The increasing need of eliminating pesticides and substituting chemical-based weeding techniques with manual and mechanical methods have gave way to the development of agricultural robots that will greatly improve and reduce the time spent on growing more healthy plant fields. Because most crops are cultivated in rows, the development of robust and reliable algorithms for the detection of plant rows and individual plants is the foundation for autonomous navigation within the plant fields. A number of field robots have been developed by various research groups and companies during the past decades. Although they use expensive sensors for detecting rows, these methods lack a certain degree of robustness with regard to the variability of different fields. They are typically built with a specific project or purpose in mind and their design limits the possibility of using them for other purposes. Using big robot platforms that usually span several plant rows limits greatly the size of the rows and the size of the plants in many types of fields. This thesis proposes instead an algorithm that makes use of cheaper sensors and has a higher variability by combining row detection algorithms with graph-based localization methods as they are used in Simultaneous Localization and Mapping (SLAM). The considered method focuses on row detection in different types of fields and on using graph-based localization to improve individual plant detection and deal with exception handling, like row gaps, which are falsely detected as end of rows. Testing the developed algorithm in a variety of simulated fields shows that the additional information obtained from localization provides a boost in performance over methods that rely purely on perception to navigate. The framework built within the scope of this thesis can be further used to integrate data from additional sensors, with the goal of achieving even better results. This Master thesis achieves the goal of implementing a robust perception algorithm that uses a LiDAR sensor and graph-based localization techniques. The entire framework allows a small-sized robot to navigate autonomously inside tree nurseries populated by trees of different shapes and sizes.