Literature study on robotic tree harvesting and pruning : The Next Fruit 4.0 - Multifunctional robot deliverable 1

Abstract

This report contains three literature studies. In the literature study on pear detection methods the most relevant and recent publications on deep-learning based image analysis methods for fruit detection have been studied. Four software methods are popular: instance segmentation + 3D vertex normal estimation; extremely accurate instance segmentation; key-point detection and panoptic segmentation. The literature study robotic tree fruit harvesting groups the literature in methods for approaching the fruit/target, gripper requirements and manipulation patterns. For the literature study on machine vision systems for robotic pruning, a closer look is taken to literature describing the sensing system (mainly 2D or 3D cameras), data pre-processing (to create a high quality point clouds and to improve the accuracy) and key-point detection (to define pruning locations).