Self-driving Machines — Is It A Service Already?

Since the beginning of the XX century, there have been made some attempts to develop vehicles that could move autonomously, without human intervention. Car manufacturers, research institutes and development teams are trying to get people off the wheel, giving control to dispassionate automatic systems. The field of self-driving cars technology received powerful impetus for development at the beginning of the XXI century, when computer systems, detectors and sensors, machine learning algorithms and artificial intelligence began to evolve rapidly, invading all spheres of human life.

The man behind the wheel is dangerous

An invention that should have made life easier, more convenient and more comfortable becomes a cause of death for a huge number of people. According to the World Health Organization (WHO), around 1.35 million people a year die as a result of traffic accidents, and from 20 to 50 million people suffer, losing health and time to recover. As a result, states lose about three percent of gross domestic product.

The desire to reduce these figures drives the governments of different countries and leads them to the toughening of legislative rules as well to greater control and severe penalties for violation of traffic rules. Vehicle manufacturers spend impressive amounts on passive and active safety systems that preserve the lives of passengers, drivers and pedestrians.

However, there is a radical way to improve the situation: the exclusion of one of the main causes of danger on the road — the drivers. According to the WHO, the main causes of traffic accidents are:

• overspeed;
• driving under the influence of alcohol or other psychoactive substances;
• neglect of safety belts, car seats and helmets;
• distraction:
• unsafe road infrastructure;
• unsafe vehicles;
• inadequate medical emergency after accidents;
• inadequate laws and regulations.

Having given control of cars to computer systems, you can get rid of the largest and most dangerous part of these factors. This is one of the main reasons for the accelerated offensive of autonomous transport technologies. Well, of course, one of the main marketing messages of developers and manufacturers.

Are robots behind the wheel really safe?

Self-driving cars equipped with computers and many sensors that continuously monitor the state of the road, analyze, calculate the best maneuvers and tactics, are designed to increase overall safety. However, accidents and traffic accidents with autonomous vehicles are by no means uncommon.

According to the Axios report, for the period from 2014 to 2018, 88 accidents involving self-driving vehicles occurred in the US state of California. Nevertheless, drivers of other vehicles, and not autonomous control systems, were found guilty of these accidents in the vast majority of cases. That is, the robots worked well meanwhile their biological colleagues ruined all the statistics.

In March 2018, a fatal accident occurred in the North American city of Tempe. During the test drive a self-driving Volvo XC90 equipped with an autopilot from Uber hit a woman crossing the road at night outside the crosswalk. The investigation showed that not only the pedestrian didn’t take care of her safety. A backup driver who was responsible for control of the autopilot got distracted and didn’t follow the road. At the same time, a machine control system didn’t take measures to put the brakes on immediately.

Robots are still imperfect. There are several cases when drivers fail autonomous driving system in Tesla cars. This is another evidence that it is not yet possible to rely too much on robot drivers. In defense of the Tesla computer system, one can say that it is not intended for a fully autonomous driving — only to help a driver. The driver should not be distracted while the autopilot helps him.

Different types of robots

Autonomous driving systems are divided into several levels according to the degree of control over the vehicle and the degree of human attention required: from zero — when the driver is fully controlling the car, only accepting clues about the road situation and car behavior, to the fifth — moving in a completely autonomous mode, there is no driver, people in the cabin are passengers.

According to this classification, for example, Tesla autopilots belong to the second level: the car takes on all the functions of the driver — it speeds up, slows down, turns, but the person must completely control the situation and be ready to take control.

How robots see the road

Compared to humans, computer systems see further, better and control much more nuances of the road. As a rule, they do this with the help of several cameras supplying a video image; radars scanning the environment around the vehicle using radio waves; and lidars — “optical radars”, in which light rays are used (including ones from lasers).

In addition to the systems described above, sonars can also be used — they emit sound waves instead of radio waves or light rays, like radars and lidars. Naturally, automatic systems use any other parameters and car systems available to them that help them better navigate and control the vehicle: GPS, engine sensors, microclimate, humidity and many others.

It’s good to see the road environment but it’s not enough for a successful autonomous ride. It is also necessary to correctly recognize detected objects on the road and beyond: other cars and their signals, markings, signs, traffic lights, pedestrians, motorcyclists and cyclists, obstacles and damage to the roadway, roadside noise barriers. This is the task for pattern recognition systems. Such technologies are quite well developed in our time, and the involvement of artificial intelligence and machine learning makes this task quite simple.

