This article was taken from the February 2011 issue of Wired magazine. Be the first to read Wired's articles in print before they're posted online, and get your hands on loads of additional content by subscribing online.
Crash avoidance and automatic parking are just the start.
Tomorrow's cars will be brains on wheels.
The 300km trip from San Francisco to Lake Tahoe can be a frustrating slog in the wintertime traffic on Interstate 80. Speeds in the fast lane swing from 120kph to 50kph for no discernible reason. Slow, fast, faster, slow. Hit rush hour in Sacramento -- or Donner Pass on a snowy day -- and you'll see the speedometer's needle tapping the 15kph mark like a woodpecker on a tasty log.
Driver of manual cars collapse with dead legs on the side of the road; even the automatic-enabled P-R-N-D crowd can be seen massaging their sore knees at roadside burger joints and woodsy rest stops.
Not me. I'm playing the licence-plate game and humming through playlists with a few friends, happy and comfortable in a borrowed
Mercedes-Benz S550, a luxury car that's currently justifying the pants off its $100,000 (£63,600) price tag. We're bopping through the same unpredictable range of velocities as everyone else, but I haven't touched a pedal in hours.
The Benz is doing most of the driving, keeping us a comfortable distance from the cars ahead with its next-gen cruise-control system. The core of the set-up is a pair of radar emitters -- a narrow-banded one that pings vehicles up ahead, and a wide-angle unit that watches the rest of the traffic and keeps a sharp eye out for road-hogs weaving into our lane. All that locational info is fed to the car's vehicle-control unit, a computer that smoothly modulates the brakes and throttle to keep us moving with traffic.
The driver specifies a maximum speed, and the car does its best to hit that number -- without hitting anything else.
The first time you let the car do its thing is a magically scary experience. You see the cars ahead closing at a rate that activates the "I'm going too fast" reflex; your foot hovers over the brake pedal as your frontal cortex strenuously attempts to override your survival instinct. Cognitively, you know that this system has been meticulously tested by obsessive German engineers who would never let an unsafe car cross the threshold of their shiny factory.
And then, just as you're beginning to contemplate the various safety regulations that the car must have complied with on its way to the dealership, you feel yourself slowing -- gently, autonomously, in perfect control. The cold cannonball in your stomach turns back into warm muscle, and you chuckle softly to yourself for being so silly as to doubt such a well-engineered system. Getting used to these autonomous systems takes time. It turns out that we have to adapt to the machines more than they have to adapt to us.
Cruise control is just the most obvious sign of a particular kind of AI that has been accelerating for decades. Think about it.
Anti-lock brakes know when to back off the pedal. Airbags know that you just smacked into something. Stability control knows that you just overcooked your Volvo into that hairpin and need a little help to stay out of the ditch. Your satnav system knows where you are, your wipers know it's raining, that annoying seat-belt chime knows you're flouting the law. In short, modern cars are loaded with sensors and computing power. The 2011 Chevy Volt, for example, runs on some ten million lines of code -- more than Lockheed Martin's new F-35 Joint Strike Fighter.
The marquee innovation that made intelligent cruise control possible is the drive-by-wire throttle: the introduction of motor skills to the automotive body. The throttle is a flap that lets air and fuel enter the engine. In the conventional setup, it's linked to the accelerator by a thin metal cable threaded through a grooved wheel. This went unchanged for decades -- but many newer cars have done away with the cable. Instead, there is a sensor on the accelerator and a small electric motor on the throttle.
Step on the accelerator and an electrical impulse travels to the computer, telling it how far the pedal is depressed; the computer then tells that little electric motor how wide to open the flap.
Electronics and software are mediating the whole process. Voilà: you're driving by wire.
Of course, by-wire technology isn't just for throttles. The same exquisitely sensitive actuation systems are finding their way into brakes and steering as well. And where there are electronically controlled systems, there are sensors and software and processors that can command them. In other words, by-wire technology is paving the way to truly smart cars.
Drive-by-wire didn't start in the automotive industry. It's a descendant of an aerospace technology called, yes, fly-by-wire. The first aircraft to fly with it -- a Canadian fighter jet called the Avro Canada CF-105 Arrow -- took off in 1958. Most of the pilot's controls, from the elevators to the rudders, were triggered electronically.
