NARRATOR: California Speedway: Mecca for NASCAR auto racing. In October of 2005, 43 teams have brought their vehicles to compete in a unique event, a race, driven, not by testosterone, but computer code. These machines are robots, and they're going to run an obstacle course with no one behind the wheel and no remote control.
It's called the Grand Challenge and for good reason. On a two-mile course, the robots are on their own, and there's plenty to trip them up. They'll encounter gates and tunnels and all kinds of obstacles, including parked cars and hay bales.
The competition comes in all shapes and sizes. The smallest robot is Ghostrider.
ANNOUNCER: Ladies and gentlemen, the world's first autonomous motorcycle, Ghostrider, approaching Obstacle 1.
NARRATOR: As though raised from the dead, Ghostrider bravely forges on.
Now the largest entry, a 16-ton truck called Terramax, approaches the first gate and hesitates.
TERRAMAX TEAM MEMBER: Come on, baby. Crank it, crank it, crank it. Do it, do it, do it!
NARRATOR: Like anxious parents, its creators look on in agony, but all they can do is watch, as Terramax backs up and tries again.
TERRAMAX TEAM MEMBER: ...like a bull in a China shop. Come on!
NARRATOR: On the eighth pass, Terramax squeezes through the gate and heads for the finish.
Tommy has a lovable retro appearance, but looks are a bit deceiving. A computer glitch turns the robot into a wild beast. There are no serious injuries, but Tommy is out of the competition.
Team DAD has arrived from Silicon Valley with a roof ornament that looks like a relic from an old disco. It's a homemade laser scanner created by Dave Hall and his brother Bruce. Even as they drive to the start, they're writing computer code to make their scanner work. Now it's show time.
Almost immediately, they hit the hay. For this run, Team DAD, like disco in the '70s, is dead, but they'll have another chance.
Many universities are fielding teams, and Stanley, Stanford's robot, looks like a high tech version of another V.W., Herbie, "The Love Bug." Stanley is the brain child of Sebastian Thrun. For him, the Grand Challenge is a Woodstock for robo-geeks.
SEBASTIAN THRUN: There's very little in the way of competitiveness here. Although everyone understands, in the end of the day, everyone wants to win, it's been a big feast. It's, like, a thousand people who care about the same thing—making cars drive themselves—and they all get together. It's a big festival. It's just really great.
NARRATOR: The largest team at the competition is from Carnegie Mellon in Pittsburgh, and for them, the Grand Challenge is no love fest.
WILLIAM "RED" WHITTAKER (Carnegie Mellon University): You are either thoroughly rested, 200 percent effectiveness, or you got some work to do.
WILLIAM "RED" WHITTAKER : No one knows what it takes to succeed in the Grand Challenge, since no one, anywhere, has done it. The Red Team's prepared for anything.
NARRATOR: Led by former marine, Red Whittaker, the Red Team is a well-funded army of computer wizards and engineers. The team has invested two years and millions of dollars to get Sandstorm to the starting line.
Sandstorm is aggressive on the straightaway, but today, the robot has a fatal attraction to hay bales.
Like all the entries, Sandstorm has four chances to demonstrate agility and speed for the judges. Based on their performances, 20 robots will qualify for the main event, a long distance desert race and a chance to win $2,000,000.
The race is sponsored by DARPA, the Defense Advanced Research Projects Agency. Their goal is to stimulate technology that can be used by the military.
RON KURJANOWICZ: If we can use unmanned vehicles in dangerous missions, like delivering supplies to front lines, then we can take drivers out of the vehicles, and they're less vulnerable. If you remember Private Jessica Lynch, she was a driver of a supply vehicle, and if that was an autonomous vehicle, that episode would have never happened. True autonomy, that's what we're after. It's a hard problem.
NARRATOR: Just how hard a problem it is to create a vehicle that can drive itself became painfully clear a year earlier, at the first Grand Challenge race.
Carnegie Mellon was one of 15 teams at the starting line. Racing against the clock, Sandstorm was the first robot out of the gate. To win, it had to run a course across 142 miles of rugged desert terrain in less than 10 hours. But just seven miles from the start, Sandstorm drifted off course and got stuck on a rock. In a valiant effort to keep moving, the robot spun its wheels until its tires burst into flame.
Back at the start, other robots were having their own problems.
WAYT GIBBS (Senior Writer, Scientific American): Everything was set up for this high-speed rumble of all these vehicles, just careening through the desert, and that's not how it happened.
NARRATOR: All 15 robots died of mechanical failure, half within the first mile. In this group of losers, Sandstorm went the farthest, leading many to believe no robot could ever win the Grand Challenge.
RON KURJANOWICZ: The fact that the vehicles only traveled a short distance shows that it was a tough technical problem, but where the success really lies is in the amount of people that actually responded to the challenge.
NARRATOR: After this fiasco, Carnegie Mellon now has a year to prepare for the next race.
RED WHITTAKER: The five main things that we said we would get done today aren't done, so it's not enough to wave hands and say, "Voila, this works." I signed up to succeed in the Grand Challenge. This time around, the Red Team will be more like a Red Army. It is a machine to be reckoned with.
NARRATOR: Red Whittaker's resume is a veritable history of field robotics. His robots are battle-hardened machines built to clean up Three Mile Island and Chernobyl. These rustbelt brutes can map dangerous mines, load trucks, remove asbestos and mow hay. They're global explorers that have ventured across volcanoes, deserts and glaciers.
Now Whittaker has staked his reputation on winning the Grand Challenge. To double his chances, he's planning to run two robots in the next race. In addition to their veteran, Sandstorm, the Red Team will transform this new Hummer into a more aggressive machine called Highlander.
For more than a year, Wayt Gibbs has been documenting Whittaker's Grand Challenge quest.
WAYT GIBBS: Red Whittaker is nothing if not an ambitious man. He's a roboticist who, on his weekends, goes out and drives a tractor for 10 hours on his farm and who likes to actually get in and get his hands greasy and dirty inside an engine. Red gets excited about seeing his robots do powerful things in the real world.
NICK MILLER: I think it's going to work.
RED WHITTAKER: The thing...at this stage in the game, it's making it work, right? Making it work.
NARRATOR: The Red Team will spend several months replacing and modifying Highlander's hardware.
RED WHITTAKER: A tremendous amount of these desert racing machines is heavy metal, iron mongering. It's heating and beating and bashing and cutting and making. It's as important as any of the software. And it's a lot of fun.
NARRATOR: To replace the human behind the wheel, team members have installed an electro-mechanical system to steer, shift, throttle and brake. They're installing sensors, so Highlander can see, and GPS receivers, so it can follow the race course. And they're giving the robot a brain and filling it with software so it can drive by itself. Now, it might seem easy for robots to drive.
It happens all the time in the movies.
WAYT GIBBS: If you think of the Terminator movies...science fiction has given us a really distorted view of the capabilities of robots. Those robots are able to identify specific individuals, but, in reality, robots have very limited senses of perception.
NARRATOR: Robots know nothing about their environment, and they can't think like humans, but they're really good at crunching numbers. Most rely on laser range finders to perceive the world.
