📣 🐍 🏴󠁧󠁢󠁷󠁬󠁳󠁿 Why it is necessary to robotic agricultural combines, what are the difficulties, and how we did it in two years Ⓜ️ 🐬 👨🏻‍🌾

In normal times, the machine operator earns about 30 thousand rubles a month. But everything changes dramatically during harvesting, when the machine operator temporarily becomes the operator of the combine harvester - the combine operator, during this period he will receive up to 150 thousand rubles. There are literally two weeks when you need to collect everything that you invested a huge amount of money for work, fertilizers, diesel fuel and so on for a whole year. You can work from about eight in the morning (set up the car, start at nine) until dark, because dew and night humidity sharply worsen grain quality. Wear. And on the third or fourth day, problems begin with accidents or grinding the wrong and wrong.

From the outside, it seems that the task is to drive along the field with a “snake” and “grind” all the wheat or another crop on the combine. In fact, this is far from the case. The operator must monitor hundreds of things and at the same time constantly look at the edge of the field in order to move smoothly. Imagine that you drive 12-14 hours on the highway at a speed of 120 km / h behind a person who suddenly slows down every half hour. The operator feels about the same thing: the work is incredibly monotonous, but at the same time you constantly need to be prepared for a surprise.

Surprise may look like this. While we were driving "to the fields", we did not see a single whole combine without traces of welding.

In fact, the combiner monitors the grain processing process (one workplace) and at the same time conducts equipment (another workplace). But this is only one person. Consequence - one thing suffers. Since, if it is bad to lead, you can accidentally grind a stone or a person on the field, and the quality of grain harvesting usually suffers.

Part of the work is easily automated. Now I’ll tell you exactly what we did and how we modify even fairly old combines with our robots.

Turns

So, for two weeks, the operator of the combine works for wear. His payment consists of a salary (the same 30 thousand in the region) and a bonus for the amount of grain received (which allows you to get up to 150 thousand rubles in the region). At the same time, the agronomist looks at the uniformity of the field coverage - if “cocks” remain in the process (that is, mowed areas due to an uneven turn), they will need to spend a lot of extra resources on collecting them: fuel and lubricants, depreciation, transport run, time. As a rule, such areas are “being finalized” in the morning, and the efficiency of the combine is very low there. And the whole business effect of the economy depends very much on reducing the cost of cleaning: they cannot affect the market, but they can reduce the cost of grain for themselves. Therefore, there are consequences:

The field is covered by combine routes in such a way as to ensure continuous movement for complete filling.
— . . : . , — . — . , . — , , , — . — 30–50 , — . .
. ( , 90 % ) .

According to statistics, the machine operator due to fatigue by the end of the shift leaves out of the grapple up to one meter from the header length. With the speed of the combine harvester at 10 km / h and a decrease in capture by one meter in four hours, it will “underestimate” four hectares. In addition to losing time for cleaning, fuel consumption increases. At a fuel consumption level of 20 l / ha, the direct loss in fuel (which will be required to clean up the lost space) will be: 4 ha x 20 l x 45 rubles / l = 3,600 rubles / shift. The cleaning of the remaining area in the next shift will take an additional three hours of work. For four days, the machine operator increases the duration of his work by one day. During the season, the harvesting time increases due to a machine operator error of 25%.

For 20 days, the loss on fuel will be 20 x 3,600 rubles. = 72,000 rubles, a fleet of 10 cars loses 720,000 rubles during this time on fuel and lubricants.

Here is a video where we show the au pair how the robot works. The operator of the harvester is filming the actions of the machine. This is the reaction of a person suddenly confronted with the future:

Another important thing - machine operators often come on a rotational basis. The field owner or agronomist does not know them, does not know where they worked, how they worked. He understands that they will chase the fields for the maximum bonus. He needs this system to secure his equipment.

Clashes

On average, according to our statistics, one or two serious collisions occur annually. Not one combine harvester works on the field, but several. Along the way, the combine may encounter another combine, a tractor, a tree, a power pole unexpectedly entering the field, a person (including another combine who went to the toilet), a cow and other obstacles.

I know the case when the operator watched the edge, and then suddenly saw that something red was starting to fly from the combine. A man was lying on the field, he was picked up by the reaper and began to process. The machine operator was acquitted: the investigation found out that it was as if the local lads had thrown the corpse onto the field. But he went to a psychiatric hospital. Here are a couple more cases: I drove up to refuel, hit the driver , and this is the mysterious "Adam" .

