The cyclist crept unnoticed

Unmanned cars on the road - a unique achievement of progress! Moreover, the developers have prevented the danger of collisions with other cars, pedestrians, even squirrels and birds, which is especially important for American cities. But still formed a group of vehicles, which remained vulnerable due to its criteria. These are cyclists - quiet, light and maneuverable members of the movement. The unmanned driving systems that exist today are mediocre in their task of detecting them in the vicinity of themselves, and also hardly predict their actions. Although cyclists on the roads are relatively few, but do not ignore the same problem.

According to a number of researchers, unmanned vehicles today are the most difficult to recognize cyclists on the road. This is due to their uncharacteristic speed, small size and dissimilarity to each other. Cars are much more similar than these two-wheeled road users.

If you walk through any city and look closely at the bikes, their diversity becomes obvious. Here is a young man on a green sports, but the girl drove on a city pink ladies' bicycle. This woman is going to her dacha, therefore, in addition to a basket, another 2-3 bags of seedlings are hanging on her iron horse. Here, the mom takes the child with her in the bicycle chair on the trunk, but this dad also carries her daughter in the bicycle chair, but already installed right behind the wheel.
')
Any drone in the headlights charges from such a palette and cultural diversity!



The electronic driver collects all the necessary information about cyclists using three devices: a camera, a radar and a laser locator. After that, it evaluates the parameters of the object using the previously obtained data on these participants in the movement. In order for the autopilot to accurately know the intentions of the cyclist, the dimensions of the head and arms, the distance between them, the distance from the cyclist's head to the pavement, and the angle of the bend of the elbow are measured in detail. With a good angle on-board computer is able to determine the model of the bike.

It is already possible to trace the relationship between the problem of detecting cyclists and the process of creating a new generation of cars. Scientists teach autopilot systems on a variety of vehicle images that show various cars, but almost no bicycles. In this lies the flaw. Not all unmanned vehicles are taught to recognize bicycles, and therefore they have difficulty finding them on the roads. Fortunately, this problem is temporary.

Recently, the auto industry began to use the algorithm Deep3DBox . He owes his appearance to researchers at George Mason University in tandem with the development of unmanned taxi cabs Zoox. In the test mode on two-dimensional images, the algorithm is able to determine 89% of cars. Last but not least, he is very good at predicting which way other machines are heading. Moreover, the algorithm even draws a box-like area around each object. Today Deep3DBox is considered one of the best car recognition algorithms. But, unfortunately, in a comparative analysis, he is able to see only 74% of bicycles, and only 59% of them can guess the further trajectory.



Note that this is the result of one of the best algorithms, not to mention the less advanced developments. The situation is quite sad for cyclists: for unmanned vehicles, they are likely to look like an unpredictable moving unidentified object. And this contributes not only to the "untrained" algorithms, but also the compactness and openness of bicycles compared to cars. Because of this, in recent years, automakers had to increase the frequency of scanning the road with radar and lidar, so that the system could more confidently detect bicycles, determine the distance to them and their trajectory. Together with an increase in datasets for learning algorithms, this allows an increase in the proportion of correct recognitions .

Another unique development is designed to save the lives and health of cyclists - 3D maps of ultra-high detail. The computer sees all the objects on the road and next to it, all the marking lines and road signs. Such awareness will help to quickly identify and mark a bicycle on the map.

Perhaps, after this you start to relate differently to the point of view that the computer drives the car better than an inattentive and always in a hurry person. Quite a few cyclists get into accidents because of the carelessness of drivers. So for someone, the appearance of cars driven not by scattered and tired people, but by invariably attentive computers, which also learn to recognize the gestures of cyclists - a real golden age!

For example, today, Google unmanned taxis are equipped with sensors that allow you to discern and decipher cyclists' intentions. When cornering, cyclists must give their hands the appropriate signs according to the rules of the road. Google drones have learned to read and recognize these signs in order to properly maneuver or change speed. Also robotaksi able to see two-wheeled participants in the dark.

But there are also difficulties created by the cyclists themselves. As developers of Deep3DBox note, it is much easier to predict car maneuvers than a bicycle. The vast majority of car drivers act much more predictably, while bicycle owners usually have a very vague idea of ​​traffic rules, and every second cyclist behaves on the road as an impulsive and charismatic person. If this person decides to suddenly appear out of nowhere among the stream of cars, then even the smartest algorithm may not be able to react correctly.



This characteristic feature - unpredictability - today is taken into account by the developers of bicycle detection algorithms. For example, today cyclists can be less wary of Jaguar UAVs, which are equipped with a cyclists detection system. And Volvo in 2013 introduced the technology of emergency braking when detecting a cyclist (AEB) in front of the car. The system will stop the car before the driver to avoid a tragic accident. However, AEB is not perfect: it is still difficult for algorithms to predict the trajectory of a cyclist. So it is strongly not recommended to use this feature Volvo on the streets of Beijing: the daily flow of 9 million bikes can cause damage to the electronic brain.

In addition, cyclists themselves can make fun of electronic drivers. A couple of years ago in Austin there was an incident with a cyclist and unmanned taxi Google. The car was embarrassed by a man made surprise. The cyclist stopped to skip the car without removing his feet from the pedals. The car moved, but noticing that the bike had moved 1 centimeter, it braked sharply. For the sake of the experiment, the Texan repeated the trick several times, which made the taxi passengers much fun. And now imagine these cars on the streets of Russian cities. Knowing the tendency of our fellow citizens to humor and jokes, it is not difficult to guess that the drones will not go far away.

Diversity - a feature of the modern world, which is transferred to the road. In many cities, more and more citizens choose a bicycle. And if the cyclists of Sweden for many years angrily waving gagging drivers, then in connection with the advent of unmanned vehicles, the policy should be built in a different way. For example, in the United States, Uber unmanned taxi services are in full swing. Naturally, the transition to this level of service has significant advantages for both the client and the company. A taxi ride without a driver should be cheaper, the computer is not rude to customers and does not violate the rules of the road, does not get tired to work from morning to evening, etc. Uber itself, naturally, is more convenient to work with machines whose actions can be improved and controlled. In a number of countries, quite a number of crimes have already been recorded with the participation of the company's drivers, since they do not pass a special selection when applying for a job.

While testing is underway, rides on Uber UAVs are completely free. As for an attraction, you will sit in the back, and the receptionist will record the entire route. So far, the machines do not behave perfectly, but in Uber they are not upset about this, considering such violations as a working moment. Whether it is carelessness or constant readiness to innovate, this issue provokes numerous debates between supporters of the latest technologies and conservative views.

By the way, in San Francisco, one Uber drone was caught driving through a red light, while several others turned around on bicycle lanes. But at that moment people could pass there.



A number of experts condemn the zeal with which Uber seeks to market the unmanned taxi. It is difficult to argue with the statement that such crude technology must first be thoroughly tested before carrying passengers. You can not do business on technology with such a high probability of failure. Hopefully, other companies will more responsibly approach the introduction of UAVs. In the meantime, cyclists beware!