Agriculture faces a range of demands from several directions at once: farmers are expected to produce more food, use fewer chemicals, cut fuel consumption, protect the soil and biodiversity, and manage with a small workforce. These demands complicate the picture, but they also make it clear that agriculture needs more precise and more sustainable ways of working.
Not just a distant idea
Autonomous robots are increasingly seen as part of the solution, but safety has to come first. Before we ask what they are capable of, there is a more important question: “How can we be confident that they will work safely in real fields, in the presence of real people and in real conditions?”
This question matters, because agricultural robots are no longer just a distant idea. They are being built for sowing, weeding, preventing crop disease and harvesting. The NMBU robotics group, for example, has developed the Thorvald platform – a modular field robot designed for a variety of agricultural tasks. What makes this invention significant is both that the platform is autonomous and the fact that it points towards a different approach to fieldwork.
Safety first
Rather than relying solely on large, heavy machines, agriculture can move towards lighter and more flexible systems that work with greater precision, reduce wasted effort and place less strain on both workers and the soil. That is the potential of agricultural robotics, and potential alone is not enough. If these systems are to be used in vineyards, orchards and open fields, safety must come first.
Why do we use robots in agriculture?
Robotics has become indispensable in agriculture for practical reasons, because, as noted above, a great deal is asked of farmers. Automation is a matter of precision, timing and sustainability. The use of autonomous vehicles offers useful experience. Those vehicles have shown how convincing autonomous driving can be, and how fragile it becomes in unusual, chaotic situations.
Engineers sometimes call this the “long tail” – an enormous number of situations that are individually rare but collectively unavoidable. The layout of roadworks, glare from the sun, snow covering road markings, or a child running out from behind a parked car – all of this may be easy for an ordinary driver to handle, yet difficult for an autonomous system. Acting safely depends on perception, prediction and planning under uncertainty.
The same insight applies to agriculture. A field robot may perform well in a “clean” demonstration environment, but run into difficulty when its surroundings become chaotic. The changes take many forms – shadows shift, dust rises, plants lean into the path, rain turns the soil surface to mud and deep ruts form along the route. What seemed safe a moment ago may no longer be so. In agriculture, change is not the exception but the normal state of affairs.
Risks and complications in the field. Why do they differ from the laboratory?
Field conditions are one reason why farming can be more complicated than it appears from the outside. A robot in the laboratory may seem reliable, because the world around it has been simplified. Lighting is controlled, obstacles are known, scenarios can be repeated. Simulation and laboratory testing are an essential part of that control; without them, progress would be slow and risky. But they can also create false confidence. The real difficulties begin when the machine leaves controlled “greenhouse” conditions and enters real life, in a changing environment.
Perception is not “seeing”
It is not only movement that causes difficulty, but perception too. For a robot, perceiving is not the same as seeing; it is interpretation in an uncertain situation. Cameras and sensors do not provide certainty, they supply data that the software has to interpret. Machine learning has made this process far more capable, especially in detection and classification, but it also creates significant safety problems. A model can be flawed, and its performance can degrade as the weather, the light or the condition of the crop changes. When something goes wrong, it is often hard to explain why.
This is very important, because people working near a robot should never have to guess what it is about to do. What matters most to a person is being confident about their own safety. The key technical question is not whether robots can be made as efficient as possible; what matters is whether they can be made trustworthy, and that calls for a far more responsible approach. Sensors will never be perfect, models will always have limitations, and human behaviour around machines will never be entirely predictable.
In such a situation, safety cannot rest on an optimistic outlook, on handsome robot demonstrations or on a large number of hours worked. Safety has to rest on assurance.
NMBU: safety is part of the robot from the very beginning
If we take the experience of autonomous vehicles, the next question is what a more responsible approach in agriculture looks like. The NMBU robotics group has built safety into the system from the very beginning; before a robot is trusted with real fieldwork, it is tested in stages. Safety is not “something you check right at the end”.
Methodologically, it is right to treat safety as part of the system design, not as a final inspection. The robot is developed through incremental testing, first in simulation, then in the laboratory, and only after that in controlled field conditions.
Formal methods play an important part in this process. The main parts of the safety controls can be modelled and verified before deployment, mathematically and systematically, which means that important safety details are checked at an early stage rather than taken for granted. Verification during operation provides assurance in field conditions by checking that the safety requirements are still being met.
The NMBU robotics group also uses predictive monitoring, which assesses whether the robot is approaching a situation in which it may become impossible to comply with the safety rules. That gives the system a chance to act in good time.
Why does this matter to society?
Public trust in robotics cannot be won by a perfectly staged technology demonstration alone. Trust will be earned if the systems are comprehensible in difficult conditions, if their limitations are taken seriously, and if safety is treated as a design requirement rather than a public relations add-on.
Agriculture is a field in which autonomous systems could deliver real benefits, provided they are developed cautiously, transparently and on the basis of evidence. Safety is often thought to hold innovation back. In fact, it is safety that makes innovation usable. A robot that performs superbly in ideal conditions but unpredictably in the field is not progress. The proof of progress is a robot whose behaviour can be verified, explained and constrained.
If autonomous systems are to become part of everyday work in agriculture, they must first be safe and only then be smart. That is not an obstacle to progress, but the only way technical progress will earn trust.