In the production of apples, the demand for manual labor in picking is huge.
my country is the world's largest apple producer, with an annual apple output of approximately 35 million tons. When apples are on the market, both the eastern and western apple-producing areas require a large number of pickers to pick apples within a few days. Most local farmers grow apples in their own orchards. In addition, problems such as "young people leaving the villages to work and the aging of agricultural workers" have led to the problem of "labor shortages" that often occur during the apple picking period.
Therefore, in order to solve this problem, our country has been researching robots that can pick apples around 2011, and has achieved quite a lot of results so far.
For example, an apple-picking robot designed by a domestic technology company was tested in an orchard in Luochuan, Shaanxi Province. It can determine whether and how to pick apples by analyzing the size, color, shape, etc. of the apples. It has 6 picking robot arms and adopts the popular adsorption-type picking. It can work 24 hours a day and can pick more than 30,000 kilograms of apples in a day.
Of course, this is data from an orchard test, and there will definitely be many problems encountered in practical application.
In addition to domestic research, foreign research on apple picking robots is even more "obsessed".
In 1985, France had successfully developed the first apple picking robot, which could basically pick an apple in 10 seconds in a test environment.
Later, developed agricultural countries such as Belgium, Japan, the United States, and Israel also joined the research on apple picking robots, and they were all quite successful.
For example, an Israeli apple picking robot company conducted experiments on Gala apple picking in 2022. The experimental apple picking robot has 12 picking robot arms and adopts a "3-claw" design. It picks apples one by one like a doll.
The Israeli robotics company said that the efficiency of apple picking in their experiments has been 8-10 times that of manual apple picking.
For another example, a robot picking company in the United States picked Red Fuji apples in orchards in the Columbia Basin, with a picking speed of up to 1,000 apples per hour.
If calculated at this rate, 24,000 apples can be picked in 24 hours a day at the earliest.
Whether it is domestic or foreign research on apple picking robots, from the perspective of picking speed, apple picking robots seem to have been very successful. However, in practical applications, why is it still difficult for apple picking machines to land?
This may be related to the publicity of some media, which only tell you how powerful and ferocious the apple-picking robots are. However, few people mention the actual technical difficulties encountered by the apple-picking robots.
In fact, it is still difficult for apple picking robots to become popular.
The first big problem is the eyes
If you want a robot to learn to pick apples, you must first let the robot know apples.
It is very necessary to equip the apple picking robot with a pair of "good eyes" that are not myopic. This is the vision system of the apple picking robot.
The earliest apple picking robot's vision system used a "camera" (image sensor). At first, the cameras of most apple picking robots were installed at the end of the mechanical picking arm to facilitate close observation and identification of apples. The types of cameras are mostly based on color CCD image sensors, which can clearly identify every detail of an apple.
However, in the picking test, the color CCD image sensor encountered many "embarrassing" problems.
First of all, it is often difficult for the color CCD camera to accurately determine the distance between the mechanical picking arm and the harvested apples, which results in the apples often being "empty" during picking and the picking failure rate is high.
Secondly, the color CCD image sensor is too sensitive to light and shadow. In many orchards, apples are most likely to be obscured by apple leaves. At this time, the color CCD image sensor is prone to misidentification, resulting in the loss of many apples.
Therefore, in future experiments, some companies have added new "eyes" to apple picking robots, such as laser ranging systems, which can accurately determine the distance from the end of the mechanical picking arm to the apples. Furthermore, a 3D camera is added to accurately identify the shape and size of apples.
However, it is difficult to improve picking accuracy simply by improving the "vision" of the eyes.
The second biggest problem is the brain
When an apple-picking robot can "see" an apple hanging on a tree, if it can immediately determine that it is an apple, it can issue an order to wave its robotic arm to pick the apple.
If it is judged that it is not an apple, of course it can stay still and move on to pick other apples.
But the most troublesome thing is, what should we do when something "looks like an apple but not like an apple" appears in its visual system?
This requires a powerful "brain" to make judgments, which is what we often call "algorithm".
Algorithms are now at the heart of apple-picking robots.
