Stanford University wins DARPA Grand Challenge race for robots - five complete course
The DARPA Grand Challenge race for autonomous robotic vehicles has been run and won, with five robots completing the 132 mile course and the first four all finishing within minutes of each other. History will record however, that the winner was Stanford University’s Volkswagen-based "Stanley" beating out the two Carnegie Mellon Team Red entries by 11 minutes and 21 minutes respectively, with the Gray Team a further 16 minutes behind in fourth place. Had minor circumstances played out differently, any one of those four teams could have taken the US$2 million first prize and a place in history. Read on for a full report from Gizmag’s Robotics reporter, Dan Christian with images and assistance from Eric Zbinden.
As the Stanford vehicle crossed the line after 132 miles, the team's followers cheered and lifted team leader Sebastian Thrun shoulder high. Thrun is the director of Stanford's Artificial Intelligence Lab and Associate Professor of Computer Science at Stanford University. Congratulations to Stanford Artificial Intelligence Lab and to the team, and congratulations too to Carnegie Mellon robotics professor William "Red" Whittaker who put two vehicles in the race and finished with a close second and third place.
The event was a far cry from the results of the first DARPA Challenge, where the best-performed vehicle traveled just 7 miles – all but two of the 23 vehicles that started this event bettered that performance and four completed the course inside the allotted time. Most importantly, the Defense Advanced Research Projects Agency (DARPA) sponsorship of the event over the last two years has yielded pure gold – the US$2 million prize has catalysed a dramatic acceleration in the development of autonomous ground vehicle technologies and demonstrated conclusively that autonomous robotic vehicles can travel long distances across difficult terrain at militarily relevant rates of speed.
Five robots successfully completed the grueling 131.2-mile course in the Mojave Desert on Saturday (the Gray Team finished fourth) , with the Team Terramax entry completing the course on Sunday, October 9.
The vehicle that completed the course in the shortest amount of time was “Stanley,” entered by Stanford University. The team wins the $2 million prize because it finished the entire course in the shortest elapsed time under 10 hours – six hours, 53 minutes and 58 seconds (6:53:58).
Two vehicles entered by Carnegie-Mellon University, Red Team’s “Sandstorm” (7:04:50) and Red Team Too’s “H1ghlander” (7:14:00) finished close behind. The Gray Team’s “KAT-5” finished at 7:30:16. Oshkosh Truck’s 8-ton TerraMax, also finished the course, but on Sunday. Its official elapsed time exceeded the 10-hour limit.
The first four finishers entered the history books as being the first ground vehicle robots to travel a great distance at relatively high speed within a specified time frame. Stanley’s average speed over the 131.6- mile desert course was 19.1 mph. Sandstorm averaged 18.6 mph, H1ghlander 18.2 mph, and KAT-5 17.5 mph.
“It’s incredible what Stanford and the two Carnegie-Mellon teams did today,” said DARPA Director Dr. Tony Tether. “When the Wright Brothers flew their little plane, they proved it could be done,” Tether continued. “And just as aviation ‘took off’ after those achievements, so will the very exciting and promising robotics technologies displayed here today.”
“We established the Grand Challenge program to help foster the development of autonomous vehicle technology that will some day help save the lives of Americans who are protecting our country on the battlefield,” said DARPA Grand Challenge Program Manager Ron Kurjanowicz. “The outcome of this great public event demonstrates that we have succeeded in our mission.”
“These vehicles haven’t just achieved world records, they’ve made history,” said Tether. Pointing out that DARPA’s mission is to accelerate the development of promising technologies, and then turn them over to others for the development of viable applications, Tether continued, “We have completed our mission here, and look forward to watching these exciting technologies take off.”
DARPA Grand Challenge Program Manager Ron Kurjanowicz added, “The Grand Challenge stimulated the creation of a new community of innovators – inventors, mechanics, computer scientists, engineers, and students – who typically have not been involved in Defense-related activities. The camaraderie and competitiveness that have been the hallmark of the Grand Challenge since its inception demonstrates that America’s heritage of ingenuity and resourcefulness is strong.”
It all started at 4AM Saturday when DARPA revealed the race route to the teams for the very first time. The teams could then review the course against their own maps to determine strategy for the race. At 6:40 AM, the Red Team's H1ghlander left the starting gate. Stanley from the Stanford Racing Team started five minutes later and the Red Team's Sandstorm five minutes after that.
As in the very similar format employed by car rallys, though the cars are effectively competing against each other, they are racing the clock, so the order on course is not necessarily the order of performance. You can start last and still win the race. To make it even more complicated, the race officials can tell the robot to pause to avoid congestion on the course, and when the dust had settled at the end of the day, we knew that the race had been run and won, but we weren’t sure which of the four teams would take the prize because “pause time” is not counted in the official vehicle time and so everybody held their breath (because the leaderboard was not accounting for the pauses).
