Slashdot Mirror
CMU First To Qualify For DARPA Grand Challenge
Anonymous Coward writes "As of 18:00 March 9th, Carnegie Mellon's Red Team is the only entry to successfully complete DARPA's Grand Challenge Qualification Inspection and Demonstration (QID) before the main event on March 13th. The NY Times has this article detailing this first step towards winning the Grand Challenge."
I can't be the only one who questions motives when the $1M prize is being sought after by a team with more than $2M already invested. What is the eventual payoff?
Linux: Free if your time is worthless.
The real prizes:
the knowledge gained throughout the project
getting one's name published for taking an active role in the project (which can lead to further opportunities)
the overall experience, i.e. 'Hey, I did that"
The pursuit of intellectual challenge is not about money...
I want to drag this out as long as possible. Bring me my protractor.
... doesn't this basically lead directly to the US military dropping off Robo-Tanks in foreign countries as they please? We know that a steady diet of wars figure heavily in the plan for the forseeable future. The Robo-Tank cuts down on friendly casualties, thus making conflicts more palatable to the public.
Now I find this as cool as anyone else, from a technological standpoint. And it definitely has civilian applicability. But let's face it, this contest isn't about finding cheaper ways to haul cargo or reach remote locations.
They're not all that functional either - unless a mountain should suddenly spring up on the way to the kids' soccer practice. Obviously a few people have a need for those sorts of vehicles, but I question the volume of them I see on the road.
Ita erat quando hic adveni.
The hummer gets 9mpg. You can buy cars in the US that get 70mpg (diesel golf, Prius), but not many people want them. It's not that we are less advanced, it's that we are greedy and evil.
If the military invades with these, they aren't just going to tell it to go somewhere and kill someone, they are going to give the machines very specific directions. If they dont have a map...they could probably get one in a few hours anyway, so I don't think that will be a big issue.
From what I understand, the "loophole" allows them to use 1m satellite imagery of the route and, in two hours, plan something for the robot to follow. The robot still has to see its way when it's going, to avoid ditches and rocks and other things--it needs to do "local" route planning at 35mph. Even if the route were totally pre-programmed, the problem of following that route would still be pretty hard over hundreds of miles. You can't just do "dead reckoning."
What about if they invade somewhere they don't have good maps of? Somewhere with a dynamic landscape (desert, rocks etc)?
This is in the desert, and they're doing it with only satellite imagery.
There's a huge amount of mechanical and software engineering in this thing. I think that someone must have exaggerated this "loophole" to you, because it is far from making the project easy (as far as I know, it doesn't help them in the quals at all). The robot is impressive!
I'm not sure that it is a loophole. It's not as challenging as doing it the other way, but let's face it, this is being done for the military, and you're extremely naive to think that the military doesn't have precise topography maps of the entire world, or that they can't obtain such maps in short order. Remember, a key component to cruise missile technology is topography. Remember in GWI, the cruise missiles took hours and hours to program before launch. Now, they can be reprogrammed in minutes.
So, the current method used by the Red Team may likely be how the military would implement it in the first generations of this type of equipment. Plan the best route manually and then tell the automaton what track it should generally take and let it navigate the minor obsticals.
Disclaimer: I'm not involved in DARPA in any manner.
Linux - Because Mommy taught me to Share.
It doesn't have to think about navigating - they're telling it how to do that. It has to only deal with getting round obstacles in its path. They're removing 1/2 of the problem so they can put their effort behind the other half, which the other teams aren't doing. It just smacks of unfairness, that's all.
with this is simply how cheap the US military is getting away with this. Instead of setting forth a proposal, taking bids, working in tandem with one of the big development houses, they offer up a rediculously small prize. If they had gone through someone like Lockheed Martin, they project would have easily cost them into the 100 million dollar range. Oh well. Hopefully the military will get what they paid for.
Yup, those are some pretty cool prizes. But we gotta remember other prizes, like bragging rights. I'm already sending this article all over to my fellow CMU almuni friends, to other non-geek friends, etc. All this, and I have nothing to do with robotics and graduated almost ten years ago. Woohoo! I love bragging rights...
"He uses statistics as a drunken man uses lampposts...for support rather than illumination." - Andrew Lang
The point of the Darpa project is to advance technology for driverless military vehicles, primarily for convoy work. To my mind, creating a computer system to quickly plan out routes based on intelligence is an important part of a practical solution.
Not only does it more accurately reflect the technology's intended use-case in the military field (convoy operators would lilely be given a general route a couple hours before a mission, instead of simply told, 'get it to this point and leave right now') but it also means that more of the technology is outside the vehicle.
A cost-effective solution would need to have as cheap a vehicle as possible. While a fully autonomous system might be nice for a science fiction 'technology run amok' film, in reality it's more effective to have sparse mobile systems with an ops center capable of planning routes for several vehicles.
It also costs less when one goes 'wheels up' or is captured by the enemy.
Kevin Fox
I went by the event yesterday as a spectator and got to see Red Team do their run. Of the 23 teams who made it this far, they're the only one that has completed the qualification course so far. People complain that they have a more accurate map and that they're not doing real AI, but based on their performance on this surprise course, they have a real obstacle avoidance system.
In one section there was a minivan parked in the center of the GPS path. Of the eight vehicles I saw run, only three made it past the car. Three hit it, and the rest failed before making it that far.
It seemed that the biggest problems teams had were getting GPS right. Several drifted off course or turned the wrong way, going off course. One got the next GPS coord inside of its turning radius so it kept circling a spot until they turned it off.
Lots of great designs though, and some really impressive engineering.
Caltech manages JPL, the NASA lab that developed the rovers and their associated software.
Originally, the Caltec team was using rover software. However, when DARPA changed contest rules a couple months ago, it went back on its earlier ruling and said that Caltech was no longer allowed to use the rover software because that software was not commercially available.
This led to Caltech redoing much of the work on their vision software. They are now using the modified version of a commercial vision package.
I personally think that DARPA could have done better by asking JPL to make the software available to ALL teams instead of taking it away from Caltech.