Inside, a beam of light bounces off a spinning mirror and sweeps the terrain ahead. By measuring the time it takes a beam to hit something and return, the range finder can measure its distance from the vehicle. Scanners are mounted in pairs to measure different things. Some look for smooth road ahead, others look for the road edge on each side. Some look for vertical objects, a rock or a fence post that can kill a robot.
Putting it all together, this becomes Highlander's view of the world, a roadmap, made from millions of numbers, that basically tells the robot "bumpy" or "smooth."
Off on its own, Highlander will try to avoid obstacles for the first time.
Improving its driving skills will take endless hours of programming.
WAYT GIBBS: The Red Team robots have on the order of a million lines of computer code. Even so, they do very simple reasoning about the world. They look for flat spots: "Flat is good. Drive on flat spots. Anything that's not flat, avoid it."
NARRATOR: But for the Red Team, winning the Grand Challenge involves much more than writing computer code. While some team members are testing Highlander in the field, others are refurbishing Sandstorm. Another group is building state of the art sensors, and writing software never ends.
Whittaker commands an army of 100 or more. With a $3,000,000 war chest, he's determined to have two robots at the starting line and one in the winner's circle.
RED WHITTAKER: A lot of people work hard. The Red Team is no exception, on that front. Winning a race isn't about working hard. Yeah, you've got to do that, but it's a lot more. If you haven't done everything, you haven't done a thing.
NARRATOR: Not all teams take Whittaker's all or nothing approach. For the Hall brothers the Grand Challenge is great fun. Each day Bruce and Dave program their robot as they drive in to work.
BRUCE HALL (Team Digital Auto Drive/DAD): We have about a 45 minute drive each day, and we're tired of driving it. We'd like to get into our vehicle and hit the switch and have it drive us home, frankly.
NARRATOR: The Hall brothers own a company that makes audio speakers. Bruce runs the daily business while Dave tinkers in the shop.
BRUCE HALL: Dave was the engineer and chief designer behind a new robot drive system, and that enabled us to compete in robot competitions such BattleBots.
BATTLEBOTS HOST: These guys are from Silicon Valley, and they mean business. Look at this bad boy.
BRUCE HALL: ...three hundred fifty pounds. We got second place in the world championship in Robot Wars. If you think about fighting robots where you're actually controlling them through a remote control device, the next logical step is to have a vehicle that drives itself.
NARRATOR: Instead of lasers, the Hall brothers experimented with video cameras. Inspired by the way humans see, the technique is called stereovision. On the roof, they mounted two cameras side by side. By measuring the lateral shift of specific areas in each image, the robot computes the distance of objects and creates a "difference map" to avoid hitting them.
BRUCE HALL: Nice! Ooh, sweet!
NARRATOR: Even after decades of research, stereovision is an unproven technology and at higher speeds, notoriously unreliable.
BRUCE HALL: Sorry.
NARRATOR: After a few scares, the Hall brothers scrapped stereovision and started building their spinning laser.
In the fields of Iowa, Team Terramax is a group of defense contractors. Like all competitors, they're not allowed to use government money to compete in the Grand Challenge. Still, for them, this is no hobby. Their vehicle is a big yellow goose that might lay a giant golden egg.
GARY SCHMIEDEL (Oshkosh Trucks): Terramax is a six-wheel drive vehicle. It is really the logistical backbone of the Marine Corps, and we think it's the most mobile vehicles in the world.
NARRATOR: Oshkosh has sold more than 6,000 of these trucks to the military. With the prowess of an Olympic athlete, it plunges into water, charges up hills and even takes to the air. So it's got the muscle, but what about the brains?
Without a driver, Terramax is more of a timid giant than a raging bull, but the team is confident of its future. In 10 years, the Pentagon plans to convert one third of all military vehicles to unmanned operation, and Team Terramax is very eager to help out.
Making autonomous vehicles is a tough problem, but apparently not tough enough for Anthony Levandowski, who's turning the Grand Challenge into a two-wheeled balancing act.
ANTHONY LEVANDOWSKI: A bike can do things that a truck and a car really can't do. You can do really sharp turns. You can go through a forest, really tight spaces. The agility is just there that's not on other vehicles.
NARRATOR: Anthony dropped out of grad school at U.C. Berkeley and spent two years teaching Ghostrider to stay up by itself.
ANTHONY LEVANDOWSKI: We're making a motorcycle that drives itself. Can you get any cooler than that? That's why everybody on the team puts in so many hours, because when you see it work, you say, "Wow, I made that."
NARRATOR: A year earlier, at the first Grand Challenge, Ghostrider made its world debut.
ANTHONY LEVANDOWSKI: I haven't slept in two days, and I'm super excited. I mean, this is the culmination of six months of work. And I forget to turn on the stability. And then the second I saw it start to fall and not steer, I said, "Oh, my god! I forgot the switch." That's life you know, bad things happen. But this will never, ever happen to me, ever again.
The Grand Challenge is basically...takes up all my life. Like, I don't really talk to my friends anymore; I don't have a girlfriend, right now. I think it's totally worth it, though. This could be the chance of a lifetime.
NARRATOR: Stanford University is a strong competitor, entering the Grand Challenge for the first time.
SEBASTIAN THRUN: I think the teams did great the first time around and performed very well. But it was obvious there was some missing link, some missing science to actually win this race. And these are the challenges I live for. That's why I do what I do.
NARRATOR: For almost eight years, Sebastian Thrun was Red Whittaker's colleague at Carnegie Mellon. They share a passion for robots. But Thrun has little interest in hardware and convinced Volkswagen to donate a vehicle with steering, brake and shift actuators that can be controlled by computer.
SEBASTIAN THRUN: The vehicle's name is Stanley. So Stanley is nothing else but Stanford, but it also gives the vehicle a personality. We think of the vehicle more and more as an intelligent decision maker.
NARRATOR: Thrun is a computer scientist with a deep interest in machine learning, and robots with personality are his specialty. His robots assist the elderly, conduct museum tours, fetch tennis balls, and each one embodies Thrun's desire to create machines that think.
WAYT GIBBS: Sebastian Thrun is interested in the science of robotics, in learning about new techniques that can make robots do fantastically advanced things 10, 20, even 50 years from now.
SEBASTIAN THRUN: I guess we've mounted them vertically, but we have no data, so far, right? And we have no software that can do it. Okay, so how foolish of us.
NARRATOR: Compared to the Red Team, Stanford is a small group with a single focus on software.
SEBASTIAN THRUN: There's a good number of teams who think the challenge lies in the hardware, and god bless them. It's great they're doing this, and I know there's hardware challenges. We, from the very beginning, pursued a very aggressive timeline to focus almost entirely on software.
NARRATOR: Stanford uses the same laser hardware as the Red Team, but they're experimenting with ways to use software to give their robot a greater role in making decisions about speed and safety.
SEBASTIAN THRUN: So, in this display, the white area is safe to drive, the red area is unsafe to drive, and the gray are, is...you don't know. And the robot, then, just has to steer itself so it stays in the middle of the road and avoids this red stuff, as much as possible.
NARRATOR: To test Stanley's software on the original race course, the team has traveled to the Mojave Desert, where the first Grand Challenge began. Most of Stanley's computer code has been written by Mike Montemerlo, Red Whittaker's former Ph.D. student.