The advantages -- instantaneous response and lighter weight -- were compelling. Within a few decades, many commercial airliners were using fly-by-wire technology. It made every aircraft from the Concorde to the Boeing 777 possible and was integral to improving autopilot systems -- including those that can land a plane. It's nice to have Captain Sullenberger (AKA The Hero of the Hudson) on board, but he's only needed on special occasions.
The by-wire throttle first made its way into cars in 1988, in the BMW 750iL, and it now makes radar-assisted cruise control possible in any number of Fords, Volvos, Jaguars and Mercedes. Some hybrids rely on it to switch nimbly between petrol-driven and electric power. But drive-by-wire technology has applications beyond the car-pool lane that conjure scenes from a sci-fi future: self-driving vehicles that promise the end of traffic jams and a major reduction in battlefield casualties.
In 2004, Darpa, the US Defense Department's research arm, challenged the big brains of the world to come up with a car that could navigate a complicated desert course with no human input. Employing technologies closely related to our smart cruise control -- electronic eyes, computer brains and drive-by-wire legs -- 15 teams vied for the million-dollar prize. None finished. But that didn't stop Darpa from throwing down the gauntlet again. It hosted another challenge the following year, and five of the 23 teams finished.
Moore's law hits the road.
Some 212 kilometres and nearly seven hours after the second Grand Challenge began, the first car across the finish line was a self-driving Volkswagen Touareg named Stanley -- one of the smartest cars ever built. Sebastian Thrun headed up the Stanford team that trained Stanley for its victory and ran head-on into the primary obstacle facing any self-driving car. "You literally can't even count the number of different situations a driver encounters,"
Thrun says.
That's why his team didn't try to code a solution for every situation. They taught Stanley how to drive the old-fashioned way. "We took the car out on the road and logged every time it made a mistake." Back at the lab, Thrun's team used this data to replay failures and challenges over and over again in the car's software mind as it simulated different solutions to each puzzle. Every time it failed or succeeded, it learned why, and improved.
Thrun has since taken a post at Google, where he and a team of engineers are testing a small fleet of autonomous Toyota Priuses on the streets and highways of the densely populated San Francisco Bay Area. (Someone sits behind the wheel of the Google cars, ready to take control if necessary.) Of course, you can't just go out and purchase a robo-vehicle today. Hell, you're probably still apprehensive about radar-assisted cruise control.
It turns out that the US federal agency charged with ensuring auto safety -- the National Highway Traffic Safety Administration (NHTSA) -- shares that fear. The NHTSA isn't going to green-light self-driving cars without a lot more trials and oversight. "It's not at the point of being sufficiently reliable for the consumer market," says NHTSA spokesperson Eric Bolton.
Still, the autonomous systems migrating into vehicles are impressively reliable -- they make far fewer errors than humans.
Plus, there's no convincing evidence that people will let down their guard when a robot is doing the driving for them, a phenomenon that's known as risk compensation. "Do they engage in risky behaviours -- texting, applying makeup, shaving?" asks Jim Sayer, who investigates real-world driver behaviour at the University of Michigan Transportation Research Institute. "We never see that."
The real problem arises when millions of humans are confronted with autonomous systems -- and some of them freak out. That seems to be what happened recently with some Toyota cars. In a number of well-publicised cases, drivers thought that the electronic throttle was improperly accelerating. It turned out that most of the incidents were caused by an all-too-mechanical flaw in the floor-mat or in the accelerator pedal's design -- or by driver error.
Avoiding those errors is a tricky dance that takes time to learn. Consider the new self-parking technology, brought to the US market by Lexus and since adopted by other carmakers. On a busy city street, I pull a Lincoln MKT (borrowed, again) alongside an empty space and hit a button labelled "auto |p|". A two-line LCD on the instrument cluster explains what to do: "Select reverse and take your hands off the wheel."
I follow its commands and the car takes control, whipping the wheel around and backing into the space faster than I would ever attempt. I tell myself to relax, to let go, that this SUV has more sensors than a satellite -- a beeping proximity sensor in the back, a rear-facing camera, radar sensors that inform its own magic cruise control. And just as I surrender myself to the future, the Lincoln slams into the car behind me.
A rep from Lincoln later tells me that you're supposed to work the brake as the car steers itself. And yeah, that two-line display never suggested I take my foot off the pedal; I guess I just assumed that "auto park" meant, you know, auto park. This machine-man language barrier is something we're really going to have to work on.
This article was originally published by WIRED UK
