Field robot market lacks direction and collaboration

The technology is fascinating: a robot that removes volunteer potatoes and weeds from a plot of beet or onions without any human intervention. Lightweight robots require no labour, exert extremely low pressure on the soil, and are able to work accurately 24 hours a day.

And while building a high-tech solution onto a 45-metre spray boom is prohibitively expensive, it is perhaps feasible on a robot with a 2-metre spray boom. Start-ups and robot firms are seeing opportunities in the global billion-euro crop protection industry.

_{Text continues underneath image}

Robot manufacturers too often view their robots as just a tool carrier, where farmers expect a complete solution for a certain problem. This makes introduction of robots to the market difficult. – Photo: Matthijs Verhagen

Development of field robots is a lengthy process

In practice, development is a lengthy process and there are some failures. The list of robotics projects that never or have not yet made it into practice is long. As soon as the innovation grants are used up, the free market proves extremely tough. For a start-up without equity, there is nothing in sight but the financial chasm.

In the meantime, crop farmers are very sceptical in view of the coming and going of robot manufacturers; after all, if you invest, you want to be sure that the supplier will still exist in five years time. This has also created a negative perception of businesses that do have a sound concept.

Amazone Bonirob field robot

The extreme difficulty of developing a field robot is evident in a company such as Amazone. Ten years ago, the manufacturer presented its Bonirob field robot. Amazone has since moved the management of the project to a start-up operating under the renowned Bosch brand. Even so, the robot is still not available on the market.

This is characteristic of a market in which robot projects fail to reach their predicted point of market launch, time and again. This makes the market very difficult to predict.

The fact that field robots never get off the ground contrasts with the success enjoyed with robots in other sectors. Consider, for example, the milking robot or the packing robot used by egg farmers, but also robotic mowers on lawns, welding robots in industry, and robots in logistics.

_{Text continues underneath image}

Arable farmers certainly are interested in robots. Labour is expensive, and the use of chemicals is under scrutiny, making spot spraying attractive. – Photo: Matthijs Verhagen

Robots broke out of the realm of science fiction many years ago. However, a field robot is trickier to develop because it works outdoors under conditions that vary each time, and that creates complexity. Technically, a robot is never allowed to work outdoors on freely accessible plots without human supervision.

A robot is a means to an end

Another factor standing in the way of success is that robot manufacturers offer no complete solutions; they often regard their machines as tool carriers. The prevailing belief among some manufacturers is that, after the sale, a crop farmer will attach a machine to the robot himself, and it should then work. A rather theoretical approach, as it is not that simple in practice.

A robot may be a means, but the farmer wants a complete solution

Crop farmers look for a solution to a specific cultivation-related problem, rather than a robot that will replace the tractor. A robot may be a means, but the farmer wants a complete solution. The most commonly requested solution is for volunteer crop removal and mechanical weeding, including from contemporary crop farmers, who want to reduce the use of crop protection products in onions and avoid spray damage to the crop, for example.

_{Text continues underneath image}

Camera technique on a self propelled sprayer. For a robot, typically with a limited working width, such high-tech is reasonably affordable. The small robot manufacturers however lack the development power. – Photo: Matthijs Verhagen

Little collaboration in the market

So far, robot and tool manufacturers have been working in isolation. There has been little collaboration in the market, and it is partly for this reason that robot builders do not always offer the market-ready, complete solution that crop farmers are after.

Things are beginning to change in that respect, however. Where the manufacturers fail to act, local parties take the initiative. Mechanisation firms have knowledge of electronics, tools, and the local market.

Artificial intelligence becoming a game-changer

Some start-ups and manufacturers are now opting for an approach based on a specific application. Hoe manufacturer Carré will be launching its own hoeing robot onto the market in 2020, for example. It is built for the purpose of hoeing, with the advantage of a familiar name accompanying the product.

Start-ups such as the Swiss firm Ecorobotix are developing robots with a single purpose, specifically identifying and dabbing the leaves of weeds and volunteer crops, and companies such as BASF have invested millions in them. The Ecorobotix weighs just 130 kilos and has a 2-metre working width. This means a single robot could cover 10 hectares per week.

Field robot no longer a self-propelled cart

According to reports, the robot would definitely become available on the market this spring. “It will now be 2020 or 2021,” an employee now tells us over the phone. This is typical of the market. Even so, these types of development signal a sea change. The field robot is now no longer a self-propelled cart, but rather a machine that is able to do something.

_{Text continues underneath image}

A small robot weighs less and poses less or no danger to people, even without extra security. Safety is now a problem with larger robots. In particular, the detection of people between crops. – Photo: Koos Groenewold

The crucial factor is that hyper-intelligent software is all-dominating, and this high-tech software is self-learning. According to market experts, this is the piece of the puzzle that has been absent from robot technology for the last 10 years. The robot or the tool itself is able to learn new things using images captured by cameras, data, and algorithms.

Deep learning

This is also referred to as deep learning, and is similar to the technology also found in self driving cars. It enables the robot to learn to assess unexpected situations posed by the environment and to learn on the basis of data. It learns to identify a weed or a volunteer potato plant between crops, and to kill it, of course. Deep learning is a form of artificial intelligence, or AI for short.

Tractor manufacturers play tactically

For as long as the market remains inaccessible, the tractor manufacturers will keep to themselves, and work on roboticising their tractors in piecemeal fashion. Only Fendt once presented a small field robot, but the product’s status is unclear. Perhaps it serves as a warning to field robot manufacturers?

Tractor manufacturers are playing a power game: remember that, with RTK-GPS ex-works, the ability to automatically turn on the headland, and technology that enables the tool to act on the tractor, state-of-the-art tractors are already not far from autonomous.

Full-liner companies have the capital, the dealer network, and soon also compact (electric) tractors that they will be able to roboticise

The Dutch company Precision Makers already demonstrated in the past that it is indeed possible to completely roboticise a tractor; the technology exists. Full-liner companies have the capital, the dealer network, and soon also compact (electric) tractors that they will be able to roboticise. No universal field robot will be able to compete with them.

_{Text continues underneath image}

The DLG-Feldtage in 2010. At the time everybody expected field robots to be common good by now. However, hardly anything happened in the fields. Arable farms seldom use one. – Photo: Henk Riswick

John Deere bought Blue River Technology

Bear in mind that they not only have the tools in-house, they also have both the platform and the opportunities to invest in deep learning software. The starting shot has been fired: last year, John Deere bought Blue River Technology, a start-up specialising in weed identification using AI in combination with a robot.

The robot is now nowhere to be seen, but the website now features photos of the smart spraying technology behind a John Deere tractor, and therein lies the key factor. To cultivate more effectively, it is not strictly robots that we need, but rather smart tools. Whether they will be attached behind an autonomous tractor or an autonomous robot in the future is of no particular interest.

It is all about the tool of the future. Robot manufacturers that fail to work on the basis of that practice and that realisation merely form part of fleeting hype, not a long-term trend towards greater autonomy.

Tijmen Bakker is owner of Tyker Technologies, that focusses on robotising in the building industry. He got his PhD in 2008 at Wageningen University in the field of field robots. – Photo: Henk Riswick