Advanced techniques for autonomous navigation in precision agriculture

Medientyp: E-Book

Titel: Advanced techniques for autonomous navigation in precision agriculture

Beteiligte: Fleckenstein, Freya [Verfasser]; Burgard, Wolfram [Akademischer Betreuer]; Burgard, Wolfram [Sonstige]; Pradalier, Cédric [Sonstige]; Bödecker, Joschka [Sonstige]

Körperschaft: Albert-Ludwigs-Universität Freiburg, Autonome Intelligente Systeme ; Albert-Ludwigs-Universität Freiburg, Fakultät für Angewandte Wissenschaften

Erschienen: Freiburg: Universität, 2022

Umfang: Online-Ressource

Sprache: Englisch

DOI: 10.6094/UNIFR/231925

Identifikator:

Schlagwörter: Autonomer Roboter ; Navigation ; Bahnplanung ; Serviceroboter ; Lokalisierung ; Umweltmodell ; Autonomes Fahrzeug ; (local)doctoralThesis

Entstehung:

Hochschulschrift: Dissertation, Universität Freiburg, 2022

Anmerkungen:

Beschreibung: Abstract: A decrease in farmers and the rural exodus call for increased automation in agriculture to feed the growing world population. Furthermore, sustainable and efficient farming technologies are more important than ever to be able to produce a sufficient amount of food. Thus, agriculture is moving from uniform field treatment with large machines towards treatment of individual plants, also called precision farming, using autonomous robots. Two goals of precision farming are to reduce the use of chemicals and thus relieve the burden of chemicals on the environment, and to increase the crop yield by treating each plant as needed. For any robot to be able to perform a variety of tasks on the field autonomously, it requires a reliable and robust navigation approach for agricultural environments. This poses several challenges: the environment changes quickly and drastically in the context of agriculture due to varying weather conditions and fast plant growth. This makes it hard for a robot to reliably extract semantic information from sensor data. Furthermore, agricultural environments such as fields and greenhouses are quite narrow, and finding and executing safe and efficient paths there is difficult. In this thesis, we present building blocks for a navigation approach suitable for precision farming applications that tackle these challenges. Various tasks in the context of navigation on a field, such as localization or mapping, employ the position of crops or crop rows as a distinct feature and thus need a reliable detection thereof. Depending on the lighting conditions and the size of plants, either camera or lidar sensors are more suitable for extracting vegetation features. We provide a representation of vegetation features that has the potential to make crop and crop row detection methods independent of the input sensor modality by being able to capture information from both sensor types. We run a crop row detection on the resulting vegetation feature maps and develop a quality measure for the detections that is based on their feature support. This enables us to filter out unreliable detections in applications such as mapping, localization or weed detection based on crop rows. We perform experiments using a crop-row-based localization and show that we are able to increase the robustness and avoid localization pose divergence by applying our quality measure to filter out unreliable detections. In narrow environments such as fields and greenhouses, movement flexibility is important for navigating efficiently. We therefore consider two elements for path planning that impact the flexibility in motions. First, we factor in the configuration of robot joints that impact possible movement. Second, we take the ground clearance of a robot into account, which enables it to pass over low obstacles. We evaluate a path planning approach that makes use of joint configurations such as manipulator arm angles or adjustable wheel positions to increase the motion possibilities. Our evaluation includes the planning time and the path efficiency. Additionally, we investigate how a heuristic for planning with high ground clearance affects path planning performance. We perform experiments in simulated and real world environments and test the limits of this planning approach in a particularly challenging environment with many obstacles. The results show that the path planner is able to efficiently incorporate the joints into the planning problem. We also demonstrate the necessity of including these joint configurations, as reaching some planning goals takes considerably more planning time if the joints cannot be adjusted, or they may not be reachable at all. We also establish that the studied planning approach is well suited for real world application in agricultural environments. Lastly, path execution plays a crucial role in an efficient and safe navigation approach. We present a local planning method to efficiently execute a computed global path with a robot that has independently steerable omnidirectional wheels. An advantage of omnidirectional wheels is that they provide great motion flexibility. However, the wheel angles often have underlying constraints, such as a maximum steering velocity, limiting the speed at which the wheels are able to turn. If these constraints are not considered during motion command computation, the robot may diverge from the planned path or it needs to stop to adjust its wheel angles. We introduce a method to represent common steering constraints in a compact manner and show two techniques to integrate them into local planning, resulting in motion commands that satisfy all constraints. We demonstrate the efficiency of our approach in simulation and real world experiments. In the results, we see that path execution accuracy can be retained despite the additional constraints. At the same time, our approach is able to substantially reduce the path execution time. The methods presented in this thesis contribute towards a flexible, robust and efficient autonomous navigation approach for precision agriculture. We hope to hereby advance sustainable farming technologies that are critical to ensure sufficient food production for the growing world population

Zugangsstatus: Freier Zugang

Nur in Feld suchen:

Zuletzt gesuchte Begriffe: