Autonomous Localization of Hexapod Walking Robot
Autonomní lokalizace šestinohého kráčejícího robotu
Authors
Supervisors
Reviewers
Editors
Other contributors
Journal Title
Journal ISSN
Volume Title
Publisher
České vysoké učení technické v Praze
Czech Technical University in Prague
Czech Technical University in Prague
Date
Abstract
Práce se zabývá vizuální lokalizací šestinohého kráčejícího robotu. V posledních letech bylo vyvinuto několik metod vizuální lokalizace založených na simultánní lokalizaci a mapování využívající významné body v prostředí. Některé z těchto metod využívající různé senzory se ukázaly velice efektivní pro lokalizaci kolových robotů. V této práci se zabýváme složitější lokalizací šestinohého kráčejícího robotu, kde vlivem kráčení dochází k oscilacím a rychlým rotačním pohybům, které nepřiznivě ovlivňují měření senzory robotu. V práci jsou porovnány tři různé metody simultánní lokalizace a mapování využívající tři různé typy senzorů. Na základě provedených experimentů byla změřena přesnost lokalizace, kterou jsou testované metody schopny poskytnout. Provedené experimenty spočívaly v lokalizaci šestinohého kráčejícího robotu v kancelářském prostředí. Na základě ohodnocení přesnosti lokalizace můžeme konstatovat, že metody simultánní lokalizace a mapování mohou poskytovat relativně přesnou lokalizaci, zejména v kratším časovém horizontu. Pro náročnější případy navrhujeme řešení na základě aktivní lokalizace.
Vision-based localization of hexapod walking robot is addressed in this thesis. In particular, localization methods called Simultaneous localization and mapping that are based on visual approaches have been studied. In recent years, several methods of simultaneous localization and mapping using detection of salient points in the robot's environment by various types of cameras were developed. Some of these methods were proved to be very effective for localization of wheeled robots. However, in this thesis, we address more challenging visual localization of hexapod walking robot where sensors suffer from motion blur, the robot's oscillations and fast rotations around several axes, induced by the robot's locomotion. Three different simultaneous localization and mapping methods (S-PTAM, ORB-SLAM2 and RGB-D SLAM v2) in combination with three different types of sensors are evaluated in this thesis. Moreover, we propose to evaluate performance of the localization methods by evaluation proposed in literature that has been further improved according to performed experimental deployments of the robot. Experiments with localization systems were performed on a low-cost hexapod platform in an office environment. Based on the presented results, we can conclude that the evaluated vision based localization can provide relatively accurate position estimation of the hexapod robot especially for short term localization. For more demanding cases, we propose to use enhance the position estimation by active localization approaches.
Vision-based localization of hexapod walking robot is addressed in this thesis. In particular, localization methods called Simultaneous localization and mapping that are based on visual approaches have been studied. In recent years, several methods of simultaneous localization and mapping using detection of salient points in the robot's environment by various types of cameras were developed. Some of these methods were proved to be very effective for localization of wheeled robots. However, in this thesis, we address more challenging visual localization of hexapod walking robot where sensors suffer from motion blur, the robot's oscillations and fast rotations around several axes, induced by the robot's locomotion. Three different simultaneous localization and mapping methods (S-PTAM, ORB-SLAM2 and RGB-D SLAM v2) in combination with three different types of sensors are evaluated in this thesis. Moreover, we propose to evaluate performance of the localization methods by evaluation proposed in literature that has been further improved according to performed experimental deployments of the robot. Experiments with localization systems were performed on a low-cost hexapod platform in an office environment. Based on the presented results, we can conclude that the evaluated vision based localization can provide relatively accurate position estimation of the hexapod robot especially for short term localization. For more demanding cases, we propose to use enhance the position estimation by active localization approaches.