cart was originally a test platform to study how to possibly remotely control a lunar rover from Earth. The Jet Propulsion Laboratory was developing a lunar rover that would be controlled by radio signals from Earth, and information sent back from a television camera. Grad student James Adams built the cart in 1961 to simulate the rover, taking into account the 2.5-second delay in the signals going to and coming back from the moon. He was able to prove that with the delay, the rover would not be reliably controllable at speeds over 0.2 mph, which is, as you can guess, really, really slow.17
To overcome this, experiments began to attempt to give the cart its own ability to “see” its environment, detect obstacles, and take steps to avoid them. This was the birth of nearly all computer vision systems employed by autonomous vehicles (and, really, any robot that uses some manner of camera-based synthetic vision) today.
By 1964 the cart had been re-outfitted with a low-power television transmitter that broadcast TV signals to a PDP-618 computer to process the images. With this setup, which I’m dramatically simplifying here, the cart was able to visually follow a high-contrast white line on the road at about 0.8 mph. This was a big deal, as it represented real computer vision controlling a moving machine, even if it was quite crude.
Development continued on the cart with new researchers and students re-outfitting the cart as new ideas and technologies became available. In 1977, the cart was upgraded with faster processors, an independently mobile camera, and four-wheel steering; in this configuration it was able to drive around obstacles in a controlled environment.
Sure, it only moved three feet at a time and then had to pause to figure out what it was looking at, but this was a gigantic leap in robotics. The cart was eventually able to navigate a chair-filled room in about five hours, and while it may be tempting to laugh at that idea now, I can think of plenty of times I’ve not been able to navigate a chair-filled room without running into half the chairs and looking like an idiot.
1977: Tsukuba Mechanical Engineering Lab, Japan
Arguably the first fully autonomous, computer-vision-controlled car was shown in 1977 by the Tsukuba Mechanical Engineering Lab, in Japan. The project, headed by Sadayuki Tsugawa,19 modified a full-size car to follow special white road markings and was able to drive at speeds of nearly 20 mph. While still essentially a follower of specially contrived external visual guides, the fact that this technology was implemented in a full-size car driving at a reasonable speed (compared to, say, the Stanford Cart) and using computer-interpreted visual information made this a significant milestone.
1980s: Ernst Dickmanns: The Man Who Made Cars See
If the overall concept of vehicles driven via true computer “vision” can be said to have a father, that father would have a German accent and a hilarious last name: Dickmanns. Ernst Dickmanns was the father of a particular set of Mercedes-Benz cars and vans that drove via information captured from cameras and interpreted by some very hardworking computers.
Dickmanns started out working in aerospace, including a stint at NASA, where he researched orbiting spacecraft reentry. By the early 1980s he’d migrated to focusing on the development of machine vision to allow autonomous driving.
Dickmanns’s first real application of his research was the result of a partnership with Mercedes-Benz, which was hoping to have something really exciting to unveil for their centenary in 1986: a self-driving car. To achieve this goal, Dickmanns outfitted a Mercedes-Benz L508D T2 van with a lot of computing hardware, cameras, servos, and actuators to operate the van’s driving controls.
Since it was 1986, the computing hardware, while state-of-the-art, wasn’t really fast enough to process a full visual field captured by the camera in real time—a full one hundred seconds was required to process a full-frame image from the camera. To get around this, there was a master computer, but actual processing of the images from the camera was passed to a parallel-processing system consisting of ten Intel 8086 CPUs—the same ones that powered the IBM PC AT.
These ten CPUs would only process certain tiny 32 by 32 pixel areas determined to be interesting in some way. The result was that the important parts of the camera’s feed (high contrast areas, areas in motion) could be processed much quicker, with only twenty microseconds needed to extract the important and required information from the visual feed.20
Dickmanns called his machine vision solution “Dynamic Vision,” and considered it a “4-D” approach to machine vision, incorporating both strict processing of the video feed and time delays, as well as predictive behavior information for certain classes of identified objects, things like “spatiotemporal models for motion processes of objects,” and more.21
Dickmanns’s first autonomous van, called the VaMoRs, first drove autonomously on non-public sections of Germany’s Autobahn in 1986. By 1987 the vans were on public roads driving at nearly 60 mph; collision avoidance and obstacle detection were implemented, allowing the van to follow a car in a convoy if desired.
Конец ознакомительного фрагмента.
Текст предоставлен ООО «ЛитРес».
Прочитайте эту книгу целиком, купив полную легальную версию на ЛитРес.
Безопасно оплатить книгу можно банковской картой Visa, MasterCard, Maestro, со счета мобильного телефона, с платежного терминала, в салоне МТС или Связной, через PayPal, WebMoney, Яндекс.Деньги, QIWI Кошелек, бонусными картами или другим удобным Вам способом.
