2005 National Qualifying Event (California Speedway) Results

First Run (September 28). Golem 2's first speedway attempt was a total failure as it immediately veered off to the right and crashed into one side of a gate meant to simulate a cattle crossing. It knocked down the fence beside the gate and came to a stop. On investigation we found several things that had gone wrong.

The robot's start position was forty feet outside the official GPS corridor it was supposed to stay on (unlike last year when the start position was inside the allowed corridor) and we did not handle that properly.
A missing decimal point in the calibration of one of our laser sensors caused the sensor to see lots of obstacles that weren't there. This made the robot swerve to the right to avoid the phantom obstacles.
The robot saw the real obstacle (the fence) that it was about to hit, and tried to brake. But it reacted a little late and didn't come to a stop until it had already knocked the fence down.
After it had knocked the fence down, Golem 2 could have moved on and tried to do better on the rest of the course, but instead it chose to remain at a permanent stop.

Second Run (October 1). When we fixed the bugs that appeared on the first run, Golem 2's performance improved dramatically. On its second run, it slightly grazed the same fence that it had crashed into before, but after that its performance was perfect. It sped around the high-speed section at 40 miles per hour. It passed through the dreaded tunnel that had caused problems for many other vehicles. It threaded its way around stacks of tires, three parked cars, and a tank trap, and finished the course.

This run was the fastest completion of the course at the time. Later on, only the Stanford and CMU teams would manage to complete in faster times while doing equally well at avoiding obstacles and traffic cones.

Third Run (October 2). The course was changed around a bit so that there was less asphalt, more grass, and an additional parked van to get around. Golem 2's performance was absolutely flawless. It finished in a fast time while avoiding all obstacles down to the smallest traffic cone. Stanford's Sebastian Thrun commented on our performance, "Wow, from zero to a hundred."

Only Stanford, CMU, Princeton, and Cornell managed runs that were as perfect as this one and in a faster time. Princeton and Cornell beat this run by one or two seconds.

Fourth Run (October 3). Another software bug emerged on the Monday run. As Golem 2 navigated a narrow passage next to a concrete barrier, a glitch in the planned path caused the robot to smack into the barrier and break its steering mechanism. Golem 2 was rushed back to UCLA for a quick repair by the excellent UCLA machine shop.

Fifth Run (October 4). This run was just okay, but it demonstrated that Golem 2 was working again after its accident the day before. The robot knocked over a traffic cone and a stack of tires, but it finished the course in a time that was faster than the third run.

All the vehicles that ran ahead of us that day had trouble with the final leg of the course, where a tank trap sat in front of the finish line. Tony Tether, the Director of DARPA, bet one of our team members a dollar that the tank trap would scrape Golem 2 as well. "We're not supposed to be betting on the vehicles," said one of the other DARPA officials. "Don't worry about it," said Tony Tether, "We're not betting on the vehicle, we're betting on the tank trap."

Golem 2 sailed smoothly around the tank trap and we collected our dollar. This made us the second-highest money-winning team at the DARPA Grand Challenge.

2005 Grand Challenge Race Results

Golem 2 raced out of the start chute in the seventh pole position. Our plan was to observe DARPA recommended speeds at the start of the race and gradually ramp up our speed from there, until by the end of the race Golem 2 would be driving fifty miles per hour, the maximum allowed speed, whenever it could.

On a paved bridge, Golem 2's laser sensors perceived the reflective Botts dots as obstacles and the vehicle slalomed back and forth between the reflectors. This alarmed the DARPA observers in the chase vehicle, but the robot continued to make good progress.

Once it got back onto the dirt road, Golem 2 straightened out and zoomed ahead at over thirty miles per hour. It went through the 16-mile checkpoint with the fastest time of any vehicle. Our DARPA observers began to believe they were following the winning vehicle.

Golem 2 passed the IVST1 vehicle, which had run into difficulties, and then closed in on the Axion Racing vehicle, which had left the start line twenty minutes ahead of Golem 2. In fact, Golem 2 caught up to the Axion vehicle three times, and was paused three times so that Axion could get ahead to a better passing area. At the time of Golem 2's accident, Axion was just about to be paused so that Golem 2 could pass.

Tragedy struck after twenty-two miles. It turns out that we had overallocated the computer's memory, but this did not become apparent until we had actually used a lot of it to record data. We had never driven such a long and winding course without stopping the program before, so we had never encountered this memory bug. (We were fully aware of the desirability of a hundred-mile endurance test, but with limited time and resources, we never got around to it.) The computer's memory filled up and the program quit.

The vehicle's controls remained fixed at their positions, which happened to be "steering angle 2 degrees left" and "accelerate hard." Golem 2 roared off the course into the desert brush. The DARPA observers hit the pause button but nothing happened, since the vehicle software was no longer running. The DARPA observers had a disable button that would have killed the engine but they chose not to press it. Golem 2 continued to crash through the cactus at sixty miles per hour, flying through the air, leaving devastation in its wake. The tail gate opened, wham, wham, wham, and broke off. Wham, wham, and the tail lights fell off. The battery bounced around in the engine compartment and broke its battery tray. Finally a connector pulled free from the fusebox and the engine died. Golem 2's Grand Challenge attempt was over.

The race driver who drove the DARPA chase vehicle begged us to get him on-board video of the drive through the brush, but unfortunately, we have none. We don't even have laser imaging data, since the computer had stopped working.

All things considered Golem 2 is in fairly good shape, apart from the broken tailgate, tail lights, and battery mount. Our computers and sensors appear to have survived unscathed. It looks as though we chose a good tough platform after all.

Because of our high speed, we were on pace to finish the race significantly ahead of Stanford and the other teams who actually finished. We might have been the fastest team of all (but Team DAD and Team Ensco deserve mention as other teams who drove very fast). It's entirely possible that we would have won the two million dollars with a tiny change in our memory allocation. But that's the way it goes.

[A]t the worst, if he fails, at least he fails while daring greatly, so that his place shall never be with those cold and timid souls who know neither victory nor defeat.
--- Theodore Roosevelt

NQE

California Speedway	California Speedway	Garage; wheels off
Avoiding tire stack	Avoiding parked cars	Avoiding tank trap

DARPA Grand Challenge 2005

Pre-start	In chute	Start
Chase vehicle	Thirty miles per hour	Status board; Golem 2 catching up
Damage	Damage