Driverless Cars Have Been 76 Years in the Making — A Look at Their Past and Present | QDaily Business History
Taking your hands off the wheel is closer than you think
Taking your hands off the wheel is closer than you think
In Silicon Valley, a car with nobody behind the wheel driving past you is no longer news. Here's one I ran into a couple of days ago on the streets of Mountain View, where Google is headquartered:
They usually aren't fully "driverless" — there's a test engineer sitting in the passenger seat.
At Google's annual developer conference, Google I/O, coming up shortly, we should see a major update to the car operating system Android Auto. For driverless cars, having an operating system that can be upgraded and extended with new software, the way a smartphone can, will be the first step.
On the way to the driverless car, we have already watched internet companies, automakers and chipmakers all try to make cars smarter. Besides Google, the traditional automaker BMW has launched the semi-autonomous i3, Tesla has followed with a semi-autonomous version of the latest Model S, and even Qualcomm and Nvidia, which make chips, have started supplying automakers with chips for autonomous-driving technology.
But how long has it taken the driverless car to get this far? We have to start counting 76 years ago.
1939: The First Autonomous-Driving Exhibit
You read that right. At the 1939 World's Fair, General Motors built a future-technology exhibit called Futurama to show what it thought cars would become. GM imagined that by 1960 American highways would all be laid out something like railway tracks, and that cars would carry an autonomous-driving system that switched on by itself once you reached the highway, following the track at a set speed until the exit, where the human driver took over again.
Crowds lining up to enter GM's Futurama pavilion in 1939
In 1939, people imagined autonomous driving as a car and a highway working together. That is because Germany and Italy had only begun building the world's first motorways around 1920; before that, the fastest roads on earth were still railway tracks. That was where Norman Bel Geddes, the designer of GM's pavilion, found his idea: working from Autobahn No. 1, built in Germany in 1920, he designed the Futurama hall, complete with a fake "aerial viewing platform" from which visitors could look down on how the tracked highway system of the future would cover city and countryside alike.
The aerial viewing hall inside GM's Futurama, image courtesy of General Motors
But what GM and Norman Bel Geddes hoped for never arrived. Eighty years after the Futurama exhibit, neither autonomous-driving technology nor the highway had changed much.
Concept vehicles kept coming, though. In 1958 GM unveiled an autonomous-driving concept car (really a plane) called the Firebird 3. Here is the advertising copy of the day:
"Want to sit back and relax? Fine — set the speed you want, then switch it to autopilot. Let go of the controls, and the Firebird 3 takes care of the rest."
Concept art of the Firebird 3, image courtesy of General Motors
2004: DARPA Puts Autonomous Driving Back in Motion
In 2004 the U.S. government and military began looking at autonomous driving again. Aiming to have at least a third of military vehicles driving themselves within ten years, the Defense Advanced Research Projects Agency (DARPA) started running an annual contest with a $1 million prize, drawing people from everywhere to build autonomous cars.
The contest got better every year. In 2004 the best autonomous team's vehicle wrecked itself after a few kilometres; the next year, all sorts of autonomous cars turned up, and some of them crossed California's Mojave Desert without a scratch.
This is Stanley, the team that won the top prize in 2005. The man in the blue jacket in the middle is Sebastian Thrun, then director of the Stanford Artificial Intelligence Laboratory. Thrun lost a friend in a car crash when he was young, and it was after that that he began working on autonomous-driving technology.
In 2009 Sebastian Thrun joined Google X to work on Google's driverless car, and he now runs Google's autonomous car project.
Semi-Autonomous Cars Are Already Here
Unlike Google, which is set on going straight to full automation, automakers have mostly worked step by step: fit the car with the hardware, add smart systems, and get to partial or conditional autonomous driving that way. So which cars that drive themselves to some degree are already on sale?
Germany's Daimler reached a deal with the state of Nevada to let its autonomous trucks on the road. But these new trucks still carry a driver, who takes over in certain awkward situations — pulling in and out of a dispatch centre, for instance, or driving through busy city streets. So they run mainly on long-distance highways, where they spare the driver the fatigue.
