ROSANA - RObust SemAntic NAvigation in orchards

Many industrial sectors (e.g. automotive) have undergone strong automation in the past decades. However, within the agricultural sector and more specifically for the fruit industry, such automation is still lagging behind due to the cost and limitations of existing automation solutions. The decreasing availability of seasonal workers and increasing cost of manual labour threatens the profitability of the Belgian fruit sector. This PhD aims to support this sector by developing affordable yet robust mobile navigation techniques in the challenging outdoor setting of orchards. The overall goal of this project is to make several step changes towards a fully autonomous mobile robot capable of robustly navigating outdoors despite changing illumination, seasonal and weather conditions. The platform will be able to position itself and navigate both globally in the orchard and relatively to trees or branches. To achieve this, the project pursues three main objectives: (1) robust outdoor sensing and dataset construction; these sensors and datasets will be used in the next two objectives; (2) lifelong localisation and mapping with respect to trees and orchards using a semantic, adaptable map of the orchard; (3) outdoor semantic navigation by adopting semantic knowledge about recognised objects when planning robot trajectories, and by storing the effects of robot control parameters into the map.