Autonomous agricultural vehicle design for variable rate spraying at orchards

In recent years, agricultural producers have faced significant challenges due to the uncertainty of access to the agricultural labor force, the growing demand for safe, accessible and high-quality agricultural products, a high level of competition with international producers, and an increasing need to reduce their carbon footprint. The fact that producers can continue their competitive and profitable production only by reducing labor costs means they have to invest in technology and increase efficiency. Autonomous agricultural vehicles are important in autonomous processes in fruit orchards, increasing productivity, collecting necessary data for decisions on orchard management, reducing operating costs and carbon footprint. In this study, autonomous vehicle design and simulation were done in order to carry the variable level spraying system in the orchards. An autonomous vehicle can map the orchards by using the data from the odometry and LIDAR sensors, then accurately determine the position of the autonomous vehicle on the map by using the adaptive Monte Carlo localization method to compare data from the LIDAR and odometry sensors with the map. An autonomous vehicle can avoid obstacles by using the dynamic window approach algorithm. The designed autonomous vehicle is able to give commands for starting and stopping the spraying machine to ensure that the variable-level spraying system is sprayed only within the desired locations. The autonomous vehicle designed in the study is an original study because it will work in covered orchards where GPS cannot work properly. It is also a fully autonomous tool because it does not need any data, such as external GPS data. It is expected that higher efficiency will be achieved by providing less environmental pollution, lower operating expense and less labor force in practice. The success of the mapping application depends on the map being updated after the position displacement and the number of particles used in the mapping. When the map is updated with the 50 cm displacement of the robot and 30 particles are used, 3,03 entropy value was obtained which are the most suitable for the required processing power and performance. The localization is achieved by comparing the robot sensors' data and map information through an adaptive Monte Carlo localization algorithm. The localization success depends on a position update after the amount of displacement of the robot and the number of particles used in the localization. When the position update was made with the displacement of the robot by 2 cm and the minimum 500, maximum 2000 particles were used, a mean error of 3,50 cm was obtained which was optimal in terms of the required processing power and performance. The path planning application has been developed in order to reach the desired point from the location of the robot, which is localized on the map. Dijkstra algorithm was used for path planning, planning was done in two stages as global and local planning. The dynamic window approach used in local planning allows the robot to escape obstacles.