Robotic urban warriors pass DARPA's Grand Challenge

For nearly 50 years, the Defense Advanced Research Projects Agency (DARPA) has funded scientific developments and promoted their adoption by the companies that supply military technology. For the last few years, the agency has been encouraging the development of automated vehicles through a series of Grand Challenges. The first two editions tested the ability of automated vehicles to navigate desert terrain, but this year's version raised the difficulty dramatically by changing the environment: the 2007 version was called the Urban Challenge.

DARPA's description of the obstacles sound a bit like a description of a typical commute to work: "They must obey traffic laws while merging into traffic, navigating traffic circles, negotiating busy intersections and avoiding obstacles." The competition rules, in fact, included a copy of the California Driver's Handbook. Instead of dropping carpoolers off at work, however, the vehicles were expected to run simulated supply missions that had them covering 60 miles in under six hours. As the Challenge's literature notes, "The urban setting adds considerable complexity to the challenge faced by the robotic vehicles, and replicates the environment in which many of today’s battlefield missions are conducted."

Given that no vehicles completed the first desert challenge, it's clear that experience has made a big difference for many of the teams. This time around, three teams finished within the allotted time. Computer science hotspot Carnegie Mellon's Tartan Racing took home $2 million for driving a Chevy Tahoe into first place. The Volkswagon Passat wagon of the Stanford Racing Team, winners of the second Desert Challenge, pocketed $1 million for second, while Virginia Tech's Team Victor Tango drove a Ford Escape Hybrid named Odin into the $500,000 third prize. According to Wired, all three vehicles averaged speeds in the low teens.

All of the vehicles were crammed with radar and laser sensors to detect their surroundings, used high-precision GPS sensors to figure out where they were, and they packed some serious hardware to support their AIs. For example, Stanford's vehicle used 64 lasers to create a 3D map of everything within a 65-meter radius, while using five radar sensors to track up to 32 individual obstacles at ranges of up to 200 meters; Carnegie Mellon's team put 300,000 lines of code into getting their Chevy around the course.

DARPA clearly foresees military uses for these vehicles, which is no surprise, given the military's successful use of unmanned aircraft. The fact that many of the Defense Department's recent casualties have come from bombs targeting manned vehicles has undoubtedly convinced DARPA that it is time to bring automated vehicles down to earth.

This is technology that will also attract an audience within the consumer automotive industry. Lexus is now selling a self-parking car, and many companies are testing sensors that let vehicles follow others closely at high speeds and brake safely when the car in front slows. The interest was also clear from the sponsorship; each team was sponsored by the manufacturer of the car it used, and two of the winners received money from equipment manufacturer Caterpillar.

courtesy @arstechnica.com