An optimization based planner for autonomous navigation in vineyards

Abstract

Autonomous driving systems found their first applications in the agricultural field, being a way to ease personnel of repetitive jobs and increase precision. Performing operations like harvesting or pruning requires high positioning accuracy, especially in structured environments like vineyards and orchards. In these contexts, the global reference path is dictated by the agricultural procedure to perform. The continuously-changing vegetation and reduced maneuvering space create the need to re-plan the vehicle route with respect to the global reference. Hence, the importance of local planning. This paper proposes a local planning strategy with the objective to follow a park-to-park global path while avoiding obstacles. We formulate the local planning task as a constrained optimization problem. The resulting local plans are not constrained in shape, thus guaranteeing planning freedom, and manage obstacle avoidance in an innovative way. The collision area is precisely determined taking both the vehicle and the obstacles dimension into account, and considering the vehicle approach direction. The proposed system is tested in simulation, where its performance are compared with a benchmark planner. An experimental campaign validates the local planner with satisfactory results.