Motion planning and control for selective harvesting of soft fruit

这个博士学位涉及水果收获机器人的规划和控制。该博士旨在探索最先进的数据驱动方法，结合运动规划和运动控制的分析方法，使机器人能够安全有效地移动，以推动（1）遮挡件，如草莓簇或叶子中的未成熟水果，以改善感知（例如水果检测和定位），以及未成熟水果达到成熟水果。传统机器人设置中的运动规划/控制涉及作为在机器人运动期间要满足的硬约束的碰撞避免。然而，在具有挑战性的情况下，如水果采摘或高度混乱的环境中，可行的无碰撞机器人运动可能是不可能的。因此，机器人需要与其环境进行交互，以改变场景中对象的配置。新的对象配置应允许闭塞自由的感知和碰撞自由达到采摘运动的机器人末端执行器。传统的运动规划和控制限制我们，以科普高度的非线性参与机器人和它的环境之间的相互作用。这个博士将研究使用新的数据驱动的方法进行交互式运动规划和选择性收获机器人的控制。博士将研究一系列可用的最先进的（SOTA）数据驱动的方法，并在SOTA方法上取得进展，并使其适应草莓采摘。 博士不仅将受益于Agriforwards CDT提供的培训，还将利用数据驱动方法的在线资源，例如深度NN，强化学习。本博士研究的风险包括（可能性0-3）：由于COVID 19（可能性2，影响2），由于大学实施的限制，对实验室的访问受限;缓解措施：在可能的情况下使用模拟计划工作的开发，并使用实验室设备验证和运行少量实验。由于暴露于COVID 19病毒而自我隔离（prob 2影响3）;缓解措施：学生可以使用模拟环境在家工作。错过收获季节（问题3影响3）;缓解措施：（一）在模拟环境中工作，（二）在实验室中使用玩具草莓，（三）在温室中工作，

This PhD is concerned with planning and control of a fruit harvesting robot. This PhD aims to explore state-of-the-art data-driven approaches combined with the analytical methods of motion planning and motion control allowing safe and efficient movements of the robot to push (1) occluding pieces, such as unripe fruits in a strawberry cluster or leaves, to improve the perception (e.g. fruit detection and localisation), and also unripe fruit to reach a ripe one. Motion planning/control in conventional robotic setup involve collision avoidance as hard constraints to be satisfied during robot movements. However, in challenging situations, such as fruit picking or highly cluttered environments, feasible collision-free robot motions may not be possible. Hence, the robot needs to interact with its environment to change the configuration of the objects in the scene. The new objects configuration should allow occlusion-free perception and collision-free reach-to-pick movements for the robot end-effector. Conventional motion planning and control limit us to cope with high nonlinearity involved in interactions between the robot and its environments. This PhD will investigate the use of novel data-driven approaches for interactive movement planning and control of the selective harvesting robot. The PhD will study a range of available state-of-the-art (SOTA) data-driven methods and make advancements on SOTA approaches and adapt them to strawberry picking. The PhD will not only benefit from the training provided by the Agriforwards CDT but also will exploit the available online resources on data-driven approaches, e.g. Deep NN, Reinforcement Learning. The Risks of this PhD study include (prob 0-3): Limited access to the lab due to restriction imposed by the University because of COVID19 (prob 2, impact 2); mitigation: plan the development of the work using simulation where possible and use lab equipment to validate and run a small number of experiments. Self-isolation because of being exposed to COVID19 virus (prob 2 impact 3); mitigation: the student can work from home using the simulation environment. Missing harvesting season (prob 3 impact 3); mitigation: (I) working with a simulation environment, (ii) working with toy strawberry in the lab and (iii) work in greenhouses provided by