Fitted with Volvo's City Safety system, plus a new run-off-road protection system and automatic braking at intersections, to help drivers avoid the everyday rear-end collision.
The BMW i3 carries four IBEO laser sensors on its body, which can pick out obstacles such as warning posts and make sense of the surroundings inside a multi-storey car park, steering around pillars and braking on their own. The car can also stop within a few centimetres of an obstacle, which makes automatic parking possible.
Last October Tesla said the latest Model S would add an Autopilot mode. Using long-range radar, image recognition and ultrasonic sensors, it lets the car reverse by itself, spot obstacles on the road and drive semi-autonomously on the highway, among other things.
Fitted with the Intelligent Drive system, it can avoid collisions, hold the car in the middle of its lane, spot pedestrians and brake on its own to prevent a full-on crash above 50 kph, and follow the car in front automatically in a traffic jam.
Google's Driverless Car
In 2012 the Google driverless car — built at the Google X lab by a team led by Sebastian Thrun, who had taken part in DARPA's driverless car contests — went into internal testing in the United States.
So how does Google's autonomous car actually work? At the heart of it are an assortment of sensors and Google's map data. This is what Google's previous-generation autonomous car looked like:
Photographed at the Computer History Museum in Mountain View
The parts that go into Google's driverless car:
- Laser: senses the car's surroundings through a full 360 degrees, so the car can pick up objects ahead, to the side and behind it at the same time, without a break. It also helps the car locate itself. This is what the laser looks like taken off the roof:
- Position sensor: mounted inside the wheel hub, it fixes the vehicle's position.
- Orientation sensor: sits inside the car and works like the human inner ear, which keeps the body balanced, guiding the direction the car travels in.
- Radar: mounted on the bumper, it measures the distance between the car and several obstacles at once and can tell the system to slow the car down. It is the small black box at the front of the car:
According to Google, these radars and sensors working together can detect traffic as far as two football fields away. Everything the sensors gather goes straight back to Google's servers, which pick the best route and steer the car along it.
How Far Off Is the Autonomous Car?
The first thing keeping autonomous cars off the road is the law.
In the United States, home of Silicon Valley, only four states currently allow autonomous cars on the road, and the questions that worry people more — how to regulate autonomous cars, who is at fault in a crash, user privacy and safety — have gone unanswered for years. Google was already pushing California's government in 2012 to revise its rules for autonomous cars, but as of this year, past the bill's own deadline, the draft legislation still has not appeared.
The good news is that the United Nations recently amended the Vienna Convention on Road Traffic, so that 72 countries in Europe and the Americas can pass rules allowing a car to drive itself for certain stretches. Road traffic law still requires that the driver be able to control the vehicle and override the autonomous-driving function by hand.
Even once the legal questions are settled, the technical and cost questions remain:
The high price of sensors holds manufacturers back and pushes them towards building "partly automatic" cars. At this year's CES, the i3 concept car BMW demonstrated used four laser sensors made by IBEO. According to published figures, sensors like these cost more than $20,000 each.
Sensor prices as they stand do not suit mass production, but Chris Urmson, who runs Google's driverless car project, thinks the cost will come down soon. Ragunathan Rajkumar, a professor at Carnegie Mellon University, reckons that within three years the cost of Google's autonomous-driving technology will drop to $10,000-15,000 a car — which would do more than anything else to pull the autonomous car into the mainstream market.
Beyond that, making a car's operating system upgradable and extendable the way a smartphone's is will be the groundwork for autonomous cars, and Android Auto looks like the system moving fastest. At this Google developer conference we may well see a new shape for Android Auto — using the full range of Google services in the car without plugging in a phone, for instance — and the Volkswagen Santana could become the first mass-production model to carry it. Google's autonomous-driving technology, six years in the refining, will also put its latest prototype on public roads for testing this summer.
For Google, the point where Android Auto and autonomous-driving technology run together is getting closer. For more, look out for QDaily's live coverage from the Google developer conference tomorrow morning.
Translated from the original QDaily article. Original publication date retained.