Garcia-Lechuz Sierra, Juan (2023) Bayesian Optimization for robust robotic grasping. DLR-Interner Bericht. DLR-IB-RM-OP-2023-97. Masterarbeit. Technical University of Munich (TUM). 82 S.
![]() |
PDF
14MB |
Kurzfassung
Among the most complex tasks performed by humans is the manipulation of objects. In robotics, automating these tasks has applications in a variety of environments, such as the development of industrial processes or providing assistance to people with physical or motor disabilities. Using bio-inspired robotic hands is helping the emergence of increasingly robust and dexterous grasping strategies. However, the difficulty lies in adapting these strategies to the variety of tasks and objects, which can often be unknown also involving the computational overhead of identifying them and reconfiguring the grasp. The brute-force solution is to learn new grasps by trial and error. This method however is inefficient and ineffective, as it is based on pure randomness. In contrast, Bayesian optimization allows us to turn this process into active learning, where each attempt adds information to the approximation of an optimal grasp, in a manner analogous to a child learning. The present work aims to test Bayesian optimization in this context, providing some techniques to enhance its performance, and experimenting not only in simulation but also on real robots, as well as studying different grasp metrics that allow the grasp evaluation during the optimization process and how they behave when computing from a real system. For this, along this work, we implemented a realistic simulation environment using PyBullet, which emulates the real experimental environment. This work provides experimental results using the Light Weight robotic arm, designed at the German Aerospace Center (DLR), and two tridactyl robotic hands, the CLASH (DLR) and ReFlex TakkTile (Right Hand Robotic), demonstrating the usefulness of the method for performing unknown object grasping even in the presence of noise and uncertainty inherent in a real-world environment. Consequently, this work contributes with practical knowledge to the studied field and serves as a proof-of-concept for future grasp planning and robotic manipulation technology.
elib-URL des Eintrags: | https://elib.dlr.de/197869/ | ||||||||
---|---|---|---|---|---|---|---|---|---|
Dokumentart: | Berichtsreihe (DLR-Interner Bericht, Masterarbeit) | ||||||||
Titel: | Bayesian Optimization for robust robotic grasping | ||||||||
Autoren: |
| ||||||||
Datum: | September 2023 | ||||||||
Referierte Publikation: | Nein | ||||||||
Open Access: | Ja | ||||||||
Gold Open Access: | Nein | ||||||||
In SCOPUS: | Nein | ||||||||
In ISI Web of Science: | Nein | ||||||||
Seitenanzahl: | 82 | ||||||||
Status: | veröffentlicht | ||||||||
Stichwörter: | robotic gasping, bayesian optimization, robotic hands, robot manipulation | ||||||||
Institution: | Technical University of Munich (TUM) | ||||||||
Abteilung: | Department of Informatics | ||||||||
HGF - Forschungsbereich: | Luftfahrt, Raumfahrt und Verkehr | ||||||||
HGF - Programm: | Raumfahrt | ||||||||
HGF - Programmthema: | Robotik | ||||||||
DLR - Schwerpunkt: | Raumfahrt | ||||||||
DLR - Forschungsgebiet: | R RO - Robotik | ||||||||
DLR - Teilgebiet (Projekt, Vorhaben): | R - On-Orbit Servicing [RO] | ||||||||
Standort: | Oberpfaffenhofen | ||||||||
Institute & Einrichtungen: | Institut für Robotik und Mechatronik (ab 2013) > Autonomie und Fernprogrammierung | ||||||||
Hinterlegt von: | Roa Garzon, Dr. Máximo Alejandro | ||||||||
Hinterlegt am: | 09 Okt 2023 08:34 | ||||||||
Letzte Änderung: | 09 Okt 2023 08:34 |
Nur für Mitarbeiter des Archivs: Kontrollseite des Eintrags