Slashdot Mirror
DARPA's Autonomous Vehicle Challenge Too Popular?
Tim writes "Mobilerobotics.org has an editorial accompanying a copy of a letter to one of the teams entering the DARPA Grand Challenge 1 million dollar autonomous vehicle race, in which DARPA admits to underestimating the number of teams that can actually partipate in the actual race. They figure they've only got room for 20 teams, and more than 100 have applied. The writer of the editorial argues that if more than 20 teams can qualify safely and technically, DARPA should have to chose the 20 cheapest financed teams. What should DARPA do to sort out these problems?" CNET News has more on the high turn-out, while DARPA ponders its next step.
100 contestants, room for 20 on the course... run 5 heats! Top 4 from each go on to final heat of 20...
Don't blame me, I voted for Kodos
They need to have a pre-contest, something akin to Robot Wars. I mean, it is the Defense Department after all.
I Am My Own Worst Enemy
The bottom line of the whole thing is that people have invested time and money in the research and development behind their vehicles. What DARPA should do is run the course multiple times with the max # of participants, or an even division of participants, but run all of the contestant vehicles through, and time/rate them on their traversal of the course. Then pick the X top competitors and run them all through again to pick a winner. Or something. But you can't honestly expect people, who, on the word of DARPA, undertook to research and build something as difficult and complex as an autonomous vehicle, to just walk away because DARPA didn't consider limiting the number of entries before they announced the contest. Adapt the competition to suit the response..... and you'll be certain you didn't throw the best idea out arbitratily to cut down the field.
Comment removed based on user account deletion
If each team starts at a random (or sensibly positioned) part of the course, and has to navigate to a central point or points the vehicles would need to navigate similar but not identical terrain, and also have to deal with traffic that is coming at it from many directions rather than a convoy type traffic scenario with everyone starting from the same place.
Alternately, if the weather conditions in this part of the world are stable enough it should be possible to run the course over several weeks. The only problems that occur to me would be that evidence of previous vehicles would mean that the latter teams would have tracks as markers as to where others went. If the area is reasonably windy, or has lots of rain, these could be washed away but that is sheer speculation. Just my $0.02 worth
Nobody expects the Spanish Inquisition!!
They're supposed to be autonomous vehicles, right? If they can't keep out of each others way, they're not very autonomous, are they?
Can You Say Linux? I Knew That You Could.
Well, I think they should evaulate all the projects, the experience of the team members, cost-effectiveness and feasability of the project...
And then give the contract to Haliburton.
Just admitting that they underestimated the interest in this competition, and change it to make the rules harder? I doubt that the current requirements include everything that they would like an autonomous vehicle to accomplish.
I like the "tournament" ideas discussed so far, as DARPA should really test ALL the submissions. Find the best technology now, and further development WILL bring the cost down in the long run. Simply saying 'Oh, but this one is too expensive' has too much potential to eliminate superior technology.
There's no wrong way, to eat a Rhesus...
If I were a contestant, I'd go above and beyond the call of duty to win the contest. The rules of the competition say that the vehicles need to navigate a 250 mile long course without human steering. I'd say that all the other contestants are going to try to do just that, with various degrees of success.
But, I am gifted with the ability to see the forest for the trees. The people who are running the contest are with the Defense Department. Among other things, that department is responsible for prosecuting wars. And wars are just formal and legal ways to kill lots of people.
So, my strategy, in line with seeing the forest for the trees, would be to read between the lines. The rules talk about navigating a course, but why do they want to do that? TO KILL PEOPLE. The larger goal here is not to just navigate a course, but to win a war by killing lots of people.
My hypothetical entry would be a very large limousine, with a fully stocked bar, lots of hookers, a disco ball, and a bomb in the trunk. Everyone would take a look at my entry, and say "I gotta take a ride in that thing." The limousine would be very large, and could hold hundreds of people. When they are all inside, the autonomous function would take over. The contest does specify "autonomous" so the car would know what to do automatically. It would blow up, killing a lot of people, and hopefully anyone close by too.
