recharging and vision-based improved localisation for a quadrotor UAV

Date
2021-03
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: In the growing field of autonomous multirotor unmanned aerial vehicles (UAVs) applications, one of the biggest challenges to achieving full autonomy is the limited flight duration. While improvements to existing battery technology is on the way, a solution is required to allow any UAV to perform self recharging without any human intervention and indefinitely increase any mission duration. Such a system could potentially unlock more UAV applications which currently seem impractical. This could allow UAVs to autonomously perform any task in remote locations and extend the radius of operation to far beyond what is currently possible. This thesis presents an autonomous recharging solution which involves a UAV performing an accurate vision based autolanding on a custom charging platform prior to charging. A direct contact approach design is proposed which makes use of onboard electrodes and charging pads located on the charging platform for electrical energy transfer. This system is designed to enable any UAV to self recharge with zero human intervention. A quadrotor UAV was built for the purpose of practically demonstrating the proposed solution. A Pixhawk flight controller using PX4 was chosen as the avionics of the vehicle. Custom controller gains were designed for the UAV in order to satisfy the desired flight characteristic requirements. An autonomous vision based autolanding strategy for this vehicle was designed and simulated using software-in-the-loop (SITL) and hardware-in-the-loop (HITL) simulations in Gazebo simulator. This allowed for the actual PX4 flight control software to be executed on an accurate simulated model of the UAV prior to practical flight tests. A ROS package was developed for the autolanding which was responsible for image processing, publishing control setpoints and processed visual feedback data to the flight controller for accurate control during the autolanding. An average landing accuracy of 4cm was achieved in practical flight tests using this strategy. In many UAV applications highly accurate localisation of the UAV is required which is unachievable using traditional Global Navigation Satellite System (GNSS). This thesis presents a solution which involves using the charging stations as visual landmarks in order to localise the UAV. A marker based Simultaneous Localisation and Mapping (SLAM) strategy is implemented which aims to map the charging stations in the inertial frame while simultaneously localising the UAV within the same reference frame. The SLAM output is published to the flight controller where it is included in the UAV’s state estimation process using PX4’s EKF2. Simulation and practical results are provided which showed that the localisation accuracy could be greatly improved provided that the stations were accurately mapped and that the stations were clearly visible.
AFRIKAANSE OPSOMMING: In die groeiende veld van outonome multirotor UAV toepassings, is die beperkte vlugduur een van die grootste uitdagings om volle outonomie te bewerkstellig. Terwyl verbeterings aan bestaande battery tegnologie op pad is, is ’n oplossing nodig om enige UAV self te laat laai sonder menslike ingryping en die duur van die missie onbepaald te verhoog. So ’n stelsel kan moontlik meer UAV-toepassings ontsluit wat tans onprakties lyk. Dit kan UAV’s in staat stel om enige taak op afgele¨e plekke outonoom uit te voer en die werkstraal verder te brei as wat tans moontlik is. Hierdie proefskrif bied ’n outonome herlaaioplossing aan wat beteken dat ’n UAV ’n akkurate visie-gebaseerde outolanding op ’n aangepaste laadplatform moet uitvoer voordat dit gelaai word. ’n Ontwerp vir direkte benaderings word voorgestel wat gebruik maak van aan boord elektrodes en laaikussens wat op die laaiplatform gele¨e is vir die oordrag van elektriese energie. Hierdie stelsel is ontwerp om enige UAV in staat te stel om self te herlaai sonder menslike ingryping. ’n Quadrotor UAV is gebou om die voorgestelde oplossing prakties te demonstreer. ’n Pixhawk vliegbeheerder en PX4 vlugbeheer sagteware, is gekies as die lugvaart van die voertuig. Aangepaste beheerderstygings is vir die UAV ontwerp om aan die verlangde vlugkenmerkvereistes te voldoen. ’n Outonome visie-gebaseerde outolandingstrategie vir hierdie voertuig is ontwerp en gesimuleer met behulp van SITL en HITL simulasies in Gazebo-simulator. Hierdeur kon die werklike PX4-vliegbeheersagteware volgens ’n akkurate gesimuleerde model van die UAV uitgevoer word voor praktiese vlugtoetse. ’n ROS-pakket is ontwikkel vir die outolanding wat verantwoordelik was vir beeldverwerking, die publisering van kontrolepunte en verwerkte visuele terugvoerdata aan die vlugbeheerder vir akkurate beheer tydens die outolanding. ’n Gemiddelde landingsakkuraatheid van 4 cm is in praktiese vlugtoetse met behulp van hierdie strategie bereik. In baie UAV-toepassings is baie akkurate lokalisering van die UAV nodig, wat onhaalbaar is met behulp van tradisionele GNSS. Hierdie tesis bied ’n oplossing aan wat die gebruik van laaistasies as visuele bakens gebruik om die UAV te lokaliseer. ’n Merkergebaseerde SLAM strategie word ge¨ımplementeer wat daarop gemik is om die laaistasies in die traagheidsraamwerk te karteer terwyl die UAV terselfdertyd binne dieselfde verwysingsraamwerk geplaas word. Die SLAM uitset word aan die vlugbeheerder gepubliseer, waar dit ingesluit word by die beraming van die UAV’s met behulp van EXF2 van PX4. Simulasie en praktiese resultate word verskaf wat getoon het dat die lokaliseringsakkuraatheid aansienlik kan verbeter, mits die stasies akkuraat gekarteer is en dat die stasies duidelik sigbaar is.
Description
Thesis (MEng)--Stellenbosch University, 2021.
Keywords
Localisation, UCTD, Quadrotor helicopters, Unmanned aerial vehicles, Rechargeable batteries
Citation