It is clear that when something is hiding in culture, it is almost impossible to prevent a collision with it. Therefore, by the way, there are no large stones in the fields: they are removed even before sowing (and small ones fall into the stone traps of the combine). The operator sits quite high, and most obstacles are visible in advance. The question is where does he look (sideways at the edge, at the console or forward) and how good is the reaction.

According to my information, rodent trunks are most often damaged and reapers beat. One of the common reasons is that when moving in a group (for example, a wedge), the uneven speed of the combine in front.

Our robot allows you to keep a distance in front of going.

The Cognitive Pilot uses a video camera that looks from the cockpit just above the operator’s seat and recognizes obstacles. The robot knows that the pillars and trees stand still, but it calculates the motion vectors of people, cows and tractors. When the situation becomes simply requiring attention, a notification occurs. In critical circumstances, the harvester will stop abruptly.

This means that we save several days of repairs during the harvest. That is, repair at such a time is not only the cost of parts and work, but also huge losses of grain.

Speed and other settings

What can an operator manage? The first is the reel speed (this is a device such as a screw that compresses the ears under the shear). The speed directly depends on the type of crop (one for corn, another for wheat) and the maturity of the crop.

The peculiarity of the work of the combiner is that it is a difficult algorithmizable task with many purely experimental assumptions. On the other hand, an almost complete action log is written in the combine itself, and you can set the percentage of grain losses due to incorrect modes. There is not enough input data: humidity, type of crop, its specific variety, region (the crop is very different in form and size from soil and climate), and the degree of maturity. If you combine all this, you can get either a neural network trained in 20 years of experience in the head of a skillfully working operator, or a robot that will do the same for him.

Skillfully working operators are now dying. Almost all effective machine operators were trained under the USSR, after which the technological secrets of this civilization were lost. Specifically, training has weakened, and newcomers, of course, work worse. This is normal for almost all working professions, but specifically here, our robot is in high demand as much as possible because of this effect.

So, if the rotation speed is higher than optimal for this situation, the grain will be kicked out by the header when even before it gets into the main processing, the grain will begin to crumble to the ground under the combine. This is from three to ten percent of losses for the wrong mode. If you give less speed - the ears will not bend to the end, and they will begin to fall to the ground, and not to the header. The same result.

If the grain is not fully ripened - already large, but still tight in the ear - the ear with it can easily be considered a single object. But if it is slightly overripe, then it’s easy enough to hit the spike, and it will begin to crumble. On one field for different passes, you need to change the settings: it happens that a slightly laden mass or non-laden. Somewhere a piece of field is on a hill, and the other end is in a lowland. The lowland has more moisture, and there the stalk is higher.

It is necessary to keep the edge, constantly change the speed of the combine depending on the characteristics of the plot (on dead grain - up to one or two kilometers per hour, on ordinary culture - on average six to eight kilometers per hour), constantly adjust the speed of rotation of the reel.

And the machine operator receives for the volume.

There are a lot of machine operators on the field. And the amount of culture is limited. Therefore, they naturally want to earn more. They have no task to save for the economy, they have a task to personally collect as much grain as possible. Therefore, instead of moving at the right speed and selecting all dozens of parameters, it is enough to move so that the agronomist does not drive out of the farm for the next crop. That is, as fast as possible before he begins to ruin and swear.

In practice, this means that the efficiency of grain harvesting is far from optimal. Old masters at the beginning of the shift show excellent performance, but it’s hard to keep them until the end of the shift. "Young" focus on movement and work with low efficiency, trying to gain more grain per shift. Harvesting has passed, the first rain - the green stripes are. It was someone who sowed the grain. This is a loss.

Spike processing

The next type of settings is what happens inside the combine. Ears on the conveyor from the header fall into the separation apparatus. It may differ in design, but in general there are two drums, under which there are sieves. Drums work on knocking out grain from ears with the help of hammers. They simply knock on the mass, and the grain spills out into a sieve.

It is very important to adjust the concave according to the volume of the mass (that is, different modes for different speeds or different speeds for different parts of the field: it depends on the density of ears, on the size and density of grain, on the relative grain size and on the ease of knocking it out of the ear, i.e. maturity and humidity). Unripe grain needs to be beaten harder, a mature one does not need to be pounded especially.

In fact, this means that before each different section of the field, you need to reconfigure these parameters. My practice shows that no one does this, because he is afraid to get into the settings or simply does not want to deal with them. On modern combines there are sensor systems that show the process on the monitor. On the old, this is not. The times have already passed when it was necessary to twist something physical in order to change the operating modes of the mechanism. Now you can adjust everything at any time. But my approximately 12-year sales experience shows that after the sale of the combine, a dealer arrives, sets up the main characteristics and writes them in a piece of paper. Of the 200 or so combines I wrote off, I saw exactly three with parameters that differed from this dealer’s piece of paper.