By the way, active Internet users take part in the development of such systems: the “reCaptcha” service (when you need to mark traffic lights, bicycles or buses in the photographs), for example, this is nothing more than training artificial neural networks of pattern recognition systems.

This is one of the moments due to which there are so many information giants and search engines among companies developing self-driving vehicles. Thanks to the various services developed by Google, Apple, Baidu or Yandex, the portfolio of these organizations has image processing technologies, cartographic systems, developed divisions of artificial intelligence and huge data arrays.

The head of Yandex’s self-driving vehicles department, Dmitry POLISHCHUK, said that the quick start and successful development in this area had been supported by a significant technological reserve. The Russian search engine had already developed and tested by numerous related services: computer vision and pattern recognition, navigation application and user localization features.

The identification and recognition of objects on the road and next to it is not all that a good self-driving car needs. An equally important part of the system is direct driving and responding to instantly changing traffic conditions. As has been shown above with statistics on accidents of self-driving vehicles, even a well-tuned autopilot is faced with unpredictable behavior of other participants — and it must be able to adequately respond to inadequate maneuvers of people driving cars around.

Computer-based vehicle control systems train these skills by conducting many test drives, both in reality and in road traffic simulators. For training self-driving vehicles in real conditions, there is a backup driver, who is obliged to intercept control in difficult situations or when there are obvious violations in the operation of the system.

Other advantages of robots

The evolution of vehicles whose drivers are not subject to fatigue, scattering attention, driving while intoxicated and emotional states leading to road conflicts — this is only one of the important advantages of self-driving cars.

Other advantages may include:

• reduction in fuel consumption: cars under control of robots will be more economical, since they drive in optimal engine operating modes, on shorter routes and with lower costs;
• increased mobility for citizens with physical disabilities due to age and health;
• acceleration and streamlining of transport routes: the more self-driving cars will be on the roads, the more order will be on highways and streets, which means that the driving time will be reduced;
• the appearance of more free time for people who often move from point to point: due to the fact that there will be no need to be distracted by driving, different activities will become possible during the ride: conducting business negotiations, watching movies, talking, sleeping;
• cost reduction: the number of owners of private vehicles will decrease due to the greater distribution of taxis and car sharing, insurance will decrease due to greater safety and fewer incidents on the road.

Potential threats

Any innovation carries a whole range of changes in familiar life: both positive and negative. The improvement and distribution of self-driving vehicles is accompanied by threats to the confidentiality of the established labor market and information security. From the point of view of the employment market, professions related to driving and servicing vehicles are at risk: robots are likely to lead to massive reductions in these areas.

Confidentiality is at stake, and each new information service digs an ever deeper hole, from which it is increasingly difficult to get out. Self-driving cars are not vehicles in the common sense — this is a service. A wagon full of electronics, computer systems and connected to all possible networks is a service that automakers hope to sell very well.

Here are all the ensuing consequences for the consumer:

• regular paid subscriptions, without which the functionality will be limited or the machine will simply be deprived of the opportunity to use it;
• a continuous stream of personal information flowing from the user device (car) to the servers of the manufacturer and the partners;
• endless advertising, which will finally become the most targeted and completely consumed (do not like the non-disconnectable promotional video — try to escape from a moving car that you don’t drive);
• the maximum loss of personal space — the car will know exactly where its owner is going, where, how often he goes in this direction, what he talks on the phone during the trip, how he feels, what emotions he experiences (sadness, excitement, distraction — they can be used for marketing purposes), and all this will be recorded and maintained.

The complexity and variety of computer systems in a self-driving machine makes it a good target for hackers. Getting outsiders access to these systems can lead to a wide range of problems for a person inside: from blackmail and theft to death. In 2015, there was already a situation with a Jeep hack.

It seems that there were no more such scandalous and vivid examples of weak protection of modern cars (which can speak of both enhanced measures taken to protect and more thorough "filtering" of leaks), but with the spread of self-driving cars they will appear: the more popular the device, the more incentives hackers have to break it.

The train is racing at full speed

No matter how terrible the threats are, it seems that this process cannot be stopped. Huge funds have been invested in the development, creation and testing of self-driving cars, and progress is evident — cars without drivers are becoming more technically advanced and more attractive to consumers. The future is, when a man driving his old-fashioned non-autopilot car on the streets, will feel like a black sheep among the rows of self-driving cars busily scurrying around.

And then for all other participants in the movement, such an unpredictable “colleague” will be as much a curiosity as the current self-driving vehicles: on the screens of the on-board computers this brave man will be marked with a flashing red frame, and all robots will pay increased attention to the last human driver.

Image: Science News