Hands-Off Training: Google's Self-Driving Car Holds Tantalizing Promise, but Major Roadblocks Remain
Driverless automobiles lack common sense but are getting better at using mapping, GPS and sensing technologies to hold the road
KING (ROBOT) OF THE ROAD: The Google fleet consists of conventional vehicles--six Priuses and one Audi--that have been outfitted with off-the-shelf components consisting of two forward-facing video cameras, a 360-degree laser range finder, four radar sensors and advanced GPS units.Image: COURTESY OF GOOGLELong a staple of science fiction, self-driving vehicles that act as robot chauffeurs have been a cultural dream for decades. For most of that time, however, the dream seemed a part of some unattainable future.
But now, led in large part by Google's sudden and unexpected charge, autonomous robot cars come tantalizingly close to reality. As various mapping, sensing and location-based technologies have converged recently, Google has begun to position itself as the leader of our robo-chauffeur future. Yet for all of the technology's promise, it still has some major—and perhaps insurmountable—hurdles to overcome.
Google estimates that one million lives could be saved around the globe by driverless cars each year. According to the National Highway Traffic Safety Administration (NHTSA), in the U.S. alone there were 5.8 million crashes in 2008. Of those, about 34,000 resulted in fatalities, 1.6 million resulted in injuries and 4.2 million entailed some sort of property damage. The NHTSA says these numbers have come down over time—attesting at least partly to the ever-increasing safety of all vehicles—but they clearly still account for a large amount of deaths, injuries and property damage that driverless cars could drastically reduce.
Gigantic leaps with off-the-shelf components
Forty years ago the first self-driving cars were little more than crude, slow-moving contraptions following lines painted on the road. The past several years, however, have seen accelerated success in the quest for autonomous road vehicles, starting with a series of self-driving challenges the Defense Advanced Research Projects Agency (DARPA) held between 2004 and 2007 to help develop robots that could replace some U.S. military personnel on the battlefield.
Last year, a group of engineers from Stanford University Dynamic Design Laboratory, the Electronics Research Laboratory for the Volkswagen Group and software-maker Oracle Corp. shifted driverless cars into a higher gear by successfully running their 265-horsepower Autonomous Audi TTS Pikes Peak research car up the 20-kilometer course of the Pikes Peak International Hill Climb race in Colorado without stopping. The sporty hatchback—Volkswagen owns Audi—carried two computers in its trunk, both using Oracle's Java Real-Time System to run algorithms that kept the car on the road and at the limits of its handling ability on the varying surfaces and conditions.
Google's self-driving car project exited stealth mode last year and now the company is lobbying Nevada to be the first state to allow self-driving cars to be legally operated on public roads. To this point, Google's robot fleet has traveled more than 240,000 kilometers with minimal human intervention and only one incident in which a test car was rear-ended by another (human-driven) vehicle. Unlike the futuristic sci-fi images splashed on movie screens, the Google fleet consists of conventional vehicles—six Priuses and one Audi—that have been outfitted with off-the-shelf components consisting of two forward-facing video cameras, a 360-degree laser range finder, four radar sensors and advanced GPS units.
The Google project is very much an offshoot of the DARPA challenges. In fact, many of the members of the Google team are alumni of those competitions and brought the lessons learned during that time with them.
"Certainly there have been a lot of advances in sensor technology that are allowing us to do this today versus 15 years ago," says Chris Urmson, technical lead for the Google project and a lead on the Carnegie Mellon team that won the 2007 DARPA Urban Challenge. "But we see the real power as being in the software that's taking that data and figuring out what to do with it."
It is a vehicle's ability to interpret what is going on in its surroundings and then react appropriately that represents perhaps the biggest technical challenge for the future of self-driving cars.
An issue of common sense
An aspect of Google's project often lost on the casual observer is that its cars are not completely autonomous, even when no human is helping drive them. In order for the vehicles to function the route needs to be driven by a human ahead of time in one of the test cars and mapped using its array of sensors. This rich set of mapping data is then stored on a Google data center and a portion of it is loaded into the car's hard drive. The location of stoplights, school zones and anything else that is reasonably static is marked so the car will acknowledge them without having to interpret them in real-time.
But now, led in large part by Google's sudden and unexpected charge, autonomous robot cars come tantalizingly close to reality. As various mapping, sensing and location-based technologies have converged recently, Google has begun to position itself as the leader of our robo-chauffeur future. Yet for all of the technology's promise, it still has some major—and perhaps insurmountable—hurdles to overcome.
