“There would be technology, but there will be an application”. But all the same, we must first set the task, and then look for ways to solve it: you can not just think of something and wonder where it will come in handy. Scientists and engineers put forward hypotheses, explore theoretical premises, check ideas, and try to implement them. Sometimes the discoveries that are considered advanced have been mentioned a while ago, although it was not about computers and high technologies.
- Autopilot. Start
- Wires on the road. Utopia of the “automatic highway”
- Citroen DS19 with air suspension. The first full autopilot
- Computer vision. The machine must be controlled by a computer
- 100 km / h. Finally, without wires
- The birth of ALVINN. Neural networks
- Million dollars for the current autopilot
- Google enters the case
- Tesla showed Autopilot
- Without AI and not there and not here
- And what about Belarus?
Let’s remember the childhood – the time when everything around seems new, unknown and full of secrets, when the focus with the “tearing off” of the finger causes a storm of delight and the desire to see it again and again. And even adults tell tales to children, but only Pushkin-Chertyak was right: “A fairy tale is a lie, but there is a hint in it!” Good fellows a lesson. ”
A little re-enact the well-known plot: Emelya blackmails the scientist Schuku, who leads a reclusive way of life, and gets at his disposal secret technologies. Being naturally lazy, Emelya makes Shchuk develop a self-propelled sleigh, and later – a much more comfortable oven.
The idea of autonomous vehicles found features in the first half of the last century. In 1925, the American inventor Francis Goudina showed a radio-controlled car based on Chandler. From the outside it seemed that the car was going by itself – after all, there was no one at the wheel, since the driver was sitting in another car behind, from where he switched speed on the radio, stuck to the horn and ruddered.
Realizing that to drive a car, though remotely, but in close proximity to it is not very effective, engineers have come up with a way to circumvent the restriction. A step forward was made by the futurist and industrial designer Bel Geddes, proposing the road with the wires laid in it.
Together with General Motors at the Futurama exhibition in 1939, the engineer showed a prototype of the system with electric cars that found their way thanks to electromagnetic induction (coils in the machine and the cables built into the roadway interacted with each other). In 1940 Geddes published a book in which he asserted: people should not sit behind a wheel – in order to save lives. According to his forecasts, the future was to come in the 60s of the XX century.
In the postwar years, the desire to make the car and the road a single whole came back, Bel worked for good reason. The RCA company joined the case, the same one, thanks to which there were connectors like “tulip”, as well as an original television test table. In the laboratory, a miniature test range with a toy car and wires in the floor was deployed, which in a literal sense became a guiding thread – the principle of organizing the “automatic road” described above was implemented.
“Very cool!” – thought local officials responsible for local roads, and offered to implement a full-sized project – with real roads and cars.
Engineers agreed. To further impress the commission, the demonstration was conducted first with the help of Chevrolet Impala in 1958, and then – futuristic for those times concept Firebird III. Everything would be fine, but … “It’s very expensive!” The officials said. The equipping of one mile of the road required $ 100 thousand (about $ 860 thousand today), so the “automated highway” by 1975, as originally planned, did not appear.
In the 1960s, the trend was picked up by other companies, trials were conducted, but the approach was always the same: cars and expensive infrastructure around should operate in tandem.
The first car with a full autopilot, probably, was the Citroen DS19. In the framework of the UK-implemented project, engineers modernized the hydraulic system of the car, adapting it to affect the pedals and steering wheel. The vehicle, without human intervention, kept the road with the “wires” laid in it at a speed of about 130 km / h even on a snow-covered highway. Effective, but again too expensive.
Then John McCarthy, about whom we spoke earlier , joined the case . He, we recall, is the author of the term Artificial Intelligence, introduced in 1955, as well as a developer of the Lisp programming language for AI.
In 1969, the scientist published an essay “Computer-Controlled Cars” (Computer-Controlled Cars). In it, the author discusses the topic of an “automated chauffeur” capable of driving a car on public roads, relying on visual information from cameras. The introduction of the route McCarthy proposed to carry out using the keyboard, it would also be used to enter current commands (slow down near the bar, for example).
McCarthy pointed out the direction in which it is necessary to move engineers. True, until the early 90s of the last century, his ideas remained a concept, until someone remembered neural networks in reference to autonomous vehicles.
In the 1980s, attempts were made to develop self-controlled machines. Having worked at NASA engineer Ernst Dickmanns (although he achieved the main heights in his homeland), which is called the “German pioneer of computer vision,” he designed his brainchild – VaMoRs based on, apparently, Mercedes-Benz L 508 D.
