Baumlinien- und Spurzentrumsschätzung unter Verwendung Neuronaler Netzwerke zur Führung eines Autonomen Arbeitsfahrzeugs

Details

Veröffentlichungs-Nr.

EP4744468

Anmeldetag

11. November 2025

Veröffentlichung

20. Mai 2026

Rechtsraum

EP

Abstract

Systems and techniques are disclosed for autonomously navigating a work vehicle within rows of trees or similar environments. A range sensor captures depth and intensity data representing the scene ahead. A trained segmentation model classifies points of the scene into features such as trunks, canopy, ground, and obstacles. Trunk features are clustered into centroids on the left and right sides of the vehicle, and a line estimation process, informed by vehicle odometry and prior row spacing, generates candidate tree lines. Weighted evaluation and resampling produce estimated left and right tree lines, from which a lane centerline is computed. The vehicle's control system uses the lane centerline to determine lateral and heading corrections and to generate steering and speed commands for navigation. The approach continuously updates the lane geometry as new sensor data is received, supporting robust, real-time operation through orchards or other structured field environments.

Anmelder

Firma

Deere & Company

Land

🇺🇸 USA