Google estimates that one million lives could be saved around the globe by driverless cars each year. According to the National Highway Traffic Safety Administration (NHTSA), in the U.S. alone there were 5.8 million crashes in 2008. Of those, about 34,000 resulted in fatalities, 1.6 million resulted in injuries and 4.2 million entailed some sort of property damage. The NHTSA says these numbers have come down over time—attesting at least partly to the ever-increasing safety of all vehicles—but they clearly still account for a large amount of deaths, injuries and property damage that driverless cars could drastically reduce.
Gigantic leaps with off-the-shelf components
Forty years ago the first self-driving cars were little more than crude, slow-moving contraptions following lines painted on the road. The past several years, however, have seen accelerated success in the quest for autonomous road vehicles, starting with a series of self-driving challenges the Defense Advanced Research Projects Agency (DARPA) held between 2004 and 2007 to help develop robots that could replace some U.S. military personnel on the battlefield.
Last year, a group of engineers from Stanford University Dynamic Design Laboratory, the Electronics Research Laboratory for the Volkswagen Group and software-maker Oracle Corp. shifted driverless cars into a higher gear by successfully running their 265-horsepower Autonomous Audi TTS Pikes Peak research car up the 20-kilometer course of the Pikes Peak International Hill Climb race in Colorado without stopping. The sporty hatchback—Volkswagen owns Audi—carried two computers in its trunk, both using Oracle's Java Real-Time System to run algorithms that kept the car on the road and at the limits of its handling ability on the varying surfaces and conditions.
Google's self-driving car project exited stealth mode last year and now the company is lobbying Nevada to be the first state to allow self-driving cars to be legally operated on public roads. To this point, Google's robot fleet has traveled more than 240,000 kilometers with minimal human intervention and only one incident in which a test car was rear-ended by another (human-driven) vehicle. Unlike the futuristic sci-fi images splashed on movie screens, the Google fleet consists of conventional vehicles—six Priuses and one Audi—that have been outfitted with off-the-shelf components consisting of two forward-facing video cameras, a 360-degree laser range finder, four radar sensors and advanced GPS units.
The Google project is very much an offshoot of the DARPA challenges. In fact, many of the members of the Google team are alumni of those competitions and brought the lessons learned during that time with them.
"Certainly there have been a lot of advances in sensor technology that are allowing us to do this today versus 15 years ago," says Chris Urmson, technical lead for the Google project and a lead on the Carnegie Mellon team that won the 2007 DARPA Urban Challenge. "But we see the real power as being in the software that's taking that data and figuring out what to do with it."
It is a vehicle's ability to interpret what is going on in its surroundings and then react appropriately that represents perhaps the biggest technical challenge for the future of self-driving cars.
An issue of common sense
An aspect of Google's project often lost on the casual observer is that its cars are not completely autonomous, even when no human is helping drive them. In order for the vehicles to function the route needs to be driven by a human ahead of time in one of the test cars and mapped using its array of sensors. This rich set of mapping data is then stored on a Google data center and a portion of it is loaded into the car's hard drive. The location of stoplights, school zones and anything else that is reasonably static is marked so the car will acknowledge them without having to interpret them in real-time.
"Computers are famously devoid of common sense, and you can think of this pre-mapping as a way to bootstrap some common sense into the car," Urmson says.
Even so, if the world changes between the time the map was assembled and the time the test vehicle drives the route, it can lead to confusion. "There are things that right now are a challenge for us," Urmson says. "For instance, if most of the world stayed the same but the lanes are shifted—so the physical road didn't move but, for whatever reason, the department of transportation decided we should drive a half lane to the left—that would probably confuse the car today."
There are two main components to Google's efforts: "The first is reliability, which means having the car do the things we expect it to do over and over again; and the second is robustness, which is dealing with unusual situations and still being safe," Urmson says.
A large part of the ability to increase the system's reliability and robustness depends on developing new sensors that can see farther and provide a denser data set, according to Urmson. But the ability for a self-driving robot to deal with unexpected or unusual situations that get thrown at it are what make some question the self-driving car's apparent inevitability.
John Leonard, a Massachusetts Institute of Technology mechanical and ocean engineering professor who led that university's team to a fourth-place finish in the 2007 DARPA Urban Challenge, thinks that major technological hurdles in robot perception need to be overcome before self-driving cars can be deployed on a large-scale.