The environment of the orchard is not the environment of the test site. On apple trees with dense branches and leaves, in many cases if a robotic arm wants to pick an apple, it must learn how to avoid obstacles and accurately grasp the apple.
It's similar to catching a doll in a claw machine, but harder.
When you catch a doll, you have to use your brain to think about how to catch the doll. The same goes for the apple-picking robot. It has to "think" about how to get around the branches, how to choose a suitable path to pick the apple, and it even has to judge whether it is It's not about using another robotic arm to move the branches.
These are decisions that the algorithm has to make.
If we look at it from our human perspective, this is very simple. A 3-year-old child may know how to pick an apple. But the earlier apple-picking robots all had "mechanical thinking". One is one, and two is two. They will react to any instructions they encounter, which is almost unchanged.
This rigid "algorithm" also leads to low picking accuracy of apple picking robots in actual orchards.
Currently, there are also apple picking robots that use "AI algorithms". Similar to the popular artificial neural network system based on deep learning, they can intelligently pick apples hanging on branches, and can gradually optimize their picking methods.
However, this type of apple-picking robot is still in its infancy. If we really want to realize commercial harvesting, it is estimated that it will take a lot of time and different orchards to conduct "training".
Sometimes if you think about it, teaching a robot to pick apples accurately may be more difficult than teaching a monkey to climb a tree and pick apples.
The third biggest problem is hands
When the eyes find the apples and the brain makes the decision to pick them, what actually picks the apples is the end effector of the apple picking robot, which is actually the "hand".
Designing the hands of an apple-picking robot is difficult.
The initial design of the mechanical claw was a bit "hard". Not to mention the fruit drop rate during picking, there were a lot of apples that were scratched and rotten. Basically, the orchards could not afford such losses.
Later, silicone software "hands" were designed, which reduced the damage rate when picking apples, but the fruit drop rate was still very high.
This is a material problem.
We also encountered many difficulties in terms of shape.
For example, some of the current apple-picking robots have "two claws" and "two fingers" at the end of the robot's execution. When picking apples, these two fingers pinch the apple and then twist it to twist the apple off the branch. There are also ones designed with "three claws", which all use a twisting method to pick apples.
There is also a simpler design, where one hand is designed as a pair of scissors to cut the apple directly. Extend a tray or mesh bag under the falling apples in advance to catch the falling apples.
Each of these designs has advantages and disadvantages.
At present, the most common "hand" design at home and abroad is the funnel-shaped vacuum suction cup.
Just suck the apple firmly, then shake off the stem and pull it off. This funnel-shaped vacuum suction cup should currently cause minimal damage to apples, so it has been widely used. But the flexibility seems not as good as the "claw", and sometimes it is difficult to suck apples in tricky places.
How far are we from?
It seems that apple picking robots are not far away from us, but "a person who travels a hundred miles only takes half a mile". Many previous research and research directions may be focused on speed. However, in complex orchard environments, accuracy and low Damage is arguably more important.
If apples are missed or too many apples are damaged during picking, the expectation of such an apple robot in saving labor will be greatly reduced. Just like some so-called automated equipment, it is said that it saves labor, but in actual production, a person has to be hired to keep an eye on the equipment, otherwise it is easy to make mistakes.
This instability is what apple picking robots need to focus on overcoming.
As for whether the picking speed is faster or slower, it is easy to say. After all, the robot can pick 24 hours a day, and the longer working time can offset the lack of work efficiency.
However, even if these problems are solved.
In actual orchard production, many of our orchards still have no way to use this kind of robot.
The reason is simple. The apple cultivation model in many old orchards is "arbitrary". The canopy of the trees is very large, and the gap left between the two rows of trees is too narrow, and the picking robot cannot enter at all. Furthermore, many orchards are located on mountains, and the planting is done one tree at a time, which is not conducive to the work of picking robots.
In the future, apple picking robots will be more suitable for large-scale modern apple plantations.
Apple picking robots are still far away for small orchards.
It may indeed not be too far away for modern, large-scale apple plantations.