global-orchard-vineyard-planner

Global Path Planning for Autonomous Vehicles in Orchard and Vineyards

Timo Schönegg†, Turcan Tuna†, Fan Yang†, Gabriel Waibel†, Matias Mattamala‡, and Marco Hutter†

†Robotics Systems Lab, ETH Zürich

‡Dynamic Robot Systems Group, University of Oxford

Supplementary Video

System Overview

The proposed global planning framework generates a global path for a given environment and vehicle. The planner consists of five components:

1. Map creation, 2. Topological graph extraction, 3. Mission user input, 4. Finding a high-level path, 5. Fine-tuning solution ensuring kinematic feasibility.

Starting from a 3D point cloud map of the environment (1), an occupancy grid map indicating the navigational space is created (2). Using the Voronoi diagram, the field's graph structure is extracted, covering the open space (3). Taking into account the row selection from the user input (4), the Rural Postman Problem is solved to find an Euler tour covering the selected rows to ensure that all necessary paths are covered efficiently (5). Finally, a hierarchical A* approach ensures feasibility given kinematic vehicle constraints (6).

Results

Results of the framework applied to an orchard in autumn (top images) and a vineyard in spring (bottom images).
Left images: The generated occupancy grid map Mog with the global path Pog where green points are required paths and blue points are optional paths.
Right images: The corresponding global path PENU. Red arrows are required paths, and blue arrows are not required paths.

Results of the framework applied to forest environment with (top images) and search and rescue environment (bottom images).
Left images: Occupancy grid map Mog with the global path Pog, required paths in red, optional paths in blue.
Right images: The global path PENU visualized in the global point cloud Mpc. For better visualization, the forest canopy has been removed from the point cloud.

Page updated

Report abuse