With that kind of performance, I am quite sure that DARPA would be very impressed with my entry, and I would win.
If tits were wings it'd be flying around.
I am sure that a few short heat races would weed out 95% of the competitors.
love is just extroverted narcissism
Autonomous vehicles roaming the countryside, tracking down the stray humans who haven't been corraled into the pod camps so our bodies can be used as batteries. Like we haven't seen THAT before.
Hmm. An editorial in an amateur robotics site recommends using the cheapest projects. Who'll have the cheapest projects? Why, the amateurs who don't have to pay for (or, at least, account for) labor, project space, etc., of course!
DARPA is looking for people to push the envelope on autonomous vehicle research. However, this is also a very political project that involves a lot of cross-department cooperation. They don't want to have to talk to the press about how an out-of-control "giant robot" crashed into the home of the last colony of purple spotted pigmy desert lizards and exploded. That means, effectively, that talented amateurs with a go-cart and a spare PC are not welcome. They want people who either have a track record or who seem to really be on top of things. As a result, I fully expect them to reject most of the last-minute entrants, small teams, or teams with known problems (like "it don't work yet").
Forward, retransmit, or republish anything I say here. Just don't misquote me.
Seems to me like the costs involved in simply extending the contest by a day or two(or 5) to run longer, as well as conduct any trail repairs and such, are minimal compared to the potential rewards from much more competition.
However, I suspect that the "big boy" defense contractors will get first pickings even if they have shit for entries, and if the armed services -really- want to stack the deck, they'll pick the independent teams they think have the least chance to fill out the other 15 or so slots.
If you don't believe me, just look at some of the wonderful moves the armed services have made in the past when things were supposed to be open to fair bidding etc. Or, look at the current bids for Iraq stuff- one does wonder what sort of commission Cheney gets these days- oh wait, that would be his Haliburton retirement account...
Please help metamoderate.
We already have "car-sharing" programs, and this could take it a step farther.
Instead of driving the car to work and leaving it in a parking lot all day, the car could spend the day running errands for other people. Not everyone would need to buy a car and those who do could chose to lease them into this system.
Sure, Americans like the freedom of driving down the road wherever and whenever they want. We also don't like being pressed in with a bunch of other people. Given a choice between an hour alone in our cars or an hour stuffed in a train or bus, most Americans will choose the car.
In my personal case, the mass-transit option takes more than an hour to get to and from work. I can make the drive in 15 minutes. Working an hourly job, the hour of commuting time saved more than pays for my parking costs.
Instead there will be loads of wrecks from overworked programmers messing up a few lines of critical code. At that point terrorists will have no need to hijack an airplane, they'll just release virus "InsaneTruck.vbs", or "WackAPedestrian.asx" and every vehicle on the road becomes a remote weapon of terror.
Just taking 20-40 of the entrants sucks. You follow the rules, come up with a design that costs a million dollars, then they say "sorry, you're not the cheapest - bye bye"...
That's the wrong way to do it, anyway. You pick the winner(s) out of the best ones to do it successfully for the lowest cost, not the ones that only have a low up-front cost. If they don't work, you've binned 80% of your other candidates.
Also, you look for the lowest cost of building the finished unit, not the development costs put up front. Some teams may have had massive amounts of money put into them to guarantee a win, that doesn't imply the finished unit will be expensive to make.
DARPA should have to chose the 20 cheapest financed teams.
Whoa there, this is government. The 20 most expensive to make should be in the final competition.
In other news, I expect to open an autonomous vehicle dearlership soon. initial inventory is expected at about 80 vehicles.
what sig?
Reading between the lines of DARPA's letter...
