Re:any robotics experts? (Score 1)

I am the author of the article on the Grand Challenge in the March issue of Scientific American. While it is true that much of the difficulty in completing this race is due to the AI required, a big part of what makes the AI hard is the speed at which the vehicles must process the data streaming in from their many sensors. How many gigaflops can you get out of two four-processor Itanium2 machines (running code compiled by hand-optimized compilers) and three dual-Xeon machines? Not enough, is the short answer. There are good reasons that autonomous off-road vehicles built by seasoned and very well-funded robotics teams for DARPA's and the Army's and the Navy's many UGV programs all move slower than the 20-30mph needed to win the Grand Challenge race, and insufficient sensor data processing speed is one of those reasons. Some teams will try to get by using just one type of sensor, but that is unlikely to work. Dust clouds, inevitable on desert roads, can make a LIDAR scanner useless. Direct sunlight dazzles them, too--and the racers will start the day heading into the sun. Shadows, dust and the generally low contrast terrain in the desert can confuse stereo vision systems. Radar updates relatively slowly and offers relatively low resolution.

So speed of operation is key, and the need for ruggedness follows from that: the more stable and well-shielded your sensors are, the less work the computers have to do to clean up the data before they can use it to make sense of the world looming before the racer. AFAIK, only one team (the Red Team) has built a high-speed gimbal in order to stabilize some of their sensors in all three axes against the jolts and jitter of a high-speed desert traverse. But will the gimbal itself be able to withstand the journey? Component endurance will be critical, and the race will in a very literal sense be a shake-out. Any team that has less than a month of rigorous desert testing under its belt by race day will have essentially no idea how well its racer will endure.

Top-of-the-line components are sometimes more fragile and often more buggy than standard COTS parts. So a big budget doesn't guarantee success by any means, and in truth not once of the competing teams has a big budget by DARPA standards. (The $5 million budget often attributed to the Red Team in lackadaisacal news accounts is spurious; that figure was tossed out early on as what would be required to field two vehicles and have spares for every critical part. Needless to say, the project quickly lowered its amibitions considerably.)

According to the rules, DARPA has no claim to the intellectual property developed by the GC teams. Based on my conversation with DARPA director Tony Tether, my hunch is that what DARPA gets out of this event is primarily PR and exposure to a part of the "fringe" engineering community that would not normally think of taking its out-of-the-box ideas to the DOD. Frankly it seems very unlikely that any GC vehicles we'll see this year could serve as the starting point for a standard DARPA contract program. But they will be exploring a much larger design space than DARPA could explore through its normal programs.

Nor is DARPA the only beneficiary. One very competent team pulled out early on because an Asian firm swooped in and paid it eight figures for its design. So there is one vehicle that is more likely to end up in the arsenal of South Korea, say, than in any Pentagon warehouse. Not exactly what DARPA expected, I imagine.