High-throughput robotic phenotyping of fruit traits for automatic strawberry harvesting

英国的软果行业，总生产成本的50%用于劳动力，非常关注采摘劳动力的可用性和劳动力成本的通货膨胀。解决这些挑战的一个有希望的解决方案是开发水果采摘机器人，实现水果生产的自动化。进一步发展成功的机器人采摘系统的主要障碍是缺乏对哪种水果品种最适合这项任务的普遍理解。该博士项目建议开发新的自动化表型技术，部署在移动机器人上，提供高通量，高保真度的草莓品种指标，表明它们适合机器人收获。该研究将研究基于植物/果实几何（即3D）的技术，提供有关品种物候和外部果实和植物特征的特征。该方法将通过利用自主移动机器人来克服基于实验室的表型系统的局限性，从而能够快速识别该领域的多个性状。这一目标将通过以下目标来实现：验证现代机器人传感技术，直接在环境中对草莓植物/果实进行精确的3D传感；开发软件技术，使机器人能够从多个角度对环境进行精确的3D重建；发展获得单个水果和植物特征的技术，最重要的是它们的空间安排，如分离和分组，这将决定它们是否适合机器人采摘；将系统部署在移动机器人上，以实现快速和自动化的操作；验证现有草莓品种在机器人采摘中的使用，以及新品种种植者技术的适用性。项目成果：至少三个草莓品种的标注数据集；重建草莓环境的软件技术（即三维地图），测量单株成分特征和草莓植株和果实的空间排列；基于计算特征与业界专家比较的评估和验证结果；通过一个真实的机器人拾取实验验证了技术。本提案中概述的研究将使软果行业、机器人公司以及更广泛的学术界受益，为表型分析提供基础知识和实用解决方案，推动未来的浆果育种计划和新型机器人设计。或者，水果采摘机器人是解决软果行业面临的劳动力问题的一种很有前途的技术解决方案。进一步发展成功的机器人采摘系统的主要障碍是缺乏对哪种水果品种最适合这项任务的普遍理解。该博士项目建议开发新的自动化表型技术，在移动机器人上部署内场，提供高通量，高保真度的草莓品种指标，表明它们适合机器人收获。该研究将研究基于植物/果实几何（即3D）的技术，提供有关品种物候和外部果实和植物特征的特征。该方法将通过利用自主移动机器人来克服基于实验室的表型系统的局限性，从而能够快速识别该领域的多个性状。机器人表型分析的基础知识和实用解决方案将有利于软果行业、机器人公司和学术界，推动未来的浆果育种计划和新型机器人设计。

The UK's soft fruit industry, with 50% of the total production cost dedicated for labour, is extremely concerned with both the availability of picking labour and labour cost inflation. One of the promising solutions addressing these challenges is the development of fruit-picking robots enabling automation of the fruit production. The main obstacle for further development of successful robotic picking systems is a lack of general understanding which fruit variety is the most suitable for the task.This PhD project proposes to develop new automated phenotyping techniques deployed infield on a mobile robot providing high-throughput, high-fidelity indicators of strawberry varieties indicating their suitability for robotic harvest. The study will investigate techniques based on plant/fruit geometry (i.e. 3D) providing traits about the phenology of the variety and external fruit and plant characteristics. The approach will overcome the limitations of the laboratory-based phenotyping systems by exploiting an autonomous mobile robot to enable rapid identification of multiple traits in the field. This aim will be addressed through the following objectives:Validation of the modern robotic sensing technology for precise 3D sensing of the strawberry plants/fruit directly in the environment;Development of software techniques enabling the robot a precise 3D reconstruction of the environment from multiple views;Development of techniques which derive traits of individual fruit and plants and most importantly their spatial arrangements such as separation and groupings which will determine their suitability for the robotic picking;Deployment of the system on a mobile robot for rapid and automated operation;Validation of the existing strawberry varieties for their use in robotic picking and suitability of the techniques for the growers of new varieties.The deliverables of the project:Annotated data sets of at least three strawberry varieties;Software techniques for reconstructing the strawberry environment (i.e. 3D map), measuring traits of individual plant components and spatial arrangement of strawberry plants and fruit;Evaluation and validation results based on comparisons of calculated traits against the human experts from the industry;Validation of techniques through the experiments with a real robotic picker.The research outlined in this proposal will benefit the soft fruit industry, robotics companies but also the wider academic community with the fundamental knowledge and practical solutions to phenotyping driving future berry breeding programs and novel robot designs.Or Fruit-picking robots are a promising technological solution to the labour problems faced by the soft fruit industry. The main obstacle for further development of successful robotic picking systems is a lack of general understanding which fruit variety is the most suitable for the task. Project ScopeThis PhD project proposes to develop new automated phenotyping techniques deployed infield on a mobile robot providing high-throughput, high-fidelity indicators of strawberry varieties indicating their suitability for robotic harvest. The study will investigate techniques based on plant/fruit geometry (i.e. 3D) providing traits about the phenology of the variety and external fruit and plant characteristics. The approach will overcome the limitations of the laboratory-based phenotyping systems by exploiting an autonomous mobile robot to enable rapid identification of multiple traits in the field. The fundamental knowledge and practical solutions to robotic phenotyping will benefit the soft fruit industry, robotics companies and academia, driving future berry breeding programs and novel robot designs.