Monday, October 10, 2005

DARPA Grand Challenge

A post by Rob M on his blog reminded me of this event, which finished up yesterday. The DARPA Grand Challenge is basically rally racing for geeks. Imagine this, a 240km race course in a barron dessert location (Nevada's Mojave dessert, U.S.A) - basically, as far away from population centres as possible - where fully autonomous vehicles attempt to complete the course without collision with obstacles, or getting lost. This event was run for the first time in 2004, and no team got anywhere near finishing. It was quite a depressing snap shot of the state of the art for autonomous mobile robotics. Well, this year, I am most pleased to see that five teams (out of 23) managed to complete the course this year, with Stanford Universities entry, "Stanley", the quickest at 6 hours, 53 minutes (ave speed: 34 km/hour).

Despite being a researcher in the general field of autonomous robot navigation, I have only a limited interest in this event. This is not to say it isn't worth while and informative, (although DARPA's interest in it is hard to ignore), ultimately its an exercise in Engineering, and spending an incredible amount of money on sophisticated hardware and sensors to solve the problem. From what I understand, there is little or no use of vision systems in any of the top performing entrants (I might be wrong though). This is disappointing, but by no means surprising given many of the challenges to overcome can be achieved using things like laser range finders, GPS systems and other wizz bang devices. I tend to be more interested in achievements made with only a minimal number of sensors (ultimately, less hardware and more embedded intelligence is what interests me, but of course, I'm a computer scientist, not an electrical or mechanical engineer), that do not cost the equivalent of a small south Pacific island to design and build. Vision is good because its cheap (what's a good digital camera worth these days?) and incredibly versatile. If we could overcome these challenges using only vision, then much of the technology being developed would become far more accessible, and affordable.

Just as an aside...(from a comment I posted on Rob's blog) ...

.. autonomous vehicles, or more precisely, driver assistance systems are becoming big business for roboticists this days, particularly for robot vision researchers. While fully autonomous vehicles on our roads are still a long way off, vision sensors for obstacle avoidance, drive-by-wire, driver monitoring (such as tiredness through eye gaze tracking) and road sign detection are either already here, or just on the horizon. Its great to see this research finally finding a market for its practical use by everyday (though perhaps wealthy) users (rather than just on factory floors and in Toys'R'Us).

0 Comments:

Post a Comment

<< Home