The equipment installed in the bead controlled the car, receiving information in real time from the cameras. The software was engaged in processing sequences of images and translated what they saw into teams, and the robot “ruled”. The first completely self-contained arrivals of the updated VaMoRs began in 1986, and the maximum speed reached almost 100 km / h. True, on an empty road.
In late 1986 – early 1987, the Prometheus Project program was launched with the aim of creating a self-moving car. The project was funded by the European Research Coordination Agency EUREKA, which allocated about 750 million euros in equivalent.
It was decided to abandon the previously proposed technology for placing “navigation” cables on the road and to take Ernst Dieckmann’s work with his machine vision as a basis. Later, thanks to Prometheus, Mercedes had a prototype VITA – Vision Information Technology Application (technology uses computer vision to analyze the position of the car on the road and manage it).
The problem was not only computer power limited for those times, but also in their sizes. It was not for nothing that the tests were carried out on a cargo bus – the body was given to the equipment and air conditioners for its cooling. By the way, GPS was also unavailable to mere mortals, so it was necessary to ensure the storage of terrain maps. A new goal is to reduce all equipment to acceptable dimensions.
In October 1994, a couple of Mercedes cars, including the Mercedes-Benz 500 SEL, drove at speeds of up to 130 km / h on the highway in the usual traffic for the road, independently changing the lanes and outperforming other cars (with confirmation of the driver’s actions). And in 1995, the robot, looking at the road with saccades (a constant cursory inspection of the environment) and utilizing the available capacities of the multiprocessor system, accelerated to 175 km / h, having traveled more than 1,5 thousand km. The maximum distance covered by the car without human intervention was 158 km on this route.
In the 1980s, work was carried out in the United States with the financing of DARPA. The American project was named ALV – Autonomous Land Vehicle. Their main achievement was the development of a navigation system capable of driving a car when driving on rough terrain, and not along an asphalt road with clear markings. The computer, using a map and sensors, dispersed the car to 3 km / h and drove about 600 meters. While a little.
This was the beginning for ALVINN – the so-called armored car, which became the basis for the new technology. The name of the vehicle consists of two abbreviations – ALV and NN (Neural Network, or neural networks). The idea of using neural networks was promoted by Carnegie-Mellon University and its student Dean Pomerlo in particular. Then he completed the next stage of education with a specialization in machine learning, machine vision, robotics and neural networks.
The training of the ALVINN autopilot, which became classics, looked like this: a man-driver drove the car along a path, at which time the cameras collected data around, the actions of the driver were recorded. The computer defined various parameters of interaction of the person and the machine, receiving the answer to a question “that it is necessary to do to keep the car on the road” for different situations.
Each neural network was trained for a particular type of road, and then ALVINN was able to pick the right one, based on an assessment of its effectiveness and reliability in a particular case. Actually training took a couple of minutes, after which the computer with a capacity of a tenth of the capabilities of the smart clock processor Apple, acted independently. The man could only squeeze the pedals.
The next steps to improve the technology were made in the early 2000s. The US government has launched several projects involving the army and the DARPA agency. The agency, in turn, in 2004, offered a million dollars for a functioning management system capable of carrying a vehicle along a multi-kilometer route in the Mojave Desert. Unsuccessfully.
The second and third attempts in the Grand Challenge were more effective, and in 2007 the city car won a car of the University of Carnegie – Mellon and General Motors. Even then it became clear that a set of a pair of cameras and a computer is not enough, as indicated by the first two Grand Challenge. At the same time, “interest clubs” were formed, the theme of autonomous machines became widespread.
And in 2009 Google entered the business. While secret, calling the head of the project Sebastian Trun. He participated in the competitions DARPA Challenge II and III and took high places, besides being the director of the Laboratory of Artificial Intelligence in Stanford.
Here we will make a digression. The principle of finding a way in modern autonomous vehicles is approximately the same. Sensors constantly collect information around, generating a map. Ready-made cards can be used.
Speech, as a rule, is about a set of lasers, radars of different types, sonars, GPS, high-resolution cameras and internal sensors. The laser, for example, measures the distance to objects and their dimensions; Radars and sonars are used for adaptive cruise control, but are also needed to determine the distance; cameras receive images for their subsequent study by a computer (signs, markup, etc.). Then the software analyzes the incoming data, is guided by rigidly defined rules, and sets of changing algorithms, predictive models, distinguishes objects that are important for further actions, or ignores them.
Based on the information received and processed, the car carries out further actions on the road, all decisions are made in fractions of a second (the same motorcyclists always appear out of nowhere).
