Red Team falls to its own offspring
Former CMU professor and students overtake Red Team for Stanford
Carnegie Mellon?s Red Team and Red Team Too autonomous vehicles crossed the finish line in a historic race on Saturday with times of 7:04:50 and 7:17:00, one right after the other. The two Humvees finished just behind the Stanford Racing Team, which had the fastest time, clocking in at 6:53:58.
?Sometimes it?s the performance, sometimes it?s the quality, sometimes it?s raw speed. Sandstorm and H1ghlander did superbly in all,? said William ?Red? Whittaker, CMU?s Red Team leader.
Though Whittaker refrained from naming teams, the real surprise of the pack was Stanford, which entered Stanley, a Volkswagen Touareg. Averaging 19.1 mph, Stanford gained the lead just before the third-to-last obstacle of the course.
The Stanford team, ironically, has many CMU connections: Team leader Sebastian Thrun is a former computer science professor at CMU. Several CMU grads, including Michael Montemerlo, a former Ph.D. student under Whittaker, are on the team as well.
?There are plenty of my students and our [CMU] grads that are part of the culture and in lots of these teams,? Whittaker said. ?In a lot of ways, no matter who wins, I win.?
Sponsored by the Defense Advanced Research Project Agency (DARPA), the Grand Challenge sought to bring together the brightest minds to create the first fully autonomous vehicles that could cross miles of off-road terrain successfully. The first challenge in took place in the Mojave Desert in 2004, with disappointing results. None of the teams finished the race, though CMU?s entry, Sandstorm, traveled the farthest: 7.36 miles.
This year saw CMU return a veteran, entering Red Team with Sandstorm, and a second team, Red Team Too, with H1ghlander. ?This isn?t CMU?s first rodeo,? Whittaker said. ?We are in the robot business, period.? Carnegie Mellon was the only organization to submit two entries in the Grand Challenge. They were two among 195 total entrants from 36 states and five countries vying for 20 spots on the start line on race day.
Carnegie Mellon?s vehicles performed exceptionally, with one-of-a-kind hardware and technology. Red Team?s Sandstorm is a triumph of reinvention. The 1986 Model 998 HMMWV (High Mobility Multi-purpose Wheeled Vehicle) ?is a 20-year-old junker that [Whittaker] bought from a farmer.? Red Team Too?s H1ghlander represents the ?latest, greatest, and also commercial technology. It has all the bells and whistles you expect in a real hot truck or a real hot off-road machine,? Whittaker said.
While there is an extensive amount of attention to hardware, ?[the] most powerful technology is preplanning the curves in the hours or so before the race when the route is revealed,? Whittaker said. ?We compute the terrain and the details of where we?re going to drive and how fast we?re going to drive, and that is definitely the thinking part of the race. It?s a long, long day in the dirt.?
The technology incorporated into both cars is like nothing done before. In layman?s terms, it could be MTV?s ?Pimp My Ride? to the millionth power. Both vehicles use similar navigation technology to pick out their paths over terrain. Plan sensors on the vehicles include GPS, radar, four cameras, and seven LiDAR scanners.
LiDAR, which stands for Light Detection and Ranging, is a particularly useful feature of the vehicles. Light is emitted from the LiDAR and interacts with a target object. The light is subsequently changed, reflected back to the lidar instrument, and analyzed. The incoming light is then used to determine the size, shape, and distance of the target.
A basic LiDAR device consists of a transmitter, a receiver, and a detection system. LiDAR uses radiation at wavelengths 10,000 to 100,000 times shorter than radiation used by ?conventional? radars. LiDAR can be bistatic ? which requires two locations to operate, one for the transmitter and one for the receiver ? or monostatic, which only requires one location to operate. Monostatic LiDAR is preferred over bistatic because there is only one location to staff, one instrument to operate, and the ?aligning and pointing? feature is more straightforward.
Both vehicles use aerial LiDAR. Three cameras and one LiDAR scanner are suspended in a floating gimbal within the vehicles. The camera and LiDAR sensor are mounted on an extremely stable swivel, giving the vehicles rotating ?necks? with which to scan their environment and point them along the intended route.
The sensory mechanisms are constantly working to provide navigation feedback and pick the best path along the terrain. A ?path corrector? will alter the preplanned route if any obstacles get in the way. As the vehicles continue, 6000 candidate paths are generated each second. Ideal paths are put together so quickly that they merge into one smooth continuum, allowing the vehicle to simply plow ahead on its course.
The race course, announced two hours before the start time, covered 132 miles of Nevada desert. The most treacherous obstacle awaited challengers at the Beer Bottle Pass. Only 1.5 miles long, the Pass is so narrow that a wrong turn here could have cost challengers the race ? and even their vehicles, sending them plummeting 100 feet over a cliff into a canyon. Then again, this is precisely the kind of terrain that the Red Teams were built to take on. ?We have rock-solid position estimation,? Whittaker said. ?Our robots really know where they are; they know where they are going; they know how to get there; and they are tremendous at driving and maneuvering.?
Though Stanford walked away with the $2 million grand prize, up from the promised $1 million last year, the benefits of even participating in the Grand Challenge, much less completing it in unprecedented times, are far reaching. There is a whole movement ?just hungry for these technologies.... All those things are in huge demand,? Whittaker explained. ?The idea of robot mobility is fundamental to the advancement of robotics.... The technologies are not just mobile but also high performance. These are serious companies. You build new alliances, a new generation of people; you need the best and you need a lot of them.?
According to Whittaker, there is no limit to where CMU could go from here: ?That?s just the tip of the iceberg. Future revenue is huge in research and industry.?
Job opportunities are ripe for participants, too. The iRobot corporation, which Whittaker explained wasn?t racing, simply came ?to pick off people. They are hiring one person each day.?
Make no mistake, though, this is not just a playground for robot or car fans. This is serious business. ?Breakfast is 3 in the morning, then preplanning practice is at 4 am, and we are trackside by 6. You?ve got to be ready to go when that sun comes up,? Whittaker said. ?It?s like Buggy at CMU.?
Similar commitment and drive is required, according to Whittaker. ?There?s a lot of the CMU culture that really matters in a race like this.... Those are leaders of the future, and they are wonderful people.?
Everyone involved in the Red Teams will walk away from this experience better for it. ?Universities teach all kinds of things, including how to live life,? Whittaker said. ?Most things worth doing aren?t easy, and they aren?t fast. You play with what you got, and how things turn out, that?s the way they are supposed to be. The right thing to do is choose something you love, go after it with everything you got, and that?s what life is about.?