When Robert Whitehead invented the self-propelled torpedo in the 1860s, the early guidance system for maintaining depth was so new and essential he called it “The Secret.” Airplanes got autopilots just a decade after the Wright brothers. These days, your breakfast cereal was probably gathered by a driverless harvester. Sailboats have auto-tillers. Semi-autonomous military drones kill from the air, and robot vacuum cleaners confuse our pets.
Yet one deceptively modest dream has rarely ventured beyond the pages of science fiction since our grandparent’s youth: the self-driving family car. Unlike Mars rovers or sailboats, cars need to navigate the complex world of city streets, passing inches away from fragile, litigious human beings. This article explores both the history of autonomous vehicles in general, and that elusive goal of a car that drives itself. Several groups say they are now close to making it a reality. If they succeed, how will they change our world? Could autonomous cars replace public transportation? Would they make our cities more walkable, or supersize them with unimaginable sprawl?
Draft animals and distracted pedestrians can usually keep to a path on their own. But with the first self-propelled vehicles came the need to have an alert human guide the craft at every moment, or risk disaster. The modern experience of driving was born – that peculiar mix of anxiety, alertness, and boredom.
Sailboats were likely the first self-propelled vehicles, and possibly the first to have some form of automated steering, the auto-tiller. This device uses ropes to connect something like a weathervane to the boat’s tiller, so that the craft stays on course even with shifting winds.
The first widely used motorized vehicles were steamboats and trains. The latter adopted their guiding tracks more to support their huge weight than for directional control, but tracks serve both ends. Just a decade or so after its invention, the airplane got its first autopilot.
The kind of self-guiding that carried torpedoes to their targets was repurposed for another medium – the air. By the early 1940s the German V-1 drone bomb was buzzing its way to London on stubby wings. Its successor, the V-2 rocket, touched the edge of space itself.
Driverless cars and taxis have been improving the lives of millions in the pages of science fiction since 1935. Joined by GM’s automated highway plans in its seminal 1939 Futurama ride, the basic driverless dream has changed little in the ensuing decades. Besides reducing accidents and congestion, such cars might liberate city centers by eliminating the need for most parking.
Of course, in the pre-computer days of the 1930s, giving cars meaningful smarts was literally the stuff of science fiction. But there might be other ways….
Much of the danger of early motoring was not the cars but the era’s narrow, ill-marked roads, designed mostly for local travel. Railroads were still the superhighways. By the 1920s, a few began to dream of transforming roads into something more like a modern freeway system, where controlled access would simultaneously raise speeds and reduce accidents.
Italy’s Autostrada and Germany’s famous Autobahn plans stopped there. But American designer and futurist Norman Bel Geddes mated the Autobahn vision with the sorts of electronic speed and collision control systems common to railroads. His spectacular Futurama ride for General Motors at the 1939 World’s Fair also imagined trench-like lanes that would keep cars apart in their own “tracks.” The idea was to drive to the freeway normally, then engage the automatic systems and kick back until your exit. Related visions involved magnetic trails built into the road’s surface, or physical slots or troughs, or train-like rails engaging hidden steel wheels on the inside of each tire.
80 years on the two basic ideas – smart cars, and/or smart roads – have changed little. Prime goals remain safety, speed, access, more cars sharing the road, intelligent intersections, and reducing congestion.
If you’ve ever seen a cockroach, you know that even insect nervous systems are capable of navigating through a complex environment at tremendous relative speed. A car-sized cockroach would be running – and turning, and dodging – at over 200 mph. Giving cars just a fraction of those navigating capabilities has taken 50 years.
It’s no accident that autonomy came to other vehicles first. However distant or exotic, the sea, the air, and even the surface of Mars are relatively forgiving environments for self-guiding vehicles. There are no children to dart out in their path; no traffic lights, or distracting billboards. Mostly, there’s just a lot less delicate stuff rushing by in close proximity – other vehicles, pedestrians, outdoor restaurants, flimsy wooden buildings.
