Successful farming is founded on control. Indeed, the task of growing crops is one of the world’s oldest control loops. Our ancestors’ realisation that plants could grow faster and produce greater yield with an optimal amount of water and nutrients – but no pests – sparked an arc of agricultural evolution that continues to this day.  

Closing the loop for true ploughing automation

In this article, the latest in a series that has already explored ploughing and planting, I want to unpack crop growing from a control engineering perspective – and identify some of the challenges and opportunities as we contemplate the coming generation of transformative farming autonomy. 
Let’s start with the basics. Plant growth can be controlled indirectly by regulating environmental conditions affecting photosynthesis and respiration. If farming is conducted indoors, it’s possible to control most of the environmental parameters. In an outdoor environment, on the other hand, only parameters such as irrigation, fertilisation and pest control can be regulated practically and with significant variation.  
Farmers act as controllers by processing input information and producing desired actions. The control aim is usually to maximise yield and speed up growing while preserving the condition of soil, so it remains fertile next season. Here’s a control diagram to show what that looks like: 

In this simple example, a controller – the farmer – knows that he can maximise yield if the amount of water is just right – not too much, not too little. He also removes insects and weeds if needed, increasingly using automated and in some cases, autonomous means. The desired signal then, is the optimal amount of water and no pests. The farmer observes the field and produces several types of control actions based on the deviation between the desired outcome and input information related to pests and water. The possible activities are pouring or removing water, extracting weeds or clearing away insects. 
With time, farmers learn not only how to react to the current state of the growing process, but even predict how particular action or lack of action will influence the system. In control theory, we would call them model-based predictive controllers. Model predictive control is often challenging to implement when it comes to controlling machines since most of the processes are very fast and characterised by complex, non-linear models with intrinsic stochasticity. When considering farming, the model of a field is also non-linear. Still, fortunately, the system response is characterised with a considerable delay (days or weeks) that allows for the proper reaction of a farmer. 
A good example of predictive control action is cultivation. Farmers know that preparing the land to grow crops will improve yield, so they do it often, many months before planting. A similar situation applies to a different kind of spraying. Based on the current state of crops, environmental conditions and predictive state, farmers spray fields. They provide water, fertilisers and pesticides to protect plants, accelerate growth and increase yield. The model-predictive control applied to a farm is explained in this diagram: 

Contemporary farm development 

Contemporary farming developed from a deeper understanding of optimal growing conditions and technological advances such as tractors and harvesting machines. The industrial revolution marked the first significant sea change as control action adoption became more efficient on a larger scale. Sensory advancement, driven by mass production and related cost savings, can be seen as the second revolution. Traditionally, farmers processed visual information to gain insight into the system. Now, they use sensors that can give them additional insight into soil and environmental conditions. Examples include metrological stations and soil sensors which measure humidity, pH and nutrients in the ground (NPK levels). 

The third revolution contributing to farming transformation is happening now. It is related to the development of controllers which can advise on specific control actions. The algorithm can process visual information acquired from cameras and suggest particular actions, such as spraying with herbicides or adding water, based on the deducted state. 

The diagram below shows a general control architecture for a well-organised farm in a developed country. Farmers can use specialistic tools to observe and gather data from the environment and computers to process and visualise information. They can then automate and scale-up control actions by using machines.   

As before, their intention is to control irrigation and pests, but it now extends to soil pH and fertilisation. In contemporary farming, workers can gather information by using drones, metrological stations and remote IoT soil sensors. What’s more, information can be presented in an easy to interpret form by combining environmental information with GPS data. In this diagram, we see a contemporary farm that uses IoT sensors, GPS data and a computer to process information. The control action is exerted by using machines: 

Introducing automation and autonomy  

It would be possible to minimise human involvement by building a highly automatic farm with autonomous control. The straightforward effect of automation is productivity growth due to optimisation of environmental conditions. At the same time, a product that comes from automatic farms is usually of better quality. Automated farming is also more environmentally friendly since the amount of water, heat and nutrients can be regulated more precisely. Water waste is lessened, and energy use optimised. Substituting humans with machines increases profitability by scaling up production while minimising the number of employees.  