MIKE MONTEMERLO: Give me steering and throttle.
SEBASTIAN THRUN: Steering is yours; total; you're going.
MIKE MONTEMERLO: Great. That's up to 15 now.
SEBASTIAN THRUN: What's the mile marker?
MIKE MONTEMERLO: Seven point two eight.
SEBASTIAN THRUN: Seven point three is the critical place. In fact, right after this turn over here, the Red team rolled off the road. And there's this final destination in the race, it's a very exciting place for us, it's a historical place. This is the end of the previous Grand Challenge.
MIKE MONTEMERLO: Swerve to the right.
SEBASTIAN THRUN: Sorry, we're having a problem.
MIKE MONTEMERLO: It's the end of the race for us.
SEBASTIAN THRUN: Yes. What happened, Mike?
MIKE MONTEMERLO: It just...the long lasers can't see us. They're blind in the corner.
SEBASTIAN THRUN: We're stuck because of a software bug, which we're currently debugging. So our challenge, between April...now and October, will be to get rid of these bugs.
NARRATOR: Unbelievably, Stanley faltered in the same place that killed the Red Team in the first Grand Challenge. But Thrun is confident about going head to head against a team he knows well.
SEBASTIAN THRUN: The Red Team is a seasoned competitor, and they have a lot of money, a lot of equipment. They have a fantastic team. I used to be at Carnegie Mellon. I know Red very well. He's an amazing person.
I wouldn't be surprised if the Red Team and the Stanford Team and maybe two or three other teams were the big teams in the end. We'll see. It's a race. It's a true race.
NARRATOR: The Red Team is teaching its robot to go fast, up to 50 miles an hour. Software guru Kevin Peterson has been spending long days and nights improving Sandstorm's performance at race pace.
KEVIN PETERSON: We've been working on Sandstorm for the last month. I just cut sleep out, you know? It's not that tough. You get three hours of sleep a day, drink some coffee, you're okay.
WAYT GIBBS: As a robot drives, it has to constantly update its picture of the terrain ahead. The faster it goes, the faster it has to update this picture or the terrain just gets on top of it. This is easy for humans to do because we can see out 300, 400, 500 feet ahead. But robots have sensors, like lasers, that operate at fixed ranges, 60 feet, at most 150 feet. It's like driving through a fog.
NARRATOR: Each team must lift this fog to win the race. The Red Team's secret weapon is hidden inside the dome on Sandstorm's roof, where a high-powered laser is mounted on a device called a gimbal.
RED WHITTAKER: The gimbal is a motion platform that's a little bit like an animal's neck. It can point sensors left, right, up, down, in such a way that the pounding motion from the vehicle is smoothed.
NARRATOR: Like binoculars, this special laser can see into the far distance. By damping vibration, the gimbal gives it a steady gaze and can point it to follow turns in the road.
The gimbal-mounted laser is unique to the Red Team, which has made several from scratch. But it's a fancy piece of hardware that's hard to build and easy to break.
SEBASTIAN THRUN: There's been a lot of discussion whether you want to gimbal your sensors. In fact, we've actually decided against it. We decided against it, because we wanted simplicity in the system. Every physical system that moves can break. And there's a lot of stuff you can do in software.
NARRATOR: The Stanford Team is putting its hopes in a set of off-the-shelf lasers linked through software to a simple video camera. The camera can see all the way to the horizon.
The trick is to give Stanley some way to distinguish the road in the video image. Stanley's lasers pinpoint smooth terrain right in front of the vehicle. By overlapping the laser and video data, Stanley can identify the color of the smooth area and search for it in the rest of the video image. The area Stanley believes is safe is now shown in red. By matching laser and video data on the fly, Stanley can adapt to changing road conditions and determine safe driving speeds all by itself.
SEBASTIAN THRUN: Making a car really intelligent requires to give it an ability to adapt. It's just like in human life. We don't enable kids to have a rule for every contingency. We tell them how to learn from contingencies and make them smarter, and then we trust them to learn.
NARRATOR: Stanford is the only team using adaptive vision to achieve race speeds. Whether it will work well enough for the Grand Challenge will soon become clear.
Back at the California Speedway, after more than a year of fevered preparations, the Grand Challenge is on. Officials from DARPA, the Defense Advanced Research Projects Agency, are judging the final runs of the qualifying event. Forty-three teams are competing for just 20 slots in the second, and perhaps final, Grand Challenge desert race.
After failing three times to complete the two mile obstacle course, Ghostrider is making its final attempt. It shows surprising agility, until it enters the tunnel. Looking like a wounded animal, it falls down and cannot get up. For Anthony Levandowski, the quest for the Grand Challenge is over.
ANTHONY LEVANDOWSKI: That was our chance, and it was unfortunate, but that's life.
NARRATOR: For the Hall brothers, it's also the moment of truth.
BRUCE HALL: This is looking good, this is looking solid. Barring any hardware issue, I think we're going to make it.
NARRATOR: At opposite ends of the track, the brothers Hall hold their breath as their new laser sensor guides the truck along the course.
DAVE HALL: All right, let's see if this obstacle detection works.
BRUCE HALL: Here's the test. Did all the years of work pay off? Turn, you. Thank you.
DAVE HALL: ...right around. It's almost as though it was programmed to do that.
BRUCE HALL: Dave, I hate to tell you, you're now in the race. You're in the race.
NARRATOR: By finishing the course, the Hall brothers have now earned a place in the main event.
BRUCE HALL: Hoo yeah, baby. You've got to love it.
RACE ANNOUNCER: Ladies and gentlemen, Terramax.
NARRATOR: Team Terramax also qualifies for the race.
Sandstorm's first run was not a dainty performance, but after the Red Team recalibrates its lasers, the robot loses its appetite for hay.
Sandstorm's younger sibling almost failed to make it to the start. Ten days earlier, on a practice run, Highlander swerved sharply and rolled on its roof. For the Red Team's prized robot, it looked like the end of the road.
Miraculously, Highlander is back in action, though hardly tested. It has a shaky start, but over the course of three runs, Highlander clocks the fastest times of any robot at the qualifier.
JOSH (RED TEAM): That thing's hauling.
KEVIN PETERSON: Frigging sweet, man. That's more like it.
NARRATOR: A first time competitor, Stanley is an unknown quantity. No one knows what to expect. The robot passes cleanly through the first gate. It shows impressive speed on the backstretch but is not as fast as Highlander.
Like a proud parent, Sebastian records the moment for posterity.
SEBASTIAN THRUN: Yes. So far, so good.
NARRATOR: Aided by adaptive vision, Stanley is the only vehicle that never hits an obstacle. In 36 hours, the race, the real Grand Challenge, will begin. DARPA announces who qualified and what the starting order will be.
TONY TETHER (D.A.R.P.A.): Highlander is Number 1 on Saturday morning; Stanford, Number 2 on the pole position; and Number 3, Sandstorm.
NARRATOR: The race course is a circular maze that zigzags for 132 miles across the Mojave Desert near the outskirts of Las Vegas. It's mostly flat, but rough and unpaved. It includes tunnels, cattle gates and steep mountain switchbacks. With a 10-hour time limit, the race will demand speed, agility and endurance that no machine has ever demonstrated on its own.