Because making cars smart is so hard, early self-driving plans focused on special freeways for guiding suitably equipped cars safely along them, more railroads than robots. The technology was projected in the 1939 Futurama and showcased by the 1950s. But getting the massive consensus needed to build public infrastructure never happened.
The digital computer promised to make vehicles smart in ways rarely imagined outside of fiction. One of the first uses was guidance computers for nuclear missiles. Bulging cold-war budgets let designer build these with still bleeding-edge semiconductors instead of fragile vacuum tubes.
By the late 1960s experimental robots were navigating through novel environments at SRI and Stanford, [testing out still-new AI techniques]. By 1971 semi-autonomous space probes were landing on other worlds. Had a parachute not failed to deploy, the Soviet Mars 2 rover might have been crawling the surface of Mars on its own that year. Today, autonomous underwater vehicles can roam the seas for years at a time. The Voyager space probe, launched in 1977, recently became the first human object to travel beyond our solar system.
By the 1960s, enthusiasts of artificial intelligence (AI) on computers began dreaming of cars smart enough to navigate ordinary streets on their own. The challenges were daunting – essentially to reverse-engineer the relevant systems in a moving animal like a cockroach:
1) Sensing 2) Processing (modeling the outside world, making decisions) 3) Reacting, with appropriate movement The first and last steps were feasible with known technology. The unknown part was the processing, the machine intelligence needed in between.
Much of that challenge was about interpretation. A car whose ”mental” model mistakes a pedestrian for her reflection in a puddle can be a dangerous thing indeed.
Early AI pioneers dreamed of breakthroughs that would bring human-like robots by the millennium. But real progress was more incremental than revolutionary. In the 1980s, German pioneer Ernst Dickmanns got a Mercedes van to drive hundreds of highway miles autonomously, a tremendous feat especially with the computing power of the time. Around the world, dozens of other pioneers added their own improvements.
In 2004, the U.S. Defense Advanced Research Projects Administration (DARPA) challenged dozens of teams then working on autonomous vehicles to compete for a $1 million prize. The hope was that a third of military vehicles would drive themselves by 2015.
The first year’s crop of entrants failed miserably, traveling barely a few miles before crashing. But the next year an odd flotilla of driverless cars and trucks were crossing huge swathes of California’s Mojave desert with nary a scratch. By 2007 the Urban Challenge had extended those successes to a mock city environment. While European researchers had laid the groundwork in self-driving, the U.S. was now a serious contender. Several factors made the difference: Better software for road-following and collision avoidance, and improved radar and laser sensors. Good mapping also helped. While machines lag behind animals in interpreting their environments, a car that always “knows” what’s around it can focus its interpretive skills on variables that change.
You can buy a self-driving car today. It’s called the Navia, and there are some limits. It’s only designed for closed environments, like a resort, and its top speed is 12 miles an hour, or about the same as gasoline powered cars in 1895. If you’re not ready to buy your own robocar – the Navia costs $250,000 – you can still ride in another example at London’s Heathrow airport.
Like many emerging technologies, self-driving has found uses in specialized applications long before reaching the general public. In the pit mines of northern Australia, trucks the size of a spacious house rumble over gravel roads without a human touch. Combine harvesters and other farm vehicles are increasingly outfitted with self-driving capabilities, as are specialized vehicles in warehouses, factories, and other industrial environments.
Don’t forget that more and more self-driving features also come as options on high end conventional cars, like the BMWs and Volvos that keep lanes, self-park, and brake for emergencies. While their manufacturers are eager to point out that such cars augment your skillful driving, rather than replace it, some systems are getting so powerful that distinctions blur.
Google, of course, is famously working on self-driving systems for the open road, with full autonomy as an explicit goal. But from Toyota to Nissan, several other companies are quietly chasing very similar dreams. Three states now permit self-guided vehicles. Will the autonomous future arrive not with a bang, but so gradually we barely notice?