"I have tremendous admiration for my colleagues at Google," Leonard says. "The performance that they have achieved is amazing—for example, their ability to drive at highway speeds. However, because they are building the maps in advance and then having humans pick out stop signs and street lights and crosswalks and so forth, it's very different than turning a robot loose autonomously in the world with very little prior information."
Dealing with the extremes
Much of Leonard's work has been focused on Simultaneous Localization and Mapping (SLAM). Unlike the pre-mapping that Google's system requires, SLAM would allow a vehicle to drive through the world at the same time that it is mapping it. This holy grail of autonomous driving would greatly increase the self-driving car's ability to deal with dynamically changing information—but even SLAM would not be able to solve all problems.
More than 15 years ago the No Hands Across America team drove from Washington, D.C., to San Diego in an autonomous vehicle and made it 98.2 percent of the way without human intervention, Leonard says. What about that last 1.8 percent? "A key challenge even today is dealing with those unexpected moments," he adds. "To try to get to that 100 percent level of performance there's a common-sense reasoning—one of those elusive goals of artificial intelligence—that no amount of pre-mapping is going to prepare you for."
Even Google admits that they have no good way for dealing with these unexpected moments yet, which is why every test vehicle has two backup humans on board to monitor and take over when the car reacts strangely. "Our program is very much a research program at this point, and we haven't really addressed that issue yet," Urmson says. "If we were to take the people out of the cars today, they'd drive pretty well, and you probably wouldn't notice them on the road until something unexpected happened or some element of the [unreliability] of the system appeared."
Even so, if the world changes between the time the map was assembled and the time the test vehicle drives the route, it can lead to confusion. "There are things that right now are a challenge for us," Urmson says. "For instance, if most of the world stayed the same but the lanes are shifted—so the physical road didn't move but, for whatever reason, the department of transportation decided we should drive a half lane to the left—that would probably confuse the car today."
There are two main components to Google's efforts: "The first is reliability, which means having the car do the things we expect it to do over and over again; and the second is robustness, which is dealing with unusual situations and still being safe," Urmson says.
A large part of the ability to increase the system's reliability and robustness depends on developing new sensors that can see farther and provide a denser data set, according to Urmson. But the ability for a self-driving robot to deal with unexpected or unusual situations that get thrown at it are what make some question the self-driving car's apparent inevitability.
John Leonard, a Massachusetts Institute of Technology mechanical and ocean engineering professor who led that university's team to a fourth-place finish in the 2007 DARPA Urban Challenge, thinks that major technological hurdles in robot perception need to be overcome before self-driving cars can be deployed on a large-scale.
"I have tremendous admiration for my colleagues at Google," Leonard says. "The performance that they have achieved is amazing—for example, their ability to drive at highway speeds. However, because they are building the maps in advance and then having humans pick out stop signs and street lights and crosswalks and so forth, it's very different than turning a robot loose autonomously in the world with very little prior information."
Dealing with the extremes
Much of Leonard's work has been focused on Simultaneous Localization and Mapping (SLAM). Unlike the pre-mapping that Google's system requires, SLAM would allow a vehicle to drive through the world at the same time that it is mapping it. This holy grail of autonomous driving would greatly increase the self-driving car's ability to deal with dynamically changing information—but even SLAM would not be able to solve all problems.
More than 15 years ago the No Hands Across America team drove from Washington, D.C., to San Diego in an autonomous vehicle and made it 98.2 percent of the way without human intervention, Leonard says. What about that last 1.8 percent? "A key challenge even today is dealing with those unexpected moments," he adds. "To try to get to that 100 percent level of performance there's a common-sense reasoning—one of those elusive goals of artificial intelligence—that no amount of pre-mapping is going to prepare you for."
Even Google admits that they have no good way for dealing with these unexpected moments yet, which is why every test vehicle has two backup humans on board to monitor and take over when the car reacts strangely. "Our program is very much a research program at this point, and we haven't really addressed that issue yet," Urmson says. "If we were to take the people out of the cars today, they'd drive pretty well, and you probably wouldn't notice them on the road until something unexpected happened or some element of the [unreliability] of the system appeared."
The way M.I.T.'s Leonard sees it, these elements of unreliability are what hinder a place for self-driving cars in our future. "Imagine a situation where a box falls on the road in front of you because it wasn't strapped down properly," he says. "The system needs to make a split-second decision to either go straight through it or to swerve left or right—which might have worse consequences than just going forward. The crux of the problem lies in those extreme situations at the tails of the curve that get harder and harder to deal with."
"Despite all the best efforts of the robot designers, humans still do stupid things," Leonard says. "Suppose 10 human-generated fatalities are replaced with five robot-generated fatalities, is that an ethical trade that society wants to make?"