The exact route is a deeply guarded secret, but now, at 4 a.m., on race day, it's released to the teams. On a compact disk is a roadmap defined by 2,900 G.P.S. waypoints.
BRUCE HALL: All your hopes and dreams are on this disk.
NARRATOR: Teams have two hours to analyze the map and plug it into their robots before the race begins. In several minutes, the Stanford Team loads the course into their robot. Stanley will choose its own speeds on the fly, adapting to different types of terrain.
The Red Team takes a far different approach. Concealed in their trailer, Whittaker's troops look like they're planning an attack. The race course is layered on top of elevation data and satellite images stored in the Red Team's computers. The route is broken into sections and distributed to team members who will examine it in minute detail. Thousands of waypoints are added to fill gaps in DARPA's map, and each point can be adjusted to optimize the robot's path. Speeds are set for every inch of terrain. Red believes his robots must have as much information as possible to gain the winning edge.
RED WHITTAKER: Our preference is to err in the direction of overkill. I wouldn't want to reflect, Sunday morning, that we low-balled it.
NARRATOR: On a course that appears easy, Red makes a critical decision. He decides to run Highlander at a fast and risky pace. Sandstorm will be the tortoise, running 40 minutes slower, in the hope that at least one Red Team robot will finish.
RACE ANNOUNCER: The first time it's ever been done...
NARRATOR: As dawn breaks over the desert, robots prepare to boldly go where no robot has ever gone before. Vehicles will be launched at five minute intervals and race against the clock.
SEBASTIAN THRUN: Hey, Red. Red. All the best!
NARRATOR: Highlander leads the pack, followed by Stanley and Sandstorm. The cheering teams are now spectators. Their work is done. After months of tireless effort, there's a lot at stake. A vision they all share will now be put to the test.
During the first eight miles of the race, Highlander gains two minutes on Stanley. At times the Stanford robot seems confused and disoriented. Sandstorm is keeping pace with Stanley, five minutes behind.
It will take two hours to launch all the robots. Each one leaves the chute with confidence, a far cry from the first Grand Challenge when many faltered within sight of the start, and no robot went beyond seven miles.
KEVIN PETERSON: The start of the race was pretty cool. It was really exciting to see all the robots go. I was excited and apprehensive at the same time. You just don't know what's going to happen with the robot. It was entirely out of our hands.
NARRATOR: Far from the crowd, Highlander now leads a fleet of driverless machines racing the clock across the desert. Team DAD has passed two robots and is now heading to the front of the pack.
Eighteen robots have now passed the seven-mile mark that ended the last race. Miraculously, the Gray Team, from New Orleans, is holding its own, only weeks after the city was destroyed by Hurricane Katrina.
Terramax was the last to start and brings up the rear. Following each robot, DARPA chase vehicles relay information back to the start.
Here, teams gather around DARPA's displays.
NICK MILLER: It's just pretty sweet. We're pulling away in the front.
KEVIN PETERSON: We're pulling away?
NICK MILLER: For now.
KEVIN PETERSON: We're pulling away from DAD?
NICK MILLER: No, from Stanford.
NARRATOR: Thirty miles into the race, Highlander is making time on Stanley. Behind Stanley, Sandstorm is closing the gap and may pass the Stanford robot. Behind them, robots are starting to drop like flies.
At mile 26, Team DAD's laser comes unbolted from the roof. Officials in the chase car hit the kill switch, and for the Hall brothers the race is over.
At a media stop, a group of journalists sees Highlander charging toward them in the distance.
WAYT GIBBS: Highlander seemed to come down the hill very smartly, but then, as it started to mount the next hill, it looked like it ran out of gas or the engine cut off or somebody in there threw it into neutral. It just stopped moving and slid back down to the bottom of the hill. Something was wrong with Highlander, and it did not have its normal get up and go.
NARRATOR: Highlander keeps on rolling, but cannot maintain its intended pace. Red and the team slowly realize that things are not going according to plan.
KEVIN PETERSON: I don't know what's going on with Highlander. It's weird.
NICK MILLER: It's running slow.
KEVIN PETERSON: It's running real slow.
NARRATOR: Highlander struggles to summon the power to complete its mission. Behind it, Stanley is rapidly approaching. But Highlander recovers on the flats, and for many miles they continue neck and neck. One hundred and two miles into the race, Stanley's video camera records a milestone event.
RACE ANNOUNCER: Stanley has passed Highlander.
KEVIN PETERSON: At that point, we pretty much knew Highlander wasn't going to win the race. We thought Sandstorm might have a chance. It was really disappointing.
NARRATOR: Five hours after leaving the starting line, Stanley now leads the pack, and just five robots remain on the course. To finish, they must wind through a treacherous mountain pass, and Stanley must now make life or death decisions.
Somehow, Highlander has managed to reach the summit of the pass, but the gimbal is turned sideways and appears to be dead. Without its secret weapon, it must rely on short range lasers and G.P.S. to avoid disaster.
Right behind Highlander, Sandstorm is in good shape, but driving slowly, according to plan.
The crowd gathers. A blue dot appears in the distance.
STANFORD TEAM MEMBERS: Stanley!
Stanley!
NARRATOR: After driving six hours and 53 minutes, at an average speed of 19 miles an hour, Stanley is about to become the first vehicle in history to drive 132 miles by itself.
Twenty one minutes behind Stanley, Highlander limps to the finish.
Minutes later, the trusty veteran appears.
SEBASTIAN THRUN: It was just amazing to see this community of people. That community succeeded today. Behind me, there are three robots that made it all the way through the desert, and all three of them did the unthinkable. It's such a fantastic success for this community, I think we all win.
NARRATOR: Before announcing the winner, DARPA officials honor the five finishing teams. Terramax is fifth. A dark horse, the Gray Team from New Orleans, is fourth. Highlander is third. Sandstorm, Whittaker's tortoise, was just 11 minutes behind the winner.
TONY TETHER: I guess it's time. And the check goes to Stanley!
WAYT GIBBS: What Stanford did, in a short span of time, is amazing in the robotics world. On this scale of time, to develop a system this sophisticated, that can navigate so well over such a wide variety of terrain, at such high speeds, it really is worth its place in the history books.
NARRATOR: The source of Highlander's problems remains a mystery, but may have been the result of its rollover.
RED WHITTAKER: The engine punked on us. The only thing that was short was not having a top-end speed on the gas pedal. Except for that, we're in like a charm.
WAYT GIBBS: This really was like a horse race; they finished in such close times. So the Red Team is justifiably proud for having both of its robots finish the course.
NARRATOR: Many believe the completion of the Grand Challenge is a giant step that will push robots out of the lab, into the real world. The forces that will make this happen are embodied in the passion and vision of two men.
WAYT GIBBS: The world needs both Sebastian Thruns and Red Whittakers. If robotics is ever going to become part of the daily fabric of our lives, it needs those scientists who can come up with those far out ideas for how machines are actually going to start to think, to reason. And it also needs the builders, the implementers, and those who are driven to make things work in the real world.
NARRATOR: There is every reason to believe all those who participated will continue their quest to innovate, even though the Grand Challenge itself is now history.
So what about the future? Will mechanical warriors take over the battlefield? Will we all be passengers on the highways of tomorrow?
