Slashdot Mirror
Still More on the DARPA Grand Challenge
The SF Chronicle has an in-depth story on the DARPA Grand Challenge, with emphasis on the several teams from the San Francisco area. The three teams covered are using a pickup truck, a six-wheeled all-terrain vehicle, and a self-balancing motorcycle...
Drones don't have to face any obstacles, they can fly almost anywhere. The desert landscape, on the other hand, is more complicated.
It would be ideal, however one of the rules for the challenge is that no government funds may be used towards development. You can of course uses technology that was developed with government funding, but ONLY if it is commercially available. Unfortunately, the JPL vision code is not.
EEPD Profive 700 Mhz Pentium III PC -104 onboard computer running Real Time Linux
http://robots.mit.edu/projects/darpa/ (with videos)
CC.
TaijiQuan (Huang, 5 loosenings)
From the article:
DARPA won't disclose the exact route of the Grand Challenge until two hours before the race March 13; it has promised a rigorous route that will include rocks, gullies and streams.
Some of the world's best dirtbike riders wouldn't be able to easily cross stuff your average Land Rover or Land Cruiser would laugh at. I think the team with [what looks like] the 6-wheeled ATV stands the best chance, at least from a vehicle-choice perspective. Those things are amazing in terms of what they can cross- some of them even float and can ford -rivers- using the tires as paddlewheels.
Description of the pickup truck entry:Two tons of steel rolled forward and made a jerky left out of a parking lot in Morgan Hill. It gained speed and settled into a lane. It followed a curve to an intersection. It stopped. Then it turned right and continued down the road.
Probably stands a better chance(and has better fuel economy than the 6-wheeler- though a MUCH higher center of gravity), but taking a trip through suburbia hardly qualifies as suitable testing grounds for what DARPA has described...and depending on the truck, it might not stand up to the abuse. A jeep(or, a Land Cruiser, or a Land Rover) would have been a much better idea than a pickup truck, which really isn't designed for off-roading.
Even the guys who do insane things with their jeeps and whatnot come fully equipped. Air suspensions. Winches. Huge tanks of air or compressors to re-seat the giant tires(did I mention giant tires? :-)
I can also think of a lot better things to spend money on than that giant LCD display they put in the truck's passenger side; that thing has got to be what, 21"? The money would have been much better spent on the truck itself. It's all fun and games until that rock takes out your transfer case and your truck's transmission rips itself to pieces.
Please help metamoderate.
The Mars rover goes a few feet a day, tops. These things have got to go 40mph average. That's a mite different...
Since when has this country used intellectual elite as a pejorative term?
with the time limit of 10 hours across 200 miles, technology from Mars Rover might not be enough.
those rovers are travelling at max 2-inch per second, which gives the processor plenty of time to build a 3D model, analyse it and make a decision.
Just because DARPA is collaborating with NASA, don't get your hopes up if you're thinking about some 'geekcool' super-Star-Trek-beam-me-up-scotty rocket their buddy. DARPA is strictly defense, and anything they can get to the benefit of a defense project is worth gold.
If DARPA is doing something with NASA, it will likely use this for the killing fields nothing more nothing less.MoFscker
Its been done a long time ago but still has soo much to be desired to do it effectivly... Basically you still need a human operator for the equipment your pulling to keep it effecient.. and since your allready in there.. might as well drive to keep effeciency up as high as you can get...
Who needs WiFi when we can have Packet Over Sheep! http://datacomm.org/PoS-InternetDraft.txt
Yes.
Each vehicle is followed by a manned one. Specifically, one of the team members and a contest official.
The team member has a "big read button" - which is a mandatory safety device - that is the vehicle is in danger of or actually goes off course can be used to shut it down.
Then you can get disqualified for it, upon the disgression of the cheif judge.
Check out the latest copy of the rules
=Smidge=
200 miles in 10 hours equals an average of 20mph.
A blog about stuff.
Some idiot already tried entering a monkey; the rules now say no living being can drive.
Sadly, we (Team Overbot) aren't going to be ready in time. We lost five members in January. Two got better jobs, and two were Stanford students who needed to get their grades back up. This left us with too few people to finish in time. We have all the hardware, and most of the software, Most of it is working, but it hasn't been integrated and tested. We'll finish the vehicle, and we'll have some public demos at some point, even if we're not at the Grand Challenge.
It's up to DARPA whether anyone wins this year. They're going to provide 5000 GPS waypoints, and if you can drive the route described by connecting the dots, somebody will probably win. If the vehicle has to find its own gully crossing, it's unlikely that anyone will win, unless somebody figured out, by hand, in advance, where the crossing is. It's all up to DARPA. As one of the DARPA people put it, "This is turning into a breadcrumb following exercise". If somebody wins by connecting the dots, this whole thing was a waste of time.
Several teams are using aerial photographs and manual planning. The general route leaked weeks ago, and it's since been oveflown by Airborne 1 in San Diego. High-resolution photos and depth maps from LIDAR scans have been obtained. Still, you won't see a fence in those depth maps. The emphasis on preplanning surprised us. The whole point of the Grand Challenge was originally that preplanning was made impossible by the large area to be covered and the release of the waypoints only two hours before the race. That all changed when the route leaked.
Nobody seems to be deploying anything new in the sensor area. Everybody with a laser rangefinder that we know of is using an off-the-shelf line scanner. Nobody has a true 3D scanner, although several teams have line scanners on tilt heads. It's quite possible to build a true 3D LIDAR depth measurement system. But it's hard to make money doing it, as the five companies that exited the field learned the hard way.
We hear talk of new vision algorithms, but no details yet. Stereo vision doesn't work well on dirt or sand; there aren't enough edges for the stereo algorithms to register the images properly. Optical flow doesn't work well for the same reason. If somebody can do good stereo from motion in this demanding environment, that will be an achievement.
Still, the Grand Challenge has done quite a bit to get autonomous vehicle work moving again. Just getting CMU off the dime (DARPA's real intent, we hear) was worth the whole effort.
If DARPA does this every year for the next decade, with a tougher course every time somebody wins, we will have battlefield robotics that works within ten years.
I'm on the Caltech team, and I can assure you, we are not using the JPL code, although we wish we could.
If I'd paid more attention to what Berkeley was doing with ACK delays, TCP would work better in that area today. Both algorithms went in around the same time, and they don't play well together.
A pickup truck, because the third team actually uses a 4WD. DAPRA said that the race can be traversed in a pickup (with a human driver), but the article doesn't say any team will actually use a pickup.
Future Wiki -- If you don't think about the future, you cannot have one.