correction: this occurred in the 2005 Grand Challenge, not the 2007 Urban Challenge.

There's a short article from Princeton team member Brian Cattle on how a memory leak caused their Urban Grand Challenge entry to fail. In their testing, the system would fail after 40 minutes. Not knowing what to do, they set the code to restart itself every 40 minutes -- sounds good, right? During the actual event their car failed after 28 minutes. Doh!

The article is essentially a marketing piece for the ANTS Profiler, while writeups over at robots.net and Slashdot are trying to blame Princeton's choice of C# and Microsoft Robotics Studio. I'm happy to use profilers and bash Microsoft, but in this case their bug was programming mistakes 101: they subscribed objects to events and neglected to add the one line of code to remove subscriptions when those objects were deleted.

In Princeton's case the objects were the obstacles that the car was seeing. In their testing, it would take 40 minutes to see enough obstacles to cause their problem. The DARPA course had more obstacles and caused a failure in 28 minutes.

I've finished processing my photos from the DARPA Urban Challenge. The latest photos feature a lot more multi-car action, like Cornell blocking Junior and photos from the four-way intersection in Area B.

• MIT's Talos
• UPenn's Little Ben
• MythBuster's Jamie Hyneman
• Cornell's SkyNet gets stuck, blocking Stanford's Junior
• More Boss, Junior, and Odin photos

Terramax, with its comically ginormous size, was definitely a crowd favorite. Someone even told me that it got the loudest cheer at the start. But as much as we wanted to see Terramax succeed, we also wanted to see Terramax destroy.

I was confronted with a sad sight as I walked past the A-course parking mission. Terramax was stranded in the middle of the lot as VT's Odin breezed through and completed a mission. The engine noise that Terramax made as it sat idling in the middle of the lot was akin to a trapped animal or visually like a beached whale.

After several minutes, the noises changed. Terramax had made a decision: forward. Slowly it rolled over the median. Slowly it crossed the road. Slowly it rolled up the entrance to the Exchange and bonk -- pillar. I'm not sure why the DARPA chase car didn't stop this Fish-Called-Wanda-esque, slow-motion collision, but as a photographer I appreciate it.

The damage to Terramax was minimal -- not even a building is worthy of more than scratched bumper paint.

Photoset

Photoset

Cornell's SkyNet vehicle started to exhibit a very bad behavior during the finals: when it came to a turn it would stop, wait a minute, move forward a foot, and then stop again. In one of the worse instances of this, it blocked Stanford's Junior for around 20 minutes before team members had to come out and get it moving.

MIT's Talos found itself behind SkyNet during one of its start/stop fits. Talos saw the stopped car and did what it thought was best: it started to pass SkyNet. Unfortunately, Talos' timing was off and SkyNet started to surge forward again just as MIT was completing the pass. Their front sensors interlocked in an expensive embrace.

By the time I arrived the cars had just been pulled apart, but I managed to snap a shot of the left-bumper damage on SkyNet.

I only had time to get a quick snap off before UCF's car disappeared up the front yard of an abandoned house. The car was coming around a bend in the road and confused the dirt yard with actual road.

Unlike the rest of the eliminated vehicles, which were retrieved by team members and taken to elimination row, UCF's was left there for the remainder of the race. I walked around the fences to a different vantage point and was confronted with an anthropomorphically sad sight -- one of the car's SICK scanners was still tilting back and forth, confused, lost, and abandoned.

CarOLO was one of the last cars to be eliminated from the finals -- I can't remember if they were before or after UCF. I didn't see the crash, which was in off limits B territory, but I did get to see the drive of shame as the car exited under non-autonomous control. Their car and team made a valiant effort and the front bumper more so: it took the brunt of CarOLO's impact and probably saved the expensive IBEO sensor from damage.

Urban Challenge Finals Photos

Winners of the DARPA Urban Challenge have been announced. Congratulations to CMU/Tartan Racing, which takes first place. Second place goes to Stanford and third to Virginia Tech.

Official results haven't been released yet, but CMU's adjusted time was about 20 minutes faster than Stanford and 40 minutes faster than Virginia Tech. Amazingly, none of the teams were penalized for traffic violations and CMU's averaged 14mph over the course.

