Position determination of a mobile robot in a precision agriculture scenario

The adoption of autonomous vehicles in agricultural scenarios can aspire to become a reality if the validation of their effectiveness is sustained by the contemporary and shared improvement of all those technological gaps identified by current research projects. In particular, a crucial point is still linked to the autonomy of robots, related to the concept of guidance, navigation and control. Using a combination of conceptually different sensing techniques and integrating the subsequent data, more accurate property estimates can be provided, leading to more robust management and increased adoptability of sensor-based crop management. For example, to properly locate and operate autonomous vehicles for in-field tasks, the knowledge of their instantaneous position needs to be combined with an accurate spatial description of their environment. In agricultural fields, especially when operating within crops, GPS data are not reliable nor always available, therefore high-precision maps are difficult to be obtained and exploited for in-field operations. Recently, low-complexity, georeferenced 3D maps have been proposed to reduce computationally demand without losing relevant crop shape information. This thesis focuses on the localization of an unmanned ground vehicle (UGV) which is moving between the rows of a vineyard performing farming operations. Due to the limited space where the autonomous vehicle is moving, an high precision filter needs to be designed to avoid collision with the crops, providing additional information on the vehicle's position and compensating the lack of accuracy of the GPS in the field. In particular, we propose an innovative approach that allows us to \textit{fuse} data collected by \textit{distance sensors} and the information obtained with the use of an a priori provided \textit{simplified map} to improve the estimation of the UGV location within crops. Indeed, starting with the distance measurements between the UGV and a crop row, measured by the ultrasound sensors mounted on the vehicle, it is possible to obtain a precise information on the UGV position by fusing it with the data coming from the georeferenced map, exploiting a trigonometric approach. This improved estimation of the UGV location can be integrated with additional data, merging it with those provided by other sensors as GPS and IMU, using classical filtering schemes. This result leads to a very precise estimation of the pose of the vehicle within the rows and allows to perform an efficient control action to make the UGV follows the designed trajectory avoiding collisions.