Automatic indoor farms moved out of the realm of sci-fi a long time ago, when large numbers of vertical farms were established around the world. In this kind of Controlled Environment Agriculture (CEA), all key parameters can be controlled via an automatic system. Actuators are connected to computers responsible for data interpretation and control.  

But autonomy on large-scale outdoor farms remains a challenge, mainly because of an unstructured environment. For example, spraying could be done autonomously by using ground vehicles, drones, or via irrigation infrastructure, but most farms are not self-sustained yet. Farmers must still refill spray, initiate the process, and make timely decisions on what kind of action to apply. 

It would be possible to substitute the farmers’ evaluation process by developing a system which can process information, make decisions by using model related to specific crops and perform actions via a network of subcontractors. The system could order services autonomously – such as a report from a company specialising in aerial imaging footage, spraying or harvesting. Communication with the third party would be automated by adopting Robotics Process Automation (RPA) principles.  

This kind of a system would help a farmer to manage large-scale crop production and, in the future, it would be able to substitute a human. Subsequently, the role of the farmer would be to monitor the system and concentrate on activities that require human interventions, such as equipment maintenance and corrective actions after disasters. Here’s a control diagram showing a fully automatic farm with an autonomous controller: 

The challenges of change 

As I’ve explained, there are many advantages of adopting automation. But like most investments, automation comes with a price. The key challenge is related to business transformation. Introducing full-scale automation changes the character of a farmer’s work – labour and toil is replaced by management, planning and engineering. I would liken the task of automating a farm to expanding a company from a sole trader to a big corporation. 

Traditionally, farming involved various detailed tasks, like gathering and analysing information and exerting control actions – often with the help of other farmworkers. Introducing automation decreases the number of workers needed. But conversely, it brings on board a variety of new engineers – system, automation, software and others – to develop and maintain the system. If the scale of production is considerable, then people responsible for sales and distribution need to be added. Effectively, the farmer becomes a manager who keeps the team together while handling production at a high level.  

There are also technological challenges related to system-level design and control. There are a plethora of various solutions on the market, but they must be integrated adequately with other subsystems. The integration often requires an online database to store information, and a variety of cooperating programs to visualise the state and advise appropriate action.  

It is usually necessary to develop control algorithms from scratch since they should match installed hardware equipment and be tailored for a specific control process. Very good algorithms are not only able to control the process in everyday scenarios but can also deal with noise and unpredicted situations by raising an alarm or producing a novel reaction autonomously. 

Delivering return on investment 

As with each control process, several trade-offs should be considered when designing an automated farm. The critical aspect is the definition of a proper cost function that is subject to optimisation. The overall aim should be to maximise the yield while minimising the cost of operation and control effort. For example, it would be possible to observe the field very precisely. But that would increase the cost of the system and energy to a level that might push reimbursement back several years, so it is better to keep it at a practical level. We should keep sustainable development in mind. 

Want to learn more about closed loop control in growing?

To sum up, the system design should be considered in a way that takes all necessary factors into account. Automating a farm will always be associated with risk. But when it’s done correctly, return on investment can be higher, and come faster, than with other kinds of ventures. The key to lowering risk while increasing profitability is to cooperate with the right partner. One who can provide diligent service, help design and optimise the farm, and plan investment to maximise return within a given time scale. On that note, lease don’t hesitate to email if you’d like to discuss any aspects of the topic. And do please check out details of Cambridge Consultants’ approach to the wider world of agri-tech here.

Artur Gmerek
Principal Engineer

Artur is a robotics engineer with over 15 years of technical experience in mechatronics and robotics fields. He has always worked on all level of abstractions, such as system-level design and theoretical analysis, mechanical and electrical design, software and control. His focus is on bringing new ideas and projects to life to delight his clients.