• Wired article
• Urban Challenge Finals Photos (more to come)

We drove down to Victorville at 1am and back at 4pm -- needless to say the words are a bit blurry right now and the photos won't get uploaded until tomorrow. Many thanks to my coworker who is a much better no-sleep driver than I.

In eight hours we'll find out who the official Urban Challenge winner is. Right now I'm leaning towards CMU's Boss, though I want it to go to VT for doing so well with far less -- perhaps there should be handicaps based on $value of sensors used.

Stanford's Junior crossed the finish line first, but it barely had time to get off the finish line before Boss crossed. Virginia Tech's Odin was close behind. The fact that those three managed to finish the nearly six hour course within minutes of each other is a sign of how well each ran the course, but reports were that CMU's run was more flawless.

The actual finish time only served as a rough grouping -- Stanford, CMU, and VT were clearly best, UPenn ran a solid conservative approach in the middle, and Cornell and MIT were last (of the finishers). CMU several minutes after Stanford and VT, but I also witnessed Junior stuck behind Cornell's car for 20-or-so minutes and another 10-or-so minutes behind the MIT/Cornell crash. It will be up to the judges to tally up points and minutes and come up with an official score, which is a bit of a flaw in the design of the current challenge -- no one really understands what the scoring system is.

Any of the finishers can lay claim to an impressive feat, so it only with alma mater pride that I take a slight dig at Cornell. Cornell's SkyNet and MIT's Talos were the worst of the best, finishing almost two hours after the top three and forty-five minutes after UPenn's slow-and-steady Little Ben. They seemed doomed to scrap it out after Talos attempted to pass SkyNet and ended up getting rammed -- the teams had to carefully pry the interlocked sensors apart.

It seemed that Cornell had the upper hand over MIT as SkyNet made it to the final traffic circle first. Then it stopped. And sat for minutes. With the finish line in sight. Talos pulled up and turned onto the finishing straight. SkyNet seemed to sense its loss as it promptly unstuck and crossed the finish line last.

I've posted my photos from the qualifying round of the DARPA Urban Challenge. As I only got to see 1.5 actual runs that day, most of my photos are from the pits. You can admire the sensory overload of CMU's Boss (above), or the seamless hiding of Team Lux, or the simplicity of Team Gray's "autonomous car in a box" (also above).

NOTE: most of the cars that I photographed in the pit are the teams' backup cars and are not always identical in configuration.

2007 DARPA Urban Challenge NQE Photos

The 11 finalists for the Urban Challenge have been announced: CMU, Stanford, Virginia Tech, MIT, Cornell, UPenn, UCF, AnnieWay, Intelligent Vehicle Systems, CarOLO, and OshKosh. Up to 20 were to be allowed, but DARPA decided that there weren't 20 vehicles that were safe enough to compete in the finals.

The finale will be a 6-hour, 60 mile race. It sounds like this will be the first time that the vehicles will be put on the same course as one another. If it's going to be a demolition derby, Team OshKosh will be able to roll over the competition. DARPA chief Tony Tether has tapped CMU's Boss as the best overall performing thus far and OshKosh as the best on the Course A left-turn-merge course.

I flew down to Victorville for the day to watch Wednesday's qualifiers. It probably wasn't the best day to visit as the good teams had already qualified -- CMU, Stanford, Cornell, UPenn, Virginia Tech, and CarOLO -- and the really bad teams (i.e. crash-worthy) had already been disqualified. What were left were the teams that, while not dangerous, didn't have all the bugs shaken out, which meant a lot of delays. In the six hours that I was there, I only saw one run, which occurred after about an hour of delays. The run was by Team Gray on the 'C' course, which seemed to go well enough for three laps before some sort of incident required the team to jump in the golf cart and go service the stopped vehicle. I also saw the comically large OshKosh vehicle get stuck trying to navigate its way to the B course.

The highlight was going through the team pits and getting to peak in everyone's cars. Team Gray's was one of my favorite cars because they've compartmentalized their processing to a small blue box with dual Pentium Ms, which stands in stark contrast the the 10-20 rack Core2Duo units that several other teams are using. MIT's and CMU's cars were covered with every sensor known to man. Hopefully I'll have time to post some photos soon.