H1ghlander had the lead early on. It came through the 68 mile point about seven minutes ahead of Stanley. But around the 100 mile mark, Stanley pulled past H1ghlander in a dry lake bed, and the crowd just went wild. By passing H1ghlander, Stanley showed that it was not only ahead, but at least five minutes ahead. When it finished 20 minutes ahead of H1ghlander (unofficially), it was really a 25 minute lead.
To complicate matters, Sandstorm was paused several times to avoid running too close to H1ghlander (Stanley had already passed it). There are only a few passing stretches in the route, so they pause the following vehicle to avoid corrupting its time due to "traffic". Since those pause times don't count, Sandstorm could have had the better overall time.
Meanwhile, there were other problems associated with the 20 other robots running the course. Some of them started very late, because Team Caltech's Alice got off course and ran up onto one of the safety barriers. The race officials towed Alice off the course and repaired the barrier before allowing any other robots to pass. Between the starting stagger and the delay to repair the barrier, some teams started more than two hours after H1ghlander.
It's easy to dismiss these teams, because they didn't rank well in the National Qualification Event, but that can be misleading. All but two robots travelled farther than the best robot from last year. In particular, two robots performed must better than the NQE would have predicted.
The Gray Team's KAT-5 started 16th and made a great run. Based in Louisiana, they have a decidedly low tech sponsor: The Gray Insurance Company. In fact, the robot was originally called "GrayBot" until the category 5 hurricane "Katrina" wiped out most of the team’s homes. So they renamed the robot, and kept on going. And going, and going. In the desert twilight, 12 hours after the first robot started, KAT-5 crossed the finish line. KAT-5 should get a prize for finishing with the lowest budget and the biggest obstacle to overcome.
Most of the other robots have broken down by this point. There was only one other robot still in action when KAT-5 crossed the finish line -- Team TerraMax's TerraMax (#21). Due to darkness, officials halted TerraMax for the night. The robot probably would have been fine in the dark, but they were worries about the safety of the chase team and course crew. So they shut it down for the night, and resumed the next morning.
TerraMax is a unique vehicle in the event. It's a specially modified 8 ton military truck. They cut most of the cab off to be able to fit under bridges and underpasses. Then they add 6 wheel steering to improve its ability to fit its 8 foot wide frame onto tight dirt roads. TerraMax had 4 good runs in the NQE, but they had to go slow to keep in the narrow routes used for the testing courses. On a wider route, it will happily go 45mph.
The robot race of the century did not disappoint – after the ignorance of the mass media was displayed so clearly with its mockery of last year’s efforts, the ingenuity of man shone through and massive gains were made in understanding what is required for a vehicle to travel on its own, without any human involvement and decision-making, across 132 miles of desert and thousands of obstacles. Watching the teams on the day before the event was a privilege. Most were sitting calmly, knowing that all the work had been done months ago, and ther robots were as ready as they’d ever be.
Most are doing last minute adjustments. A few were performing major repairs. Many teams have dozens of people that have dedicated the last few months to testing their robots. They have come to respect how tough the desert can be. If they can't win the race, they still want to conquer this hostile terrain and the field was full of immensely capable machines that 12 months ago would all represented the pinnacle of human knowledge in the area. DARPA has done the world an enormous service as the technologies that were developed specifically for this race will filter down into the robots which will populate our homes, assist us to run our lives, and most likely assist us to drive our cars more safely.
Murphy's law (anything that can go wrong, will go wrong) was in full force on the day prior. The Cornell team (#26) had their generator seize up the day before the race. They swapped in a spare in the hours prior to the flag dropping. How dedicated are the team? Well, they had a new transmission flown in from Singapore and installed during the NQE.
The top teams had spares of every major component, including a spare vehicle. They wanted every chance of taking the $2 million dollar prize. For example, the eventual winner, team Stanford (#3) has autonomous diagnosis on-board. They can power cycle problematic systems and resume racing.
The Red Team that filled the placings (#19 and #25) can run a virtual course built using detailed satellite maps. The robots had "experienced" the course before they even got to the starting gate. This cued the robot when to be careful and when to be aggressive.
The race itself was surreal – most unlike most motor racing events where if telemetry isn’t available, voice communications with the driver plays a significant role in the strategy. The start and finish were located on a dry lake bed, so the robots were far away before they hit the harder terrain. There are no direct communications from the robot allowed, so their creators could only watch the data feed provided by DARPA. Until their robot crossed the finish line or DARPA told them where to pick it up, they could only watch and wait.