Someday, the great robot race may well be remembered as a moment when the partnership between humans and the machines we create was forever changed.
It's called the Grand Challenge and for good reason. On a two-mile course, the robots are on their own, and there's plenty to trip them up. They'll encounter gates and tunnels and all kinds of obstacles, including parked cars and hay bales.
The competition comes in all shapes and sizes. The smallest robot is Ghostrider.
ANNOUNCER: Ladies and gentlemen, the world's first autonomous motorcycle, Ghostrider, approaching Obstacle 1.
NARRATOR: As though raised from the dead, Ghostrider bravely forges on.
Now the largest entry, a 16-ton truck called Terramax, approaches the first gate and hesitates.
TERRAMAX TEAM MEMBER: Come on, baby. Crank it, crank it, crank it. Do it, do it, do it!
NARRATOR: Like anxious parents, its creators look on in agony, but all they can do is watch, as Terramax backs up and tries again.
TERRAMAX TEAM MEMBER: ...like a bull in a China shop. Come on!
NARRATOR: On the eighth pass, Terramax squeezes through the gate and heads for the finish.
Tommy has a lovable retro appearance, but looks are a bit deceiving. A computer glitch turns the robot into a wild beast. There are no serious injuries, but Tommy is out of the competition.
Team DAD has arrived from Silicon Valley with a roof ornament that looks like a relic from an old disco. It's a homemade laser scanner created by Dave Hall and his brother Bruce. Even as they drive to the start, they're writing computer code to make their scanner work. Now it's show time.
Almost immediately, they hit the hay. For this run, Team DAD, like disco in the '70s, is dead, but they'll have another chance.
Many universities are fielding teams, and Stanley, Stanford's robot, looks like a high tech version of another V.W., Herbie, "The Love Bug." Stanley is the brain child of Sebastian Thrun. For him, the Grand Challenge is a Woodstock for robo-geeks.
SEBASTIAN THRUN: There's very little in the way of competitiveness here. Although everyone understands, in the end of the day, everyone wants to win, it's been a big feast. It's, like, a thousand people who care about the same thing—making cars drive themselves—and they all get together. It's a big festival. It's just really great.
NARRATOR: The largest team at the competition is from Carnegie Mellon in Pittsburgh, and for them, the Grand Challenge is no love fest.
WILLIAM "RED" WHITTAKER (Carnegie Mellon University): You are either thoroughly rested, 200 percent effectiveness, or you got some work to do.
WILLIAM "RED" WHITTAKER : No one knows what it takes to succeed in the Grand Challenge, since no one, anywhere, has done it. The Red Team's prepared for anything.
NARRATOR: Led by former marine, Red Whittaker, the Red Team is a well-funded army of computer wizards and engineers. The team has invested two years and millions of dollars to get Sandstorm to the starting line.
Sandstorm is aggressive on the straightaway, but today, the robot has a fatal attraction to hay bales.
Like all the entries, Sandstorm has four chances to demonstrate agility and speed for the judges. Based on their performances, 20 robots will qualify for the main event, a long distance desert race and a chance to win $2,000,000.
The race is sponsored by DARPA, the Defense Advanced Research Projects Agency. Their goal is to stimulate technology that can be used by the military.
RON KURJANOWICZ: If we can use unmanned vehicles in dangerous missions, like delivering supplies to front lines, then we can take drivers out of the vehicles, and they're less vulnerable. If you remember Private Jessica Lynch, she was a driver of a supply vehicle, and if that was an autonomous vehicle, that episode would have never happened. True autonomy, that's what we're after. It's a hard problem.
NARRATOR: Just how hard a problem it is to create a vehicle that can drive itself became painfully clear a year earlier, at the first Grand Challenge race.
Carnegie Mellon was one of 15 teams at the starting line. Racing against the clock, Sandstorm was the first robot out of the gate. To win, it had to run a course across 142 miles of rugged desert terrain in less than 10 hours. But just seven miles from the start, Sandstorm drifted off course and got stuck on a rock. In a valiant effort to keep moving, the robot spun its wheels until its tires burst into flame.
Back at the start, other robots were having their own problems.
WAYT GIBBS (Senior Writer, Scientific American): Everything was set up for this high-speed rumble of all these vehicles, just careening through the desert, and that's not how it happened.
NARRATOR: All 15 robots died of mechanical failure, half within the first mile. In this group of losers, Sandstorm went the farthest, leading many to believe no robot could ever win the Grand Challenge.
RON KURJANOWICZ: The fact that the vehicles only traveled a short distance shows that it was a tough technical problem, but where the success really lies is in the amount of people that actually responded to the challenge.
NARRATOR: After this fiasco, Carnegie Mellon now has a year to prepare for the next race.
RED WHITTAKER: The five main things that we said we would get done today aren't done, so it's not enough to wave hands and say, "Voila, this works." I signed up to succeed in the Grand Challenge. This time around, the Red Team will be more like a Red Army. It is a machine to be reckoned with.
NARRATOR: Red Whittaker's resume is a veritable history of field robotics. His robots are battle-hardened machines built to clean up Three Mile Island and Chernobyl. These rustbelt brutes can map dangerous mines, load trucks, remove asbestos and mow hay. They're global explorers that have ventured across volcanoes, deserts and glaciers.
Now Whittaker has staked his reputation on winning the Grand Challenge. To double his chances, he's planning to run two robots in the next race. In addition to their veteran, Sandstorm, the Red Team will transform this new Hummer into a more aggressive machine called Highlander.
For more than a year, Wayt Gibbs has been documenting Whittaker's Grand Challenge quest.
WAYT GIBBS: Red Whittaker is nothing if not an ambitious man. He's a roboticist who, on his weekends, goes out and drives a tractor for 10 hours on his farm and who likes to actually get in and get his hands greasy and dirty inside an engine. Red gets excited about seeing his robots do powerful things in the real world.
NICK MILLER: I think it's going to work.
RED WHITTAKER: The thing...at this stage in the game, it's making it work, right? Making it work.
NARRATOR: The Red Team will spend several months replacing and modifying Highlander's hardware.
RED WHITTAKER: A tremendous amount of these desert racing machines is heavy metal, iron mongering. It's heating and beating and bashing and cutting and making. It's as important as any of the software. And it's a lot of fun.
NARRATOR: To replace the human behind the wheel, team members have installed an electro-mechanical system to steer, shift, throttle and brake. They're installing sensors, so Highlander can see, and GPS receivers, so it can follow the race course. And they're giving the robot a brain and filling it with software so it can drive by itself. Now, it might seem easy for robots to drive.
It happens all the time in the movies.
WAYT GIBBS: If you think of the Terminator movies...science fiction has given us a really distorted view of the capabilities of robots. Those robots are able to identify specific individuals, but, in reality, robots have very limited senses of perception.
NARRATOR: Robots know nothing about their environment, and they can't think like humans, but they're really good at crunching numbers. Most rely on laser range finders to perceive the world.
Inside, a beam of light bounces off a spinning mirror and sweeps the terrain ahead. By measuring the time it takes a beam to hit something and return, the range finder can measure its distance from the vehicle. Scanners are mounted in pairs to measure different things. Some look for smooth road ahead, others look for the road edge on each side. Some look for vertical objects, a rock or a fence post that can kill a robot.