Some of my co-workers got to see Stanford's smooth A course run, which was brought to an early close due to its great execution. Apparently it took them two tries to pass the course -- their theory was that DARPA's radios were causing interference with the velodyne. They also saw Caltech's A course run, which was dangerously amusing: the van ran the wrong way on the course, to the terror of the DARPA stunt drivers.

DARPA has hundreds of Ford Taurus' for the event. The cars are pretty cool -- the interiors have been replaced with a roll cage and bucket race seat, and many appear to have wireless cameras attached to the roof. You could easily shoot a Hollywood movie using the Urban Challenge as a backdrop: abandoned, decrepit military housing with identical DARPA vehicles parked in the driveways as robotic cars creep around: a utopian universe gone awry.

NOVA aired a special called The Great Robot Race covering the 2005 DARPA Grand Challenge. If you missed it, PBS has been kind enough to post the entire program online -- sweet.

NOVA also has one of my favorite video podcasts. Recently, they've been airing amusing mini-programs from the DARPA Grand Challenge. Great stuff for people who short attention spans like me that want our stories in 2-3 minute chunks.

Previously: * Sebastian Thrun's talk on Winning the DARPA Grand Challenge * DARPA Grand Challenge link roundup

I went to go see Sebastian Thrun speak at Stanford about his team's winning effort in the DARPA Grand Challenge. Thrun described the contest as how to stay on the road for a very long time. It was not a general path-finding problem: DARPA gives you the route with a corridor you have to follow, as well as speed limits you have to observe for various parts of the course. Of course, DARPA didn't give perfect data. He showed a video generated from the data of Stanley driving through the most dangerous part of the course: the switchbacks of Beer Bottle Pass with a cliff on one side. DARPA's corridor was overlaid on top of Stanley's sensor data and it was easy to see that much of DARPA's corridor was actually over the cliff.

During training many traffic cones were "frequent victims of computer glitches," but Team Stanley was called "Team Boring" by the cleanup crew for their lack of incidents. The actual challenge was described as getting the data at 4AM, getting Stanley to the starting line at 6AM, and then sitting around drinking beer for several hours. The big moment came when Stanley passed CMU's Highlander as the CMU and Stanford teams listened to race radio. Thrun narrated the exciting finish for us: "[The head of DARPA] is waving his flag as if the car could see it."

Thrun said that they won mostly through luck given how close four of the teams finished. The speed limits set by DARPA for the various parts of the course were too conservative, so the cars were running below their full potential. DARPA also decided to make the course fairly easy. Asked if CMU would have won had they not had engine problems, Thrun answered, "In all likelihood, yes." Also, Team ENSCO had a faster average course time but flatted on "something really big CMU left behind" (the CMU part may have been a joke). Thrun felt that Stanford had better software than CMU and on a tougher course Stanford would have the advantage.

In the future, Thrun wants to try driving 65mph on 280, parking in a garage, convoy driving, and driving assist. Part of his motivation is to reduce traffic deaths, which a driving assist system could help prevent. He also feels that a fully automated system would change society by allowing you to use your commute time productively -- you could even drive to your destination, get out, and then send your car to go park in a parking garage farther away. These are still looking far ahead. In response to someone asking what it would take to drive at human-controlled speeds, Thrun related it to asking the Wright brothers, "If you want to fly over the Atlantic, what's missing?"

This year's DARPA Grand Challenge looks like it was pretty awesome. People from work have been sending links to their photos (see below) and it's amazing to me to think that in the year since the inaugural competition there are already five autonomous robotic vehicles that completed the 130 mile course and the $2M prize has been claimed. I'm also happy to see that students from my high school were on a team (ENSCO) that made it two-thirds of the way through the course and finished sixth (highest of the non-finishers).

Yesterday I got to talk with someone who got to drive a chase car in last year's Grand Challenge. The vehicle he was following had been programmed to avoid obstacles and had a general plan for getting to the destination, but it didn't have much else. What happens when you only teach a vehicle to avoid obstacles? It finds obstacles and then avoids them, so instead of driving along the road, it would find two bushes along the side of the road and drive between them.

• Gallery w/ wrecked vehicle photos, the giant 16-ton Osh Terramax vehicle, and a short video of the robo-motorcycle
• Gallery w/ video of a vehicle bowling over a stack of tires
• Gallery of the main contenders starting and finishing
• Official video of Stanley navigating the Beer Bottle Pass portion of the course