Self-driving machines in agriculture: Autonomously through the gray area

Contents

In a sense, autonomous driving in agriculture is old hat: for a quarter of a century now, tractors have been automatically following pre-defined lanes thanks to GPS and RTK (Real-Time Kinematic Positioning, a method with reference stations on the ground that achieves accuracy up to 2 cm). The driver can concentrate on other tasks, such as monitoring the surroundings and the attached implement.

Now agriculture is facing the next step: numerous manufacturers have been working on autonomous vehicles for a long time, which generally operate without a driver. The range extends from field robots like the Fendt Xaver GT to autonomous vehicles and upgrade kits that turn older tractors into autonomous tractors. Many of these products are still in the prototype stage, but some are already on the market.

The demand for such systems is driven by the shortage of skilled workers: farmers worldwide are increasingly having problems finding qualified drivers, as can be heard at many stands at the Agritechnica agricultural trade fair in Hanover.

However, a tour of the trade fair also shows: autonomous driving in the field may be less complex than in road traffic, but it is by no means simple. There are still numerous challenges to overcome here. This not only concerns technical but also legal issues.

Who is liable if something goes wrong?

US manufacturer John Deere, for example, is showcasing an upgrade kit for tractors at Agritechnica in Hanover, which is already being tested by farmers in corn and soybean cultivation in the USA. "Our German customers are also asking about it," says Michael Müller, Marketing Manager for John Deere's large tractors, in an interview with c't. However, he cannot give a date for a market launch in Germany, as the legal situation here is not yet clear enough.

Müller cites liability as an example: politicians still need to decide who is liable under what circumstances in the event of an accident with an autonomous tractor. The state government of Lower Saxony came to a similar conclusion two years ago: "Issues of safety and liability in the use of autonomous agricultural machinery are still under clarification," it states in a response to a parliamentary inquiry. And IT lawyer Klaus Gennen writes in an analysis on the use of AI systems in agriculture that German liability law is "not yet sufficiently adapted to the technical peculiarities, especially of autonomous AI systems." While the Road Traffic Act in Germany already contains regulations for "motor vehicles with highly or fully automated driving functions," such regulations are missing for agricultural machinery that operates in the field. They fall under the scope of the EU Machinery Directive.

But data protection is also a challenge, says Michael Müller from John Deere at Agritechnica. This is because John Deere transmits live images from the cameras of its autonomous tractors to a security center in the USA. There, human experts check the images if the tractor's AI has detected something in the vehicle's surroundings that looks like a person. The tractor then stops immediately and is only allowed to continue driving when two employees have independently given the green light. The transmission of live images with potentially personal data is easier to implement in the USA from a data protection perspective than in Europe, and even within Europe there are differences, says Müller.

Dust makes it difficult for AI to see

Müller cites dust as an example of the technical challenges encountered in agriculture. Large amounts of dust can be whirled up during field work, which can obscure the view of the upgrade kit's total of 16 cameras, which turns 5 to 6-year-old tractors into autonomous machines, and thus the AI. Radar technology would help, but is still too expensive for economic use, says Müller. John Deere does not disclose the price of the upgrade kit. It is understood from industry circles that some manufacturers are also considering subscription models instead of one-time purchase costs.

However, monitoring the surroundings is not the only task of the driver, and therefore not the future task of the AI. Implements such as cultivators, hoes, or seed drills must also be constantly monitored. Has a tine broken off? Is a branch being dragged along, destroying valuable plants? Is a tube clogged, preventing the machine from sowing?

Considering how many machines interact in agriculture and how closely they are adapted to different crop types, it becomes clear: there is still a lot to develop. John Deere's goal is to develop a fully automated production system for soybeans by 2030, including autonomous cultivators, sprayers, combine harvesters, and transfer wagons, says Müller.

"The operator doesn't have to intervene even once"

Not only corporations like John Deere or Claas are working on such challenges, but also numerous start-ups and medium-sized companies. For example, Digital Workbench GmbH, located near Ingolstadt, is showcasing its autonomous "multi-carrier platforms" at Agritechnica. These are GPS-RTK-controlled, and safety is ensured by bumpers and geofencing, among other things. They are also testing radar and lidar sensors in prototypes, reports sales manager Bernhard Limbrunner in an interview with c't.

Limbrunner names precision as an advantage of such specialized systems over tractors with upgrade kits: due to the nature of the three-point hitch on a tractor, implements have relatively a lot of play. With a rigid attachment, as used by Digital Workbench, it is possible to work much closer to the crop when controlling weeds with a hoe.

The regulatory situation for autonomous field work is "a gray area," says Limbrunner. However, like many competitors, Digital Workbench does not intend to be stopped by this. The first systems are to be delivered next year, and they have already received initial orders, says Limbrunner. The technology is mature: "When we sow sugar beets for 20 hours in our tests, an operator is still standing next to it, but they don't have to intervene even once."

Empfohlener redaktioneller Inhalt

Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.

YouTube-Video immer laden YouTube-Video jetzt laden

Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.

The German manufacturer Beinlich, a specialist in irrigation machines and sprayers - and also sensor technology for detecting anomalies of all kinds - seems to have fewer concerns about legal uncertainties. At the trade fair, the family company is exhibiting its RainRover drum irrigation machine, which is currently only a prototype and is now to be delivered to initial test customers. It is based on the manufacturer's proven technology, expanded with a crawler track and control for autonomous operation. The large drum holds up to 750 meters of PE pipe, connected to a centrally located hydrant. The farmer exports the geocoordinates of the field and lanes, which he has already measured with his tractor, and imports them into the RainRover's control system. It can then automatically drive the lanes and lay out the pipe on the way out, and roll it up again on the way back. Because the entire journey, unlike an irrigation machine attached to a tractor, does not cost any working time, the automatic device can travel over every part of the field twice and thus work with half the amount of water - which, according to the manufacturer, is better for the plants.

Beinlich has not equipped its RainRover with extensive sensors for environmental monitoring, radar, lidar, AI, or 360-degree cameras. However, unlike an autonomous tractor, the irrigation machine cannot turn the wrong way under the most unfavorable circumstances and drive for kilometers across foreign fields - after all, it is always connected to the PE pipe.

The trend towards autonomy is also evident in the sensitive areas of agriculture: a pair of humanoid robot arms is intended to automate the vegetable cultivation process for the German company Ai.Land in the future. Using VR goggles, people can control the robot arms and hands, collecting training data for full automation. However, the field robot system, which received an award at Agrictechnica, is still a prototype.

(cwo)