Machines can act as part of the “swarm” – this principle, we recall, was viewed in a slightly different form as one of the main ones at the dawn of the development of self-governing mechanisms. In the “swarm” cars are able to form a dense column (transportation of goods) or to act even more globally – at the level of the entire transport of a certain city (but only in a very distant future).
The degree of autonomy is estimated on a scale from 0 to 5. At the zero level, all major systems are controlled by humans; on the first, some systems, but not more than one in a particular period of time, can be controlled by electronics; on the second machine takes responsibility for several management components; on the third, it is able to control itself, but in some situations it transfers control to a person; on the fourth, complete autonomy is achieved in a limited number of scenarios; The fifth level, to which everyone aspires, implies autonomy in any conditions.
Now back to Google. The company set a goal to produce an autonomous machine by 2020. The project began with six cars Toyota Prius and Audi TT. They were ruled by people with an impeccable history of driving, so that the system adopted exclusively correct models of behavior.
The machines were equipped with a set of sensors, radar and sensors necessary to monitor the environment. By 2010, these cars drove more than 225 thousand km, recording all kinds of situations that can only be met on the road. The company Google, however, did not spread too much about its project, but on the streets instead of “Prius” came “Lexus”.
And in 2014, the prototypes of Google’s autonomous cars were presented to the public – such “ladybugs” without the wheel and pedals, only with a pair of seats, a power button and a rack with sensors on the roof. Called Firefly (“Firefly”), they were intended for testing, but not mass production.
The achievements of the search corporation did not give rest to the large automakers, practically everyone (almost all of them continued their research) joined the game.
In 2014, Tesla introduced its Autopilot, raising as usual a HYIP around a new direction for itself. At first it was a relatively simple system consisting of a camera under the windshield, looking ahead of the radar, a circular ultrasonic sensor and a computer for data processing. The first version of Autopilot met the requirements of the second level of autonomy.
The second version was overgrown with cameras, increasing the range and range of coverage, and received a more productive “hardware” from NVIDIA for data processing, and later – developed in Tesla processor. However, this is still a super-advanced cruise control. A similar analogue exists for Mercedes. It was planned that Audi A8 2019 will meet the third level of autonomy, but has not yet developed. The whole issue is in laws and software. Both Tesla representatives and industry experts speak about this. According to the latter, the company Ilona Mask achieved the greatest progress in the commercialization of the product.
The so-called Enhanced Autopilot of the American company is ready to provide the fifth level of autonomy (but officially on the second level), it only remained to “roll up the update” than Tesla and is engaged in gradually improving the software. On the testing grounds, the system is tested, which operates much more efficiently than developments available to electric car owners.
While ordinary drivers become hostages to their own expectations, trying to see in the cruise control Tesla full autopilot. For this reason, we regularly read about accidents involving electric cars Mask – negative role here is played by aggressive marketing of a businessman. Despite the requirement not to let go of the steering wheel, the result is sometimes very sad.
Of course, quite often “Autopilot” helps:
Google, meanwhile, is on its heels – the development of self-controlled machines is now being handled by its affiliated enterprise Waymo. Technologies and hybrid data processing systems allow test cars to operate at the fourth level of autonomy, but they have not reached public roads yet. Unlike Tesla, which abandoned the lidars in favor of cameras and machine vision, Waymo and Uber (also involved in the race of autopilots) use active optical systems. Their shortcoming is technological imperfection and high price.
Experts believe that the introduction of artificial intelligence in autopilots is inevitable. Advanced AI with a similarity of cognitive functions will be able to simulate the human style of driving. What for? This is a matter of comfort. Also, some scientists believe, the development of the concept of “Internet of things” will contribute to the emergence of autopilots of the fifth level – the global information network will be able to generate even more data for processing.
In theory, autonomous transport can significantly reduce the level of accidents, improve the efficiency of cargo delivery, make personal transport less attractive, which, in turn, will affect the ecological situation. That’s what engineers were talking about, just starting to create autopilots for cars.
According to short-term forecasts, the most active autonomous systems will be applied in the sphere of railway and sea transportation, delivery of goods on public roads, as well as in the organization of urban public transport traffic (tests are already underway).
Today the development of smart autopilots is being done by Waymo, Uber, Tesla and an incredible number of startups. Their main goal is to create algorithms for information processing, since the “hardware issue” is considered solved. Such systems will become one of the components of the global infrastructure – gradually from smart homes we will move on to smart streets, cities and states, but this next time.