Putting it all together, this becomes Highlander's view of the world, a roadmap, made from millions of numbers, that basically tells the robot "bumpy" or "smooth."
Off on its own, Highlander will try to avoid obstacles for the first time.
Improving its driving skills will take endless hours of programming.
WAYT GIBBS: The Red Team robots have on the order of a million lines of computer code. Even so, they do very simple reasoning about the world. They look for flat spots: "Flat is good. Drive on flat spots. Anything that's not flat, avoid it."
NARRATOR: But for the Red Team, winning the Grand Challenge involves much more than writing computer code. While some team members are testing Highlander in the field, others are refurbishing Sandstorm. Another group is building state of the art sensors, and writing software never ends.
Whittaker commands an army of 100 or more. With a $3,000,000 war chest, he's determined to have two robots at the starting line and one in the winner's circle.
RED WHITTAKER: A lot of people work hard. The Red Team is no exception, on that front. Winning a race isn't about working hard. Yeah, you've got to do that, but it's a lot more. If you haven't done everything, you haven't done a thing.
NARRATOR: Not all teams take Whittaker's all or nothing approach. For the Hall brothers the Grand Challenge is great fun. Each day Bruce and Dave program their robot as they drive in to work.
BRUCE HALL (Team Digital Auto Drive/DAD): We have about a 45 minute drive each day, and we're tired of driving it. We'd like to get into our vehicle and hit the switch and have it drive us home, frankly.
NARRATOR: The Hall brothers own a company that makes audio speakers. Bruce runs the daily business while Dave tinkers in the shop.
BRUCE HALL: Dave was the engineer and chief designer behind a new robot drive system, and that enabled us to compete in robot competitions such BattleBots.
BATTLEBOTS HOST: These guys are from Silicon Valley, and they mean business. Look at this bad boy.
BRUCE HALL: ...three hundred fifty pounds. We got second place in the world championship in Robot Wars. If you think about fighting robots where you're actually controlling them through a remote control device, the next logical step is to have a vehicle that drives itself.
NARRATOR: Instead of lasers, the Hall brothers experimented with video cameras. Inspired by the way humans see, the technique is called stereovision. On the roof, they mounted two cameras side by side. By measuring the lateral shift of specific areas in each image, the robot computes the distance of objects and creates a "difference map" to avoid hitting them.
BRUCE HALL: Nice! Ooh, sweet!
NARRATOR: Even after decades of research, stereovision is an unproven technology and at higher speeds, notoriously unreliable.
BRUCE HALL: Sorry.
NARRATOR: After a few scares, the Hall brothers scrapped stereovision and started building their spinning laser.
In the fields of Iowa, Team Terramax is a group of defense contractors. Like all competitors, they're not allowed to use government money to compete in the Grand Challenge. Still, for them, this is no hobby. Their vehicle is a big yellow goose that might lay a giant golden egg.
GARY SCHMIEDEL (Oshkosh Trucks): Terramax is a six-wheel drive vehicle. It is really the logistical backbone of the Marine Corps, and we think it's the most mobile vehicles in the world.
NARRATOR: Oshkosh has sold more than 6,000 of these trucks to the military. With the prowess of an Olympic athlete, it plunges into water, charges up hills and even takes to the air. So it's got the muscle, but what about the brains?
Without a driver, Terramax is more of a timid giant than a raging bull, but the team is confident of its future. In 10 years, the Pentagon plans to convert one third of all military vehicles to unmanned operation, and Team Terramax is very eager to help out.
Making autonomous vehicles is a tough problem, but apparently not tough enough for Anthony Levandowski, who's turning the Grand Challenge into a two-wheeled balancing act.
ANTHONY LEVANDOWSKI: A bike can do things that a truck and a car really can't do. You can do really sharp turns. You can go through a forest, really tight spaces. The agility is just there that's not on other vehicles.
NARRATOR: Anthony dropped out of grad school at U.C. Berkeley and spent two years teaching Ghostrider to stay up by itself.
ANTHONY LEVANDOWSKI: We're making a motorcycle that drives itself. Can you get any cooler than that? That's why everybody on the team puts in so many hours, because when you see it work, you say, "Wow, I made that."
NARRATOR: A year earlier, at the first Grand Challenge, Ghostrider made its world debut.
ANTHONY LEVANDOWSKI: I haven't slept in two days, and I'm super excited. I mean, this is the culmination of six months of work. And I forget to turn on the stability. And then the second I saw it start to fall and not steer, I said, "Oh, my god! I forgot the switch." That's life you know, bad things happen. But this will never, ever happen to me, ever again.
The Grand Challenge is basically...takes up all my life. Like, I don't really talk to my friends anymore; I don't have a girlfriend, right now. I think it's totally worth it, though. This could be the chance of a lifetime.
NARRATOR: Stanford University is a strong competitor, entering the Grand Challenge for the first time.
SEBASTIAN THRUN: I think the teams did great the first time around and performed very well. But it was obvious there was some missing link, some missing science to actually win this race. And these are the challenges I live for. That's why I do what I do.
NARRATOR: For almost eight years, Sebastian Thrun was Red Whittaker's colleague at Carnegie Mellon. They share a passion for robots. But Thrun has little interest in hardware and convinced Volkswagen to donate a vehicle with steering, brake and shift actuators that can be controlled by computer.
SEBASTIAN THRUN: The vehicle's name is Stanley. So Stanley is nothing else but Stanford, but it also gives the vehicle a personality. We think of the vehicle more and more as an intelligent decision maker.
NARRATOR: Thrun is a computer scientist with a deep interest in machine learning, and robots with personality are his specialty. His robots assist the elderly, conduct museum tours, fetch tennis balls, and each one embodies Thrun's desire to create machines that think.
WAYT GIBBS: Sebastian Thrun is interested in the science of robotics, in learning about new techniques that can make robots do fantastically advanced things 10, 20, even 50 years from now.
SEBASTIAN THRUN: I guess we've mounted them vertically, but we have no data, so far, right? And we have no software that can do it. Okay, so how foolish of us.
NARRATOR: Compared to the Red Team, Stanford is a small group with a single focus on software.
SEBASTIAN THRUN: There's a good number of teams who think the challenge lies in the hardware, and god bless them. It's great they're doing this, and I know there's hardware challenges. We, from the very beginning, pursued a very aggressive timeline to focus almost entirely on software.
NARRATOR: Stanford uses the same laser hardware as the Red Team, but they're experimenting with ways to use software to give their robot a greater role in making decisions about speed and safety.
SEBASTIAN THRUN: So, in this display, the white area is safe to drive, the red area is unsafe to drive, and the gray are, is...you don't know. And the robot, then, just has to steer itself so it stays in the middle of the road and avoids this red stuff, as much as possible.
NARRATOR: To test Stanley's software on the original race course, the team has traveled to the Mojave Desert, where the first Grand Challenge began. Most of Stanley's computer code has been written by Mike Montemerlo, Red Whittaker's former Ph.D. student.
MIKE MONTEMERLO: Give me steering and throttle.
SEBASTIAN THRUN: Steering is yours; total; you're going.
