A drone that learns to efficiently find objects in agricultural fields: from simulation to the real world

 Conference Spring 2025 

Rick van Essen, Gert Kootstra

Introduction

Drones are promising for data collection in agriculture, but their battery capacity limits the area covered in a single flight. Some dominant weed species grow in spatially distinct patterns in the field instead of randomly. Reinforcement Learning (RL) can be used to plan paths to find objects, such as weeds, in a field by exploiting these spatial object distributions. We propose to train an RL agent in an abstract simulation environment and apply it in the real world by combining it with a detection network.

Open the poster

Open the abstract