Darpa Grand Challenge     

In February of 2002, Lorin told me about an event coming up in 2004 that was being put on by DARPA. DARPA is the Defense Advanced Research Projects Agency. They build the Saturn rocket, the M-16, and the start of the internet. They build weapons. The event was a competition to create a robot car that could traverse 200 miles of desert between Los Angeles and Las Vegas in under 10 hours. The prize: One Million Dollars.

Early on I decided that I couldn't really devote the time needed to enter a bot in the contest. Instead, I eagerly stayed up with the message boards, the teams, and the announcements from DAPRA. Over a hundred teams worked on entries for the competition. Most teams were computer geeks with an idea on how to merge sensors from GPS, lasers, ultrasonics, and anything else they could think of with the motor, steering, and brakes of their vehicle.

Unfortunately, DAPRA seriously underestimated the number of teams who would have a working robot by event time. So they made a change to the rules, only 25 teams would be selected to come to the qualifying round. Many teams felt cheated, they had invested a lot of time and money only to see the government change the rules only a few months before the event.


Qualification

From March 8th thorough the 12th the 25 invited teams had to complete a 1.7 mile test course in the infield of the California Speedway in Fontana, California. The course had several examples of terrain, going from grass to pavement to emulate the change in terrain. Sand bags and metal pipes to simulate rocks and cattle guards. Even a minivan parked in the middle of the course to simulate passing other vehicles. Each vehicle was required to traverse the course without hitting any of the obstacles.

The contest was to build a vehicle with absolutely no human control. To mark the course, the teams would be given a list of 500-9000 GPS waypoints to hit. The list was to be kept secret until just a few hours before the race so that the bots couldn't just memorize a course beforehand. A leak of the proposed course map narrowed the number of potential routes to 19, Red Team did detailed maps of those routes, and ran five of them before the competition. A bit unsporting and against the spirit of the rules, but you could also say it was thorough preperation.

At the QID (Qualification, Inspection, and Demonstration) at Fontana, the first vehicle to complete the course was Red Team on Tuesday. Their 1985 Humvee cost three million dollars to develop. The 50 member team drawing heavily from Carnagie Mellon University in Pittsburg. The rest of the bots that we saw run either got lost, had mechanical problems, or had problems detecting obstacles such as a metal fence or the minivan. The largest bot, TerraMax, at around 20,000 pounds smacked the minivan square on and shattered its rear window.

By Friday's deadline, fifteen of the twenty five bots had qualified. They were cleared to attempt the race starting at 6:30am in Barstow, CA.


The Race

After staying up past midnight at an Irvine Underground meeting in Irvine, Not5150 and I drove to Redlands to catch a bit of sleep before heading up to Barstow to make the five AM press briefing. We arrived before the sun rose, ten miles into the desert from little Barstow. We joined a caravan of cars heading into the void. Topping a hill we got our first look at the starting line. In the shallow valley beyond was a pool of light created by the mobile lights used by road crews. As we got closer we passed satelite uplink trucks, military troop transporters, and several police and sherrifs vehicles. Turning into the parking lot we were directed past hundreds of cars parked neatly in the empty desert. The venue was apparently used for other off road races through the area.

We grabbed out camera gear and headed for the press tent. The drive took longer than I had expected, we missed the press briefing by half an hour. I was able to pick up my press pass and bright orange safety vest without any trouble. A man next to me tried to social engineer his way into the course map given to the press but the ladies in charge held firm. They finally said that he could copy a map by hand but he couldn't take an entire packet.

The first thing we did was to join the crowd. Hundreds of reporters were on the scene. The star of the day was Red Team. Their Hummer had had the fastest qualifying time so they were to be the first team out of the gates. We got right up close and snapped some great pictures. While they were pulling up to the starting line, we proceeded to the press box. The press box was next to the public grandstands but was isolated by an eight foot chain link fence. The best part about having a press pass is that you can take pictures from up close without anything in the way.

Of the fifteen bots that made it to race day, only nine made it out of the starting area. The best start was by ENSCO, the small six wheeled rover. It zoomed out of the starting area far faster than any of the other contenders. It tore around the corner and was well on it's way to setting the lead. Two tenths of a mile down the course it hit an embankment and rolled onto its side.

By noon the race was over. Every team had been disqualified, many due to mechanical breakdowns. Though some may consider the event a disaster, I was impressed that non professional groups were able to build driverless vehicles that traversed the desert landscape for up to seven miles.

  • Red Team: 7.4, went off course, mechanical damage
  • SciAutonics II: 6.7, hit embankment
  • Digital Auto Drive: 6.0, stuck on rock
  • Golem: 5.2, unable to climb first hill
  • Caltech: 1.3, went off course
  • TerraMax: 1.2, thought bushes were solid, kept reversing
  • SciAutonics I: 0.75, went off course
  • CIMAR: 0.45, hit fence
  • ENSCO: 0.2, hit embankment, turned over
  • Virginia Tech: brakes locked up at start
  • Axion Racing: lost GPS at start
  • CajunBot: hit wall at start
  • Palos Verdes HS: hit wall at start

For the next event I make the following suggestions:

  1. Don't do it in the desert. Start with a simpler contest, like a road course. The best teams will have reliable software that lets them do many laps. Let the fastest teams compete in the next stage, the off road course.
  2. Run the race like a FIA Rally. The race is divided up into stages. After each stage, points are awarded based on each team's time. If a vehicle is unable to complete a section then they may fall in the standings, but they're not disqualified completely. Teams are allowed an hour or two to work on their vehicle between stages. This also allows for transport, say, across a freeway or rail line, or to change terrains.
  3. Instead of arbitrarily choosing 25 teams to attempt to qualify, let anyone who wants to come come and try. Take the top 25 teams from qualifying.











  Site by Kallahar - kallahar@kallahar.com - Hosted by DreamHost