MIKE MONTEMERLO: Great. That's up to 15 now.
SEBASTIAN THRUN: What's the mile marker?
MIKE MONTEMERLO: Seven point two eight.
SEBASTIAN THRUN: Seven point three is the critical place. In fact, right after this turn over here, the Red team rolled off the road. And there's this final destination in the race, it's a very exciting place for us, it's a historical place. This is the end of the previous Grand Challenge.
MIKE MONTEMERLO: Swerve to the right.
SEBASTIAN THRUN: Sorry, we're having a problem.
MIKE MONTEMERLO: It's the end of the race for us.
SEBASTIAN THRUN: Yes. What happened, Mike?
MIKE MONTEMERLO: It just...the long lasers can't see us. They're blind in the corner.
SEBASTIAN THRUN: We're stuck because of a software bug, which we're currently debugging. So our challenge, between April...now and October, will be to get rid of these bugs.
NARRATOR: Unbelievably, Stanley faltered in the same place that killed the Red Team in the first Grand Challenge. But Thrun is confident about going head to head against a team he knows well.
SEBASTIAN THRUN: The Red Team is a seasoned competitor, and they have a lot of money, a lot of equipment. They have a fantastic team. I used to be at Carnegie Mellon. I know Red very well. He's an amazing person.
I wouldn't be surprised if the Red Team and the Stanford Team and maybe two or three other teams were the big teams in the end. We'll see. It's a race. It's a true race.
NARRATOR: The Red Team is teaching its robot to go fast, up to 50 miles an hour. Software guru Kevin Peterson has been spending long days and nights improving Sandstorm's performance at race pace.
KEVIN PETERSON: We've been working on Sandstorm for the last month. I just cut sleep out, you know? It's not that tough. You get three hours of sleep a day, drink some coffee, you're okay.
WAYT GIBBS: As a robot drives, it has to constantly update its picture of the terrain ahead. The faster it goes, the faster it has to update this picture or the terrain just gets on top of it. This is easy for humans to do because we can see out 300, 400, 500 feet ahead. But robots have sensors, like lasers, that operate at fixed ranges, 60 feet, at most 150 feet. It's like driving through a fog.
NARRATOR: Each team must lift this fog to win the race. The Red Team's secret weapon is hidden inside the dome on Sandstorm's roof, where a high-powered laser is mounted on a device called a gimbal.
RED WHITTAKER: The gimbal is a motion platform that's a little bit like an animal's neck. It can point sensors left, right, up, down, in such a way that the pounding motion from the vehicle is smoothed.
NARRATOR: Like binoculars, this special laser can see into the far distance. By damping vibration, the gimbal gives it a steady gaze and can point it to follow turns in the road.
The gimbal-mounted laser is unique to the Red Team, which has made several from scratch. But it's a fancy piece of hardware that's hard to build and easy to break.
SEBASTIAN THRUN: There's been a lot of discussion whether you want to gimbal your sensors. In fact, we've actually decided against it. We decided against it, because we wanted simplicity in the system. Every physical system that moves can break. And there's a lot of stuff you can do in software.
NARRATOR: The Stanford Team is putting its hopes in a set of off-the-shelf lasers linked through software to a simple video camera. The camera can see all the way to the horizon.
The trick is to give Stanley some way to distinguish the road in the video image. Stanley's lasers pinpoint smooth terrain right in front of the vehicle. By overlapping the laser and video data, Stanley can identify the color of the smooth area and search for it in the rest of the video image. The area Stanley believes is safe is now shown in red. By matching laser and video data on the fly, Stanley can adapt to changing road conditions and determine safe driving speeds all by itself.
SEBASTIAN THRUN: Making a car really intelligent requires to give it an ability to adapt. It's just like in human life. We don't enable kids to have a rule for every contingency. We tell them how to learn from contingencies and make them smarter, and then we trust them to learn.
NARRATOR: Stanford is the only team using adaptive vision to achieve race speeds. Whether it will work well enough for the Grand Challenge will soon become clear.
Back at the California Speedway, after more than a year of fevered preparations, the Grand Challenge is on. Officials from DARPA, the Defense Advanced Research Projects Agency, are judging the final runs of the qualifying event. Forty-three teams are competing for just 20 slots in the second, and perhaps final, Grand Challenge desert race.
After failing three times to complete the two mile obstacle course, Ghostrider is making its final attempt. It shows surprising agility, until it enters the tunnel. Looking like a wounded animal, it falls down and cannot get up. For Anthony Levandowski, the quest for the Grand Challenge is over.
ANTHONY LEVANDOWSKI: That was our chance, and it was unfortunate, but that's life.
NARRATOR: For the Hall brothers, it's also the moment of truth.
BRUCE HALL: This is looking good, this is looking solid. Barring any hardware issue, I think we're going to make it.
NARRATOR: At opposite ends of the track, the brothers Hall hold their breath as their new laser sensor guides the truck along the course.
DAVE HALL: All right, let's see if this obstacle detection works.
BRUCE HALL: Here's the test. Did all the years of work pay off? Turn, you
DAVE HALL: ...right around. It's almost as though it was programmed to do that.
BRUCE HALL: Dave, I hate to tell you, you're now in the race. You're in the race.
NARRATOR: By finishing the course, the Hall brothers have now earned a place in the main event.
BRUCE HALL: Hoo yeah, baby. You've got to love it.
RACE ANNOUNCER: Ladies and gentlemen, Terramax.
NARRATOR: Team Terramax also qualifies for the race.
Sandstorm's first run was not a dainty performance, but after the Red Team recalibrates its lasers, the robot loses its appetite for hay.
Sandstorm's younger sibling almost failed to make it to the start. Ten days earlier, on a practice run, Highlander swerved sharply and rolled on its roof. For the Red Team's prized robot, it looked like the end of the road.
Miraculously, Highlander is back in action, though hardly tested. It has a shaky start, but over the course of three runs, Highlander clocks the fastest times of any robot at the qualifier.
JOSH (RED TEAM): That thing's hauling.
KEVIN PETERSON: Frigging sweet, man. That's more like it.
NARRATOR: A first time competitor, Stanley is an unknown quantity. No one knows what to expect. The robot passes cleanly through the first gate. It shows impressive speed on the backstretch but is not as fast as Highlander.
Like a proud parent, Sebastian records the moment for posterity.
SEBASTIAN THRUN: Yes. So far, so good.
NARRATOR: Aided by adaptive vision, Stanley is the only vehicle that never hits an obstacle. In 36 hours, the race, the real Grand Challenge, will begin. DARPA announces who qualified and what the starting order will be.
TONY TETHER (D.A.R.P.A.): Highlander is Number 1 on Saturday morning; Stanford, Number 2 on the pole position; and Number 3, Sandstorm.
NARRATOR: The race course is a circular maze that zigzags for 132 miles across the Mojave Desert near the outskirts of Las Vegas. It's mostly flat, but rough and unpaved. It includes tunnels, cattle gates and steep mountain switchbacks. With a 10-hour time limit, the race will demand speed, agility and endurance that no machine has ever demonstrated on its own.
The exact route is a deeply guarded secret, but now, at 4 a.m., on race day, it's released to the teams. On a compact disk is a roadmap defined by 2,900 G.P.S. waypoints.
