The major agricultural machinery manufacturers are mostly oriented towards the development of autonomous tractors, machines that can emulate the conventional tractor without human intervention, and equipment that, although connected to a conventional tractor, is able to operate according to what it detects with a high degree of autonomy, i.e. so-called “robotic implements”.
Lorenzo Benvenuti
The technological vanguard in agriculture is moving rapidly and along very different paths. The most stimulating comes from small laboratories where researchers driven by the desire to create something new develop technologies that often have in common the ability to operate autonomously. These technologies we can call “agribot” or specialized robots for agriculture. These technologies are usually small, electrically driven machines specialized in one or a few services that almost always have to do with controlling pests and diseases in crops.
In contrast, the large tractor and agricultural machinery industry seems to be more interested in the development of autonomous tractors (again often driven by electric motors) and equipment with on-board robotic systems. This group, which is actually very heterogeneous, includes machines already on the market or soon to be introduced, developed to solve a specific problem, which, as in the case of agribots, is often linked to performing treatments on arboreal or herbaceous crops and, above all, to chemical or mechanical weeding, but also to sowing, transplanting, and harvesting produce for fresh consumption.
This equipment can be defined as robots (or containing robots) when it performs the action with a certain degree of decision-making autonomy due to its ability to process information according to artificial intelligence (AI) processes, to decide whether, how and with what intensity to perform the action itself.
Artificial Intelligence
In general, the robotic implement is a piece of equipment equipped with different types of sensors that mainly detect crop and agronomic parameters; it tends to operate in a geo-localized way and can use geo-referenced information, although it does not work only on the basis of prescription maps, it is equipped with processing systems capable of processing data in real time. A typical application is the intelligent sprayer that only carries out treatment when the conditions are right. A good example is the chemical weed killer sprayer that only sprays the surface when the presence of a weed is detected. In the most versatile equipment, pest recognition is implemented through artificial intelligence systems, and this allows the end user, after appropriate training, to implement this technology in different crop and pest environments. It is therefore autonomy, as opposed to automatism, that allows the term robot to be used properly, at least in its current meaning.
At this point, a question arises: what is the dividing line between artificial intelligence and ‘mere’ software? The question is not trivial and opens up interesting analyses, which we can ignore by settling for a simple and effective definition, useful, if nothing else, to define artificial intelligence that has a narrow range of abilities (Artificial Narrow Intelligent, ANI). We can consider the prerogative of this artificial intelligence those processes that allow the goal to be achieved from a known assumption without, however, predicting or knowing what actually happens in between. Suppose a robot is asked to recognize a face, a plant, an animal, and to do this, it is given a whole series of basic information and equipped with the appropriate tools. Suppose then that the robot achieves its goal by managing to distinguish the face, the plant, the animal among a thousand others. In this case, we know the starting point, we can evaluate the result achieved, but we do not know on which elements the program was based to give the expected answer, i.e. among the many evaluation elements available which ones were actually used by the program and according to which encoding.
In agriculture, AI is not only found on machines and equipment. For instance, in May 2023, Farmers Business Network (FBN) launched an AI-powered agronomic advisor named Norm. This advisor, based on Open Al’s Chatbot GPT-3.5, provides farmers with a wide array of agronomic information, assisting more than 55,000 members responsible for a total area of 40 million hectares in the US and Canada. Norm uses public data such as weather insights, soil monitoring, fertilizer and chemical usage for various crops, product label information, market prices, university research and farmer comments, as well as FBN proprietary data feeds. And Norm is not the only one. However, the question of whether these artificial experts provide truly reliable information has yet to be fully explored. Initial impressions suggest some expertise in offering generic advice, but they fail to tailor recommendations to individual farmers. Perhaps they appear more clever than intelligent at this stage.
Speed
There is another interesting difference between the small autonomous robot, the autonomous vehicle and the robotic implement and it concerns the speed of execution. In fact, in the former case, especially with highly specialized robots, the work is often carried out in full autonomy according to a schedule that can, at least in part, be dictated by the needs of the robot itself. These small automatons are made to work continuously, day and night, and carry out mechanical weeding, harvesting or other activities: their work capacity is essentially a problem of operating costs that can be resolved by identifying the break-even point for one’s own company between the work capacity of the individual automaton and its cost: is it better to have a few fast and expensive robots or many slower but cheaper ones?