Huge potential rewards
Even with possibly unsolvable issues, the team at Google sees the vast promise in the technology to save lives, reduce energy consumption and increase productivity as enough reason to continue pursuing it. Already, they have seen the system exhibit unexpected "intelligence" by combining bits of behavior that were fed into it and churning out surprisingly beneficial results.
Urmson recalled a moment on a recent test drive as the car was approaching a crosswalk and it came to a stop for no apparent reason. He began to tell his co-driver that it was likely a bug in the software, but at just that moment an elderly woman stepped into the road from where she was hidden behind a row of parked cars. "To me those are the most exciting moments because they show the promise of this system to be able to see more than a human can," he says.
Whereas evidence such as this certainly hints at the huge promise of robot-driven vehicles, Leonard sees a number of major technical challenges before truly autonomous cars become reality. "There are ways for self-driving car technology to have an impact and save lives without delivering completely autonomous vehicles," he says, adding that bits and pieces of the technology will likely deliver new and ingenious safety and environmental benefits. "But the question of how to deliver fully autonomous vehicles remains unanswered to me."
As the company with the most resources invested in the project and a reputation for successful technological innovation, Google's effort has certainly convinced many people that they will soon be able to nap on their morning commute while a robot car efficiently whisks them to work—and the Google team makes no attempt to dissuade this kind of wishful thinking, believing the technology will reach consumers' hands (or hands-free) relatively soon.
"With any kind of technology, making predictions for the future or how quickly things will be adopted is very difficult," Urmson says. "A decade out is a safe estimate, but I think it's really hard to say." @font-face { font-family: "Times New Roman"; }@font-face { font-family: "Verdana"; }@font-face { font-family: "Cambria"; }p.MsoNormal, li.MsoNormal, div.MsoNormal { margin: 0in 0in 0.0001pt; font-size: 12pt; font-family: Cambria; }table.MsoNormalTable { font-size: 10pt; font-family: "Times New Roman"; }span.msoIns { text-decoration: underline; color: green; }span.msoDel { text-decoration: line-through; color: red; }div.Section1 { page: Section1; }
"Despite all the best efforts of the robot designers, humans still do stupid things," Leonard says. "Suppose 10 human-generated fatalities are replaced with five robot-generated fatalities, is that an ethical trade that society wants to make?"
Huge potential rewards
Even with possibly unsolvable issues, the team at Google sees the vast promise in the technology to save lives, reduce energy consumption and increase productivity as enough reason to continue pursuing it. Already, they have seen the system exhibit unexpected "intelligence" by combining bits of behavior that were fed into it and churning out surprisingly beneficial results.
Urmson recalled a moment on a recent test drive as the car was approaching a crosswalk and it came to a stop for no apparent reason. He began to tell his co-driver that it was likely a bug in the software, but at just that moment an elderly woman stepped into the road from where she was hidden behind a row of parked cars. "To me those are the most exciting moments because they show the promise of this system to be able to see more than a human can," he says.
Whereas evidence such as this certainly hints at the huge promise of robot-driven vehicles, Leonard sees a number of major technical challenges before truly autonomous cars become reality. "There are ways for self-driving car technology to have an impact and save lives without delivering completely autonomous vehicles," he says, adding that bits and pieces of the technology will likely deliver new and ingenious safety and environmental benefits. "But the question of how to deliver fully autonomous vehicles remains unanswered to me."
As the company with the most resources invested in the project and a reputation for successful technological innovation, Google's effort has certainly convinced many people that they will soon be able to nap on their morning commute while a robot car efficiently whisks them to work—and the Google team makes no attempt to dissuade this kind of wishful thinking, believing the technology will reach consumers' hands (or hands-free) relatively soon.
"With any kind of technology, making predictions for the future or how quickly things will be adopted is very difficult," Urmson says. "A decade out is a safe estimate, but I think it's really hard to say." @font-face { font-family: "Times New Roman"; }@font-face { font-family: "Verdana"; }@font-face { font-family: "Cambria"; }p.MsoNormal, li.MsoNormal, div.MsoNormal { margin: 0in 0in 0.0001pt; font-size: 12pt; font-family: Cambria; }table.MsoNormalTable { font-size: 10pt; font-family: "Times New Roman"; }span.msoIns { text-decoration: underline; color: green; }span.msoDel { text-decoration: line-through; color: red; }div.Section1 { page: Section1; }
No comments:
Post a Comment