Navigacija poljskega robota

• Main Author: TURK, MARTIN
• Other Authors: Skočaj, Danijel
• Format: Bachelor Thesis
• Language: Slovenian
• Published: 2018
• Subjects: avtonomna navigacija; robot; LIDAR; DBSCAN; robotski operacijski sistem; konvolucijske nevronske mreže; autonomous navigation; Robot Op- eration System; CNN; sami
• Online Access: https://repozitorij.uni-lj.si/IzpisGradiva.php?id=102759; https://repozitorij.uni-lj.si/Dokument.php?id=113116&dn=

Robotski sistemi so danes že tako napredni, da so ponekod zamenjali ljudi pri opravljanju dela in predstavljamo si lahko, da bo tega v prihodnosti še več. Da so roboti inteligentni in lahko bolje opravljajo delo kot človek, se morajo biti sposobni sami navigirati v okolju in se odločati v različnih situ- acijah. Razvili smo inteligentnega robota za tekmovanje FRE (ang. Field Robot Event), ki se avtonomno navigira po polju koruze. V tem diplomskem delu smo se osredotočili na razvoj aplikacije za avtonomno vožnjo robota s pomočjo senzorja LIDAR ter prikazali, kako bi vožnjo izboljšali s pomočjo zaznavanja koruze s kamero. Podatki pridobljeni s senzorjem LIDAR, imajo običajno veliko šuma, zato jih moramo obdelati z algoritmom za gručenje DBSCAN. V zaključku opišemo eksperimentalne rezultate ter naštejemo ne- kaj možnih izboljšav in načrtov za prihodnost. Advanced robot systems have already substituted the human factor in many fields of work and we can only imagine what the future will bring. In order for a robot to be intelligent, to perform better at executing tasks than human, they need to be autonomous. We developed a robot for Field Robot Event competition, where it needs to be able to autonomously navigate through a corn field. In this thesis, we developed a robot application which is respon- sible for autonomous navigation with LIDAR sensor and proposed how to improve the navigation with camera recognition. The data collected with LIDAR usually contains a lot of noise, which we reduce with DBSCAN clus- tering algorithm. In conclusion, we present the results from the competition, DBSCAN algorithm, and camera recognition. We also propose a few im- provements and future plans.