Autonomous monitoring and control of crop growth as a feedback system

拟议的项目是在农业环境中模拟和控制作物的生长作为一个反馈系统。该项目的目标是通过利用分布式传感来优化肥料的使用和作物管理的自动化，使种植者能够最大限度地提高收成。具体而言，该项目将通过以下工作包进行开发。WP 1：生菜生长的动态模型作为反馈系统。WP 2：作物优化的反馈控制算法。WP 3：用于作物估计的分布式传感技术。WP 4：用于生长控制的机器人技术和通用自动化。所有的工作包将与G的种植者合作开发。原型和测试也将在剑桥大学人机协作观测站的Agripods控制环境中开发。WP 1将在第一年内实现。选择一个合适的控制模式将需要了解所涉及的作物的要求。具体而言，WP 1将开发开放的动态模型，其输入将在以后用于控制目的。WP 1需要收集和处理来自种植者作物的数据，以建立和测试模型。WP 2的目标是开发反馈控制算法来优化作物生长。这些算法必须考虑到资源的可用性，并且必须对不确定性具有鲁棒性。到第二年年底将完成一项完整的研究。WP 3将与WP 1和WP 2同时开发。第一个原型的生菜重量的非破坏性估计已经运作。该原型目前正在进行校准和总体设计修订。WP 3将利用G的种植者目前使用的数据和传感技术（第一年）。WP 3还将研究使用移动的机器人进行数据收集，特别关注现有的低成本技术（第2年）。最后，WP 3和WP 1将被整合。目标是利用动态模型来改善传感（故障检测、降噪）。同样，分布式传感将用于改进和校正模型预测（第3年）WP 4将研究驱动技术和通用机器人技术，以开发用于生长控制（肥料分配等）的平台。WP 4将从G的种植者现有技术的定制开始，但也将考虑基于移动的机器人技术的新技术。WP 4取决于其他工作包的完成。WP 4将在第三年完成。该项目的主要挑战包括：WP 1中复杂的系统动力学; WP 2中的大不确定性; WP 3中传感器数据的采样率和整体质量的限制; WP 4中实现高度自治的故障检测和困难。该项目的一个主要风险是依赖于种植作物，这受到季节性和该过程固有的缓慢性质的限制。风险缓解措施包括在人机协作观察员中，在Agripods的受控环境中进行可行性研究。这种室内控制设置不受季节性的限制。它允许通过模拟真实的环境条件来验证生长模型和控制算法。它还允许测试原型的传感和驱动。

The proposed project is to model and control the growth of crops in an agricultural setting as a feedback system. The project goal is to enable growers to maximise their harvest, by taking advantage of distributed sensing to optimise the use of fertilisers and of automation for crop management. Specifically, the project will be developed through the following work packages. WP1: Dynamic modelling of lettuce growth as a feedback system. WP2: Feedback control algorithms for crop optimization. WP3: Distributed sensing technologies for crop estimation. WP4: Robotics and general automation for growth control. All work packages will be developed in collaboration with G's growers. Prototypes and testing will be also developed in the controlled setting of the Agripods available in the Observatory for Human-Machine Collaboration at the University of Cambridge. WP1 will be achieved within the first year. Choosing a suitable model for control will require an understanding of the requirements of the crops involved. Specifically, WP1 will develop open dynamic models whose inputs will be later used for control purposes. WP1 requires collecting and processing data from the grower's crops to build and test the model. WP2 goal is to develop feedback control algorithms to optimize crop growth. These algorithms must take into account availability of resources and must be robust to uncertainties. A complete study will be developed by the end of the second year. WP3 will be developed concurrently to WP1 and WP2. A first prototype for non-destructive estimation of lettuce weight is already operative. The prototype is currently undergoing calibration and general design revision. WP3 will take advantage of data and sensing technologies currently used by G's growers (Year 1). WP3 will also investigate the use of mobile robots for data gathering, with particular attention to available low-cost technologies (Year 2). Finally, WP3 and WP1 will be integrated. The goal is to take advantage of the dynamic model to improve sensing (fault detection, noise reduction). Likewise, distributed sensing will be used to refine and correct model predictions (Year 3) WP4 will investigate actuation technologies and general robotics to develop a platform for growth control (fertilizer distribution, etc.). WP4 will start from the customization of available technologies at G's growers but will also consider new technologies based on mobile robotics. WP4 depends on the completion of the other work packages. WP4 will be completed in Year 3. Major challenges for the project include: complex system dynamics in WP1; large uncertainties in WP2; limitations to the sampling rate and of the overall quality of sensor data in WP3; fault detection and difficulties in achieving high level of autonomy in WP4. A major risk for the project is the dependence on growing the crop, which is limited by seasonality and by the intrinsic slow nature of the process. Risk mitigation include creating feasibility studies in the controlled environment of the Agripods, within the Observer for Human-Machine Collaboration. This indoor controlled setting is not limited by seasonality. It allows to validate growth models and control algorithms, by simulating realistic environmental conditions. It also allows testing of prototypes for sensing and actuation.