On the Way To Fully Autonomous Ground Vehicles

Well, thanks to a new landmark moment in robotics. At the recent DARPA Grand Challenge competition sponsored by (drum roll please) DARPA (Defense Advanced Research Projects Agency), the Stanford team's robot named Stanley (actually a computerized Volkswagon Tourage) successfully navigated a 132-mile course through the Nevada mountains in under 10 hours with no aid. It faced off against roads involving cliffs on the side, hills, rocks, etc...at one point, it was on a road in which on one side stood a mountain, on the other, a 100+ foot drop. The VW had only a few feet of maneuvering room (one of the other bigger 'bots, a big Terramax truck, had only a few inches).

The VW managed to navigate the course, followed closely by the Carnegie-Mellon team. The Stanford team took home $2 million dollars.

This was a landmark moment for autonomous robots because in the previous year's competition, none of the robots finished the course. Many of the DARPA folk were not enthusiastic about this competition. For a vehicle to make it halfway through the course would be great. For one to finish it would be a miracle, it was said. Well actually about 3 others finished in under 10 hours as well, but the Stanford VW finished first.

This shows a great improvement over the technology used in the previous competition. Previously, many had thought that autonomous vehicles were just too difficult to develop with the current technology, whereas now they know it is possible.

The U.S. military has plans to make 1/3 of all its vehicles autonomous by 2015. Autonomous aircraft have existed for some years now and autonomous missiles have been around for about two decades, but autonomous ground vehicles are a lot more difficult to develop.

Carnegie-Mellon is the top robotics research institute in the U.S., but Stanford is one of the top in artificial intelligence research. They said one of the things they did was to have their VW drive along and identify all sorts of different obstacles. As far as it knew, anything from a mountain to a butterfly was an obstacle. They then had a human driver drive it along and they programmed it so that whatever of the obstacles the human completely ignored and drove through, the robot was to "learn" to ignore those same objects. Thus, the computer reduced its obstacle count from around 1800 to 12.

This is pretty cool. I personally don't like the idea of robo-vehicles yet, but I am sure the military just means vehicles where a human driver isn't essential.

LINKS:

Blue Team
www.ghostriderrobot.com/index.php?id=press

Carnegie Mellon Red Team
www.redteamracing.org/index.cfm?method=members.view&memberID=59

DARPA
www.DARPA.gov/grandchallenge

DARPA Grand Challenge
www.grandchallenge.org

Palos Verde's High School Doom Buggy
.pvrw.com...

Stanford Racing Team
www.stanfordracing.com

Terrmax
www.terramax.com

Above websites obtained from ROBOT magazine, website at www.botmag.com

[edit on 26-2-2006 by WheelsRCool]