That is, losses simply go on here because they usually sacrifice this part of the work so as not to waste attention. I have seen how, under the notifications “less speed, loss”, “unfinished, reduce concave”, ignore alerts. “Grandfathers worked like that - get on a fig? If something goes wrong, then the whole season will be idle. ” As a result, they concentrate on driving.

Next comes the post-treatment of grain. There are rotary and keyboard mechanisms, there are much more keyboards on the market. Straw is pulled through a keyboard mechanism, and a separator below. The additional treatment and re-finishing of the grain results in the whole mass falling into the wind (or a similar mechanism) when the air stream blows all the lungs (small pieces of straw, pieces of leaves) to the side, and the grain falls into the hopper. Winds need to be controlled tightly, but, fortunately, you can set them before the shift in the morning. Rapeseed - minimal air flow is needed: the seed is very light. Corn - maximum flow.

Again, in practice, it looks like this: the 64-year-old Mikhail Kuzmich arrives, picks up an ear, grinds the grain in his hands, and then adjusts the wind to his site. The rest of the mechanics come up, look, copy the settings.

Very common picture.

Tractor interaction

Harvesting can go either with the dumping of grain into something that goes next to the combine along the already worn area, or you need to call a tractor from time to time to unload the harvested crop.

If we are talking about driving nearby, then you need to make sure that the grain flow gets into the transport. This means that we must now look intensely in two directions: at the remote control and forward.

If we are talking about a call, we must monitor the filling of the bunker and think about how the transport will approach, that is, solve the strategic problem of optimizing its call. As we know, people usually solve strategic problems using imperfect greedy algorithms, which is why fuel losses also occur here. Fortunately, this is the smallest of the problems on the field, so we are not touching this part yet.

The situation becomes much more interesting when the combines do not move in corners (in their sections), but in a wedge. That is together. Here you need to follow the lead and other combines in addition.

Improvement points

What can a robot do?

One of the most loaded parts of the work is to follow the trajectory of movement. That is to steer.

The past generation of automation solved the task of using GPS-devices. Here is an example of how this ends badly and regularly enough. Let me remind you, you need an accuracy of 20 centimeters. There is a problem with them in Russia: if there is no active amendment via the Internet, then you need a ground station with a radius of 15 km. A ground station costs money, and, moreover, you need to buy signal packets in order for it to work, that is, a subscription. GPS-harvesters are very “oblique” - yes, they can clean the ideal field themselves, but they make bad turns and do not monitor the situation. When the signal is distorted, “cocks” are left.

At first we thought about GPS + video analytics, but it quickly became clear that only video analytics is enough.

Now the combine is able to move according to the plan in the pen, is able to recognize obstacles and slow down in front of them, is able to coordinate movement with grain transport and other machines in the wedge.

GPS requires a long map programming on navigation points. This is a complex interface, and complex interfaces in the field do not work. We made it much easier: the autopilot sees: “Wow, field!” - and offers to transfer control to him. All.

The operator can monitor the reaper and culture. The robot steers.

Modes:

Along the edge of a mowed crop or cultivated land, in a row (corn, sunflower), felling (separate harvesting) + detour of stationary obstacles, stopping when there is a risk of collision with other equipment or people, identifying plots of the fallen crop and warning or stopping in front of them.

In addition to the harvester, it is still possible to install a co-pilot on a tractor, mower or sprayer.

Now you can roughly imagine what it takes to get bread, corn, pasta, buckwheat on your table and where you can cut it. I will dwell on this for now, because for many years I worked in agriculture with technology, and not in development and not with neural networks, but in the next post my development colleague will tell you exactly how it all works and what exactly were the difficulties to do it all in two years. And they were, believe me.

The overall economic effect is a reduction in the cost of grain by 2.5–5%. Alas, this is analytics, not an AB test (it is difficult to conduct it because all the sections are different, the machine operators too, and this is only the second year of work), but, if interested, I can tell you how we calculated it.

PS If your agronomist is not on Habré, and he is interested, you can find contacts here: promo.cognitivepilot.com and discuss in detail what combine harvester, which specific set of equipment is needed, how much it costs, and how you can quickly see it- to experience.

PS Here is the promised post - habr.com/ru/company/cognitivepilot/blog/497098 .

Why it is necessary to robotic agricultural combines, what are the difficulties, and how we did it in two years