For tractors and robotic equipment, on the other hand, the speed of work execution is conditioned by many factors, including the need to act promptly. All this requires high computing power and high reactivity on the part of the tools in order to ensure that no more than a few hundredths of a second elapses between the acquisition of the data by the sensor and the response of the actuator, which is in fact the time needed for an agricultural machine to travel the space between the point where the information is detected and the tool that is to perform the action. For example, with a speed of just one kilometer per hour, the time needed to move 2 m is 7.2 seconds; but at 5 km/h, it is reduced by 1.44 seconds; if the speed increases, the time available is reduced. There are therefore two bottlenecks: the speed of the processing system and the responsiveness of the tool. In the first case, research also tends to act on the algorithms in order to streamline processing, as well as on the power of the different components; in the second case, it tends to replace devices driven by micro-hydraulics or micro-pneumatics with more responsive electrical devices. Then it must be verified to what extent the tool can act quickly in interacting with the plant or the soil, managing to do a satisfactory agronomic job. Precisely in the case of mechanical weeding carried out on the row, the speed of the hoe acting on the soil between one plant and the next cannot enter and leave the row too quickly because it would cause the soil to be removed right next to the cultivated plant. This would obviously not only deprive the crop’s roots of the right cover, but in the case of leaf crops, such as lettuces and other head crops, it would cause them to become soiled with potential depreciation. Therefore, speed of execution is one of the stumbling blocks that cannot always be overcome and must be analyzed from all aspects.
Autonomous tractors
Autonomous tractors, capable of operating without human control, are for the time being excellent training grounds for manufacturers, who are thus acquiring fundamental knowledge for innovation and preparing themselves for the mechanization of tomorrow. However, the lack of legislations has so far discouraged the marketing of these machines, the use of which is currently relegated to experimental companies.
In fact, major tractor manufacturers such as John Deere, Case IH and Yanmar, or equipment manufacturers such as Kuhn, despite having successfully developed autonomous tractors, still do not plan to market them. This is because, although the technology is ready, legal constraints stemming from very incomplete legislation hinder the release of these autonomous wonders, especially in Europe. However, the promulgation of useful provisions to regulate the agricultural and other robotics sector could take place as early as this year. What is interesting is that it will not be a specific regulation for autonomous tractors and robots, but a real body of legislation aimed at regulating the field of machines, cybernetics, and data management. Let us mention some of the proposals for regulations pending at the European Commission, available on the web, such as Com/2021/202 which will basically renew the current Machinery Directive; Com/2021/26 on digital products containing AI; Com/2022/495 which also deal with legal responsibilities arising from failures and defects of AI-based machines and digital products.
With reference to autonomous tractors, there is a general consideration to be developed concerning the role they will be able to play and their actual usefulness. The impression is that the prototypes presented tend to reintroduce a very traditional model of agriculture without bringing process innovation. Compared to small robots or robotic equipment that are designed to solve specific cultivation problems and improve aspects of cultivation by reducing the use of substances, enhancing active ingredients, and proposing innovative agronomic activities, autonomous tractors are intended to replace the driver, replicating what is in fact performed by the conventional tractor. On the web it is easy to find pictures of these wonderful technologies pulling a plough or other equipment that is typical of a very conventional way of farming. Of course, there is much more inside an autonomous tractor than what meets the eye, but these are pluses that are already or will soon be implemented on tractors with drivers.
Undoubtedly more interesting are the robots and autonomous self-propelled vehicles proposed to manage the livestock within the farm center, by providing milking, creation and distribution of the ration, cleaning, and management of the manger. These, in fact, are time-consuming, burdensome, and repeated operations that are seamlessly throughout the year, which however require a certain amount of experience and adaptability. This is the sector where robotic development has found the greatest commercial acceptance to the extent that it can be defined as a rapidly expanding market.
What the future holds
The near future, in addition to a maturing of the current lines of development in agricultural robotics, could reserve for us the diffusion of new technological systems, intended to carry out specific crop operations, which will require artificial intelligence for their management, and which will be mounted on robotic implements or on autonomous robots. In fact, many see important synergies between robotic systems equipped with artificial intelligence systems and these technologies that we can define as emerging in their agronomic applications, such as lasers, microwaves, other electromagnetic waves, but also electrical energy or molecules with high efficacy and very low persistence. These are in fact potentially useful means for our sector that require management and control systems to be implemented in robotic devices capable of managing these technologies but also of recognizing the intervention zones within the cultivation environment.
In the longer term, we can also expect the development of robotic systems capable of improving the relationship between man, environment, and cultivation, as Soft Robotic seems to promise.
In fact, in the long term, we can envisage a growth of the interdisciplinary field that deals with robots made of soft, deformable materials that can interact with humans and their environment. So-called Soft Robotics is a new frontier in technological development because it represents an alternative way of approaching robotics by dismantling conventions and exploiting a new potential for producing an innovative generation of robots capable of supporting or replacing men in their most delicate interactions with the cultivation environment, such as, for example, in the harvesting of products for fresh consumption.
