6533b850fe1ef96bd12a80d6

RESEARCH PRODUCT

Bolts detection and a combination of conventional and reinforcement learning based control of UR5 industrial robot

Fredrik Frigstad

subject

VDP::Teknologi: 500::Informasjons- og kommunikasjonsteknologi: 550::Teknisk kybernetikk: 553MAS500

description

Master's thesis in Mechatronics (MAS500) The main objective of this paper is to investigate the possibilities for using reinforcement learning to control a UR-5 robot. The paper also looks at how well reinforcement learning works to control a UR-5 robot. These questions are answered by constructing of matlab and simulink programes. Based on different mathworks example programs and scripts. In this study, reinforcement learning only works in the situation it is trained to perform. The author believe that it could work better if it were given other configurations/parameters. This will still be an interesting subject for further studies. According to the research done in this paper, the conventional control have the best control accuracy.

yearjournalcountryeditionlanguage
2021-01-01
https://hdl.handle.net/11250/2826435