The computer systems that make self-driving cars work have gotten smaller over the years, and morphed from huge installations to discreet units mounted in trunks. They all handle roughly the same functions: taking in raw data from sensors, matching it with known models, and executing the appropriate movements by sending signals to the car’s steering, throttle, brakes, turn signals, and other controls.
If you spend much time in the Bay Area you’ve shared roads with experimental self-driving cars dozens of times, perhaps without realizing.
By the late 2000s Google had moved beyond its core search business in several ambitious directions, from scanning the world’s books to smartphones. Founders Page and Brin especially liked to take on pivotal problems a few uncomfortable but exciting notches short of being solved, what they termed “moon shots.”
An early example had been Street View for Google Maps, co-developed by driverless Grand Challenge winner Sebastian Thrun. In 2008 he encouraged another self-driving veteran on his team, Anthony Levandowski, in a side project called the Pribot – a Prius modified for the stated goal of fetching pizza on its own.
The Pribot’s success helped convince Google’s founders that self-driving, too, might be a technology on the cusp. They assigned the Pribot [and its successors?] a series of challenges such as driving 100,000 miles on public roads, and even descending San Francisco’s twisty Lombard Street. It passed, and they made Thrun co-head of a new effort, Google X, geared to launching “moon shot” ventures.
Thrun recruited Levandowski and many of the other top researchers in the field, including Urban Challenge star Chris Urmson. The team began the hard work of transforming the raw capabilities demonstrated in desert Challenges into a consumer system; one polished enough to safely carry living passengers in the real world of traffic, and commuting, and family vacations.
Google’s system has guided a fleet of Prius and Lexus vehicles over half a million miles without causing any accidents, and the firm is a leading advocate for fully self-driving cars. But as to how it might deploy the technology, Google is keeping its options open.
If all of the billion or so cars in the world magically taught themselves to drive between now and tomorrow morning, what would actually change? Right away we’d have more free time, and mobility for those who can’t drive. But deeper changes would require new business models, laws, and ways of thinking about transportation; maybe even new infrastructure.
If our cars took their own wheels, we could text, tweet, and even play driving games as we lived the lifestyle only a few chauffeured executives can afford today. But beyond that, we would have to change the ways we think about driving and transport.
Would Americans and Germans, especially, be OK with giving up their century old driving cultures? Would we send the cars to pick up the pizza, or put our children in one to get to soccer practice? Would we comfortably share a driverless car with strangers? How would we manage the transition – when self-driving cars first mix with piloted ones?
Today’s wish list for self-driving has changed little since the 1930s. We want safety, speed, frictionless transport, no congestion, new balances of public spaces; perhaps even road trains and intelligent routing to augment or replace public transport.
But that doesn’t mean building out Bel Geddes’ Futurama. It’s worth remembering that much of that ‘30s dream would appear as a nightmare today. The cities cut through with multilevel highways; the giant roads eating up distance and landscape.
While we never got self-driving, we’ve lived out plenty of the ugly aspects of the Autobahn and Futurama visions, and our tastes have moved on. We want the benefits, yes, but in forms adapted for our time. Self-driving could be a transformational tool, and that makes it an essential one to use wisely. Perhaps it’s time to look ahead with a new Futurama.
If a self-driving car was waiting outside your door today, where would you go? On a pleasure tour of the countryside, or to your next chore? If you have some distance to travel, perhaps you’d like your car to join a high speed road train, a convoy that would save on gas and road space as it whisked you to your destination at over 100 mph.
How would you while away the minutes or hours as you drove? Maybe you’d watch a movie on a flat panel TV while sipping a cold drink, or surf the Web or chat on video, or stretch out for a luxurious nap. A few years from now, you might be able to order a lovely snack with voice commands and have it delivered piping hot by a drone that could match speed with your car to transfer your order.
You can use the comments section of this blog to add your thoughts.