"
Dear Mr. Insignificant
Thank you for participating, but since this dog and pony show was really only designed to allow the top 20 big contractors (like Raytheon, Boeing, Lockheed, Rockwell, Northrop, GE, GM, Ford, et al) get a chance at a lucrative contract, we must now admit that you don't have a chance in hell. As a matter of fact, we are so nervous that one of the other 80 participants may actually produce a better, more innovative vehicle, we are going to disqualify you all before you even get a chance. Sure, in the spirit of looking fair, we will come visit your site and see your vehicle. Don't get your hopes up. Do you really think we are going to give the vehicle you and your unemployed ex-aerospace friends built in a garage the same chance we are going to give the Boeing entry we get to see at a 40,000 sq ft plant with a clean room and a reception area filled with gnosh for us to eat? Hell, Lockheed is even providing champagne and a mariachi band! Thank you for playing, now go home and let the big boys play with their billion dollar contracts.
Sincerely,
DARPA (i.e. The Federal Appropriations Enabler)"
Who needs innovation when you can get free food and a goodie bag?
"Curiosity killed the cat, but for a while I was a suspect."- Steven Wright
This competition is incredibly difficult. Travellling 250 miles in 10 hours over desert terrain, on a course which in some places is intentionally too narrow for GPS navigation, is almost certainly beyond the limits of current robotic technology. Because of the slow speeds necessary on portions of the course, the robot must drive at over 60 MPH much of the time! It will undoubtedly be several years before any team passes the test (unless they loosen the rules).
Although there are 100+ teams registered (see the team list here), that doesn't mean much. There was no entry fee to apply! At this point all the teams have to have done is supply a technical paper with their ideas for how their robot could work. There's a huge difference between doing that and actually producing a multi hundred thousand dollar vehicle.
Undoubtedly, only a small fraction of these teams will have the budgets and resources to show up with a vehicle on March 13. I doubt there will be more than 10. And none of them will meet the standards necessary to win the contest. But most of them will be back next year, with a few new entrants, and after enough years of experience they will hopefully succeed.
But for now, this is all a mountain in a molehill. People are making a tempest out of a teapot. DARPA simply failed to explicitly include a phase to weed out those contestants who won't have a vehicle. Now they are fixing that. I doubt very much that the numbers will be an issue at all.
On top of that, your car can still fail, the electronic fuel injection could go screwy and you'd lose power, causing an accident. Heck, in event of a problem, the car could broadcast an alert via some redundant system that goes "HEY! I broke, and I'm right here! don't hit me!" If you built all the cars to be interoperable, you could do all sorts of nifty things, scheduling and traffic management to alleive congestion.
All I'm really trying to say, is that I'd trust windows more than some of the drivers on the road, and these would likely be very stable embedded systems, like the kind that run assembly lines, or your car, every day without a hitch.
Douglas P. Price
The system was simple. In pre-qualifying, competitors had to exceed some pre-set minimum standard. If you didn't make it, you didn't get into the regular qualifying session. The reason for calling it pre-qualifying is that, for the bulk of the season, it only applied to the slowest N driver/team pairs.
Qualifying then required all surviving competitors to get within N seconds of the fastest car. Anyone slower was disqualified.
Similar standards are used in the Olympics, where there are minimum standards set to even be able to get there, and (as there are more countries than track space) you then have to meet additional standards to actually start.
It seems that this would be the way for DARPA to go. Now, space and marshalling will place certain limits on how they can do this. You can't do your routine pre-qualifying, if there's nobody you can have marshalling the event, and nowhere to hold it.
Static testing would therefore seem a viable alternative. Put up a prefab shop, where you can statically test the autonomous vehicles. If certain standards aren't met, the vehicle is DNQed. Yeah, it's rough, as this wasn't the original spec, but they've gotta do something and this would work better than drawing lots.
Now, you're still likely to have more than 20 meet the minimum, unless the minimum is so high that you risk eliminating too many. The answer here would seem to be to have some kind of short sprint area. That shouldn't be too bad, now the numbers are down. First 20 across the line are the entrants.
Again, this is rough, but DARPA are badly outnumbered and they don't have a choice but to cut the numbers somehow. The only fair way is to have "mini contests" that are sufficiently limited in space/time that they can manage it with the resources they have.
And, again, competitors may bitch about being DNQed with the above, because it wasn't the original contest, but they'd bitch a hell of a lot more (and with far greater cause) if it came down to drawing straws.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)