BRUCE HALL: All your hopes and dreams are on this disk.
NARRATOR: Teams have two hours to analyze the map and plug it into their robots before the race begins. In several minutes, the Stanford Team loads the course into their robot. Stanley will choose its own speeds on the fly, adapting to different types of terrain.
The Red Team takes a far different approach. Concealed in their trailer, Whittaker's troops look like they're planning an attack. The race course is layered on top of elevation data and satellite images stored in the Red Team's computers. The route is broken into sections and distributed to team members who will examine it in minute detail. Thousands of waypoints are added to fill gaps in DARPA's map, and each point can be adjusted to optimize the robot's path. Speeds are set for every inch of terrain. Red believes his robots must have as much information as possible to gain the winning edge.
RED WHITTAKER: Our preference is to err in the direction of overkill. I wouldn't want to reflect, Sunday morning, that we low-balled it.
NARRATOR: On a course that appears easy, Red makes a critical decision. He decides to run Highlander at a fast and risky pace. Sandstorm will be the tortoise, running 40 minutes slower, in the hope that at least one Red Team robot will finish.
RACE ANNOUNCER: The first time it's ever been done...
NARRATOR: As dawn breaks over the desert, robots prepare to boldly go where no robot has ever gone before. Vehicles will be launched at five minute intervals and race against the clock.
SEBASTIAN THRUN: Hey, Red. Red. All the best!
NARRATOR: Highlander leads the pack, followed by Stanley and Sandstorm. The cheering teams are now spectators. Their work is done. After months of tireless effort, there's a lot at stake. A vision they all share will now be put to the test.
During the first eight miles of the race, Highlander gains two minutes on Stanley. At times the Stanford robot seems confused and disoriented. Sandstorm is keeping pace with Stanley, five minutes behind.
It will take two hours to launch all the robots. Each one leaves the chute with confidence, a far cry from the first Grand Challenge when many faltered within sight of the start, and no robot went beyond seven miles.
KEVIN PETERSON: The start of the race was pretty cool. It was really exciting to see all the robots go. I was excited and apprehensive at the same time. You just don't know what's going to happen with the robot. It was entirely out of our hands.
NARRATOR: Far from the crowd, Highlander now leads a fleet of driverless machines racing the clock across the desert. Team DAD has passed two robots and is now heading to the front of the pack.
Eighteen robots have now passed the seven-mile mark that ended the last race. Miraculously, the Gray Team, from New Orleans, is holding its own, only weeks after the city was destroyed by Hurricane Katrina.
Terramax was the last to start and brings up the rear. Following each robot, DARPA chase vehicles relay information back to the start.
Here, teams gather around DARPA's displays.
NICK MILLER: It's just pretty sweet. We're pulling away in the front.
KEVIN PETERSON: We're pulling away?
NICK MILLER: For now.
KEVIN PETERSON: We're pulling away from DAD?
NICK MILLER: No, from Stanford.
NARRATOR: Thirty miles into the race, Highlander is making time on Stanley. Behind Stanley, Sandstorm is closing the gap and may pass the Stanford robot. Behind them, robots are starting to drop like flies.
At mile 26, Team DAD's laser comes unbolted from the roof. Officials in the chase car hit the kill switch, and for the Hall brothers the race is over.
At a media stop, a group of journalists sees Highlander charging toward them in the distance.
WAYT GIBBS: Highlander seemed to come down the hill very smartly, but then, as it started to mount the next hill, it looked like it ran out of gas or the engine cut off or somebody in there threw it into neutral. It just stopped moving and slid back down to the bottom of the hill. Something was wrong with Highlander, and it did not have its normal get up and go.
NARRATOR: Highlander keeps on rolling, but cannot maintain its intended pace. Red and the team slowly realize that things are not going according to plan.
KEVIN PETERSON: I don't know what's going on with Highlander. It's weird.
NICK MILLER: It's running slow.
KEVIN PETERSON: It's running real slow.
NARRATOR: Highlander struggles to summon the power to complete its mission. Behind it, Stanley is rapidly approaching. But Highlander recovers on the flats, and for many miles they continue neck and neck. One hundred and two miles into the race, Stanley's video camera records a milestone event.
RACE ANNOUNCER: Stanley has passed Highlander.
KEVIN PETERSON: At that point, we pretty much knew Highlander wasn't going to win the race. We thought Sandstorm might have a chance. It was really disappointing.
NARRATOR: Five hours after leaving the starting line, Stanley now leads the pack, and just five robots remain on the course. To finish, they must wind through a treacherous mountain pass, and Stanley must now make life or death decisions.
Somehow, Highlander has managed to reach the summit of the pass, but the gimbal is turned sideways and appears to be dead. Without its secret weapon, it must rely on short range lasers and G.P.S. to avoid disaster.
Right behind Highlander, Sandstorm is in good shape, but driving slowly, according to plan.
The crowd gathers. A blue dot appears in the distance.
STANFORD TEAM MEMBERS: Stanley!
Stanley!
NARRATOR: After driving six hours and 53 minutes, at an average speed of 19 miles an hour, Stanley is about to become the first vehicle in history to drive 132 miles by itself.
Twenty one minutes behind Stanley, Highlander limps to the finish.
Minutes later, the trusty veteran appears.
SEBASTIAN THRUN: It was just amazing to see this community of people. That community succeeded today. Behind me, there are three robots that made it all the way through the desert, and all three of them did the unthinkable. It's such a fantastic success for this community, I think we all win.
NARRATOR: Before announcing the winner, DARPA officials honor the five finishing teams. Terramax is fifth. A dark horse, the Gray Team from New Orleans, is fourth. Highlander is third. Sandstorm, Whittaker's tortoise, was just 11 minutes behind the winner.
TONY TETHER: I guess it's time. And the check goes to Stanley!
WAYT GIBBS: What Stanford did, in a short span of time, is amazing in the robotics world. On this scale of time, to develop a system this sophisticated, that can navigate so well over such a wide variety of terrain, at such high speeds, it really is worth its place in the history books.
NARRATOR: The source of Highlander's problems remains a mystery, but may have been the result of its rollover.
RED WHITTAKER: The engine punked on us. The only thing that was short was not having a top-end speed on the gas pedal. Except for that, we're in like a charm.
WAYT GIBBS: This really was like a horse race; they finished in such close times. So the Red Team is justifiably proud for having both of its robots finish the course.
NARRATOR: Many believe the completion of the Grand Challenge is a giant step that will push robots out of the lab, into the real world. The forces that will make this happen are embodied in the passion and vision of two men.
WAYT GIBBS: The world needs both Sebastian Thruns and Red Whittakers. If robotics is ever going to become part of the daily fabric of our lives, it needs those scientists who can come up with those far out ideas for how machines are actually going to start to think, to reason. And it also needs the builders, the implementers, and those who are driven to make things work in the real world.
NARRATOR: There is every reason to believe all those who participated will continue their quest to innovate, even though the Grand Challenge itself is now history.
So what about the future? Will mechanical warriors take over the battlefield? Will we all be passengers on the highways of tomorrow?
Someday, the great robot race may well be remembered as a moment when the partnership between humans and the machines we create was forever changed.
No comments:
Post a Comment