Navigation of an autonomous robot in a simulated corn field

Večanje človeške populacije in z njo potrebe po hrani predstavlja izziv za svetovno kmetijstvo. K večji proizvodnji pridelkov lahko veliko prispeva kmetijska robotika, ki bi v prihodnosti povečala donos, trajnost in dostop do pridelkov ter hkrati zmanjšala uporabo pesticidov. V tej nalogi smo se osredotočili na enega od mnogih izzivov preciznega kmetijstva in sicer na avtonomno navigacijo po kmetijski površini. Razvili smo navigacijski algoritem, ki je sposoben uspešno navigirati majhnega mobilnega robota po simuliranem polju koruze. Za načrtovanje poti, ki robota vodi po sredini vrst in iz ene vrste v drugo, smo uporabili meritve iz 2D LiDAR senzorja. Te podatke smo obdelali s postopkom rojenja DBSCAN in SLAM algoritmom. Robustnost navigacijskega algoritma smo preverili na poljih z različno razgibanostjo tal in gostoto posejanih rastlin. Pokazali smo, da kljub temu, da večja razgibanost tal vpliva na kvaliteto vodenja in delovanja SLAM algoritma, ima na kvaliteto navigacije bistveno večji vpliv gostota posejanosti rastlin. Naš algoritem je torej sposoben navigirati majhnega mobilnega robota po simuliranem polju koruze z razgibanim terenom, dokler je to dovolj gosto posejano.The increasing number of human population and its growing need for food presents a great challenge for global farming. In the future, agricultural robotics can greatly contribute to increased production of crops, which would increase yield, sustainability and access to crops and at the same time reduce the use of pesticides. In this work we focus on one of the many challenges of precision farming, namely autonomous navigation in an agricultural environment. We developed a navigation algorithm that is capable of successfully navigating a small agricultural robot in a simulated maize field. We used measurements from a 2D LiDAR sensor to plan a path that guides the robot through the field rows and from one row to another. These measurements were processed using the clustering algorithm DBSCAN and a SLAM algorithm. The robustness of the navigation algorithm was tested on simulated fields with different levels of terrain roughness and plant density. In our tests we show that higher levels of terrain roughness affect the quality of the robot driving and SLAM algorithm, however lowering the plant density has the largest effect on the quality of the navigation. We conclude that our algorithm is capable of navigating a small mobile robot in a simulated maize field as long as the field is sufficiently densely planted