3 Bots Win Pentagon's Robotic Rally

An anonymous reader writes "We've got a winner in the Pentagon's $3.5 million all-robot street rally, the Urban Challenge. Three, actually. Wired reports that 'bots from Stanford, Virginia Tech, and Carnegie Mellon all completed the course within the six-hour time limit. The robo-cars had to complete different missions taking varying times, so the flesh-and-blood judges will take a day to figure out who takes home first prize."

Comments

go go power-rangers

[LiquidCoooled]

They merged up into a superbot and won the contest by destroying everything around them.

Actually, 4 potential

[ZaMoose]

Ben Franklin Racing (a collaboration between UPenn, Lehigh and Lockheed Martin) also finished within the 6 hour time limit.

The judging will certainly be interesting.

Re:Actually, 4 potential

[vertinox]

Yeah, but I think Ben was penalized for its initial delay at an intersection for too long at the first mission. After they rebooted it seemed to be ok and actually did pretty good getting around the intersections it had problems with. When Honeywell was taken out of the race, Ben was waiting behind it at the stop sign until they pulled Honeywell out but I don't think they penalize for that.

It is something to note that the two teams that finished but finished last were MIT and Cornell which had a collision with each other somewhere around Mission 2. But they both finished which is pretty awesome considering what it takes to run this corse.

Re:Actually, 4 potential

[samkass]

That's unclear from the article... the article says the race started at 8am and Ben came in at 2:50pm. Do you have an alternate link to different results?

Re:Actually, 4 potential

[ZaMoose]

BFR's blog actually makes that claim (sub-6 hours).

I have to question the constest

[Anonymous Coward]

I will admit that I haven't read up on the exact rules , but I find the name Urban Challenge to be a bit misleading. From what I have seen the environments are very sterile compared to real life urban environments , yes the name gives the impression that robots can now drive in a ciry like New York. This reminds of the 60s I think it was when computer scientist claimed that because a robot could restaple boxes , we will have androids in 20 years. Then it became clear that the algorithms didn't scale well with the complexity of the environment ( to put it nicely) and Artificial Intelligence became a somewhat disappointing field for the general public atleast.
    All I am saying is that we and the tech journals should be carefull with exciting names like "Urban Challenge" or "60miles through urban landscape".
    Other then that , congratulations to the teams , I didn't expect such good results.

Re:I have to question the constest

[RespekMyAthorati]

All I am saying is that we and the tech journals should be carefull with exciting names like "Urban Challenge" or "60miles through urban landscape"

It's all PR.
But, PR is vital to getting $$ from a basically anti-scientific federal government.

We WILL have androids in 20 years

[mangu]

Then it became clear that the algorithms didn't scale well with the complexity of the environment (to put it nicely) and Artificial Intelligence became a somewhat disappointing field for the general public at least.

That's the problem with hype. They have cried "wolf" too many times. It was the same thing at the end of the 19th century, when people were researching flight. Steam engines were too heavy for their power, airplanes had to wait until engines became powerful enough. There were many people, among them some respectable scientists, that wrote articles "proving" that heavier than air flight was impossible.

At this point, computers are too expensive and consume too much power to be practical for anything that involves "human-like" intelligence. But we are making progress, at least we do have unbeatable chess-playing computers, a feat that not so long ago many people considered impossible. Of course, computers do not follow the exact path of reasoning that humans do when playing chess, but they are unbeatable anyhow. Airplanes do not flap wings either, but they fly faster and higher than any bird.

Unless Moore's law ceases to function, we can expect desktop computers with a complexity comparable to that of a human brain in twenty years or so. Given the hardware, it's only reasonable that someone will invent a way to make a computer emulate a human brain in its full power, just like people invented machines capable of flying when they got engines with enough power.

Re:We WILL have androids in 20 years

[Anonymous Coward]

So I have a choice between a robot driver that could crash my car or have an affair with my wife?
I know where my money's going.

Re:We WILL have androids in 20 years

[hyades1]

I've heard estimates of both one and three terabytes as adequate storage to accurately reproduce a neuron-by-neuron reconstruction of a human brain. Assuming they figure out how we assemble and integrate everything to produce sentience, 20 years might be longer than we need.

Remember how the Luddites used to sneer that a computer the size of Manhattan couldn't model the behaviour of a cockroach? Then somebody figured out that about 6 basic commands would do the trick?

Re:We WILL have androids in 20 years

[HUKI365]

Yeah, but we have been saying this since the 1980s! We are approaching 2010 - and have done little that science would've thought would happen. No civillian space flight, no humans on planets, no artificial intelligence, no cure to AIDs, cancer, asthma or the common cold. No voice recognition or text-to-speech worth two hoots. Only now are we moving close to reasonable touch displays and miniturised memory!

Re:We WILL have androids in 20 years

[vertinox]

No civillian space flight, no humans on planets, no artificial intelligence, no cure to AIDs, cancer, asthma or the common cold.

Actually, I don't think its not the fault of emerging technologies but rather emerging technologies must scale to the following questions:

1. Is it profitable?
2. Is is mass producible?
3. Does it follow a decentralized free market model rather than a centralized regulated model?

No one predicted cell phones and the internet in the 80s as they are now, yet if someone told me in 1989 about Youtube on my hand held device I would have scoffed at it being too Star Trek like. Yet today we have such technology.

The reason we don't have AI, Civilian Space Travel, and Flying Cars is because they meet none of the 3 criteria I mentioned. AI today would require a computer that costs billions of dollars to build, a civilian space program would cost billions to build, and the flying car industry would be too regulated and dangerous to even consider marketing to people.

Which is why cell phones and internet caught on because those things are quite decentralized in how they work (Yeah I know the cell phone companies are monopolies but you can sell someone a cell phone and it doesn't cause any problem with the rest of the system etc etc)

Re:We WILL have androids in 20 years

[LS]

I hear what you are saying - I also believe that anything is possible given enough time and hard work. Yet I think you are VASTLY underestimating the task of creating a human-like intelligence. Faster and more powerful != More Intelligent. Flight and chess are child's play compared to the human mind. It's also a false assumption to believe that a Turing architecture machine will be able to simulate the human brain with whatever specious equivalence used to compare human and computer processing power. The brain is NOT a computer. Computers themselves are simple expressions of a mere slice of how we understand our own mental processes to work. Do you know anyone who understands how any mind works, let alone their own, whether they be computer scientists, psychologists, cognitive scientists, or neuro-biologists? To put it simply, in order to expect a human-like intelligence in 20 years requires two things we do not yet have: An understanding of human intelligence, and a hardware architecture that is able to implement it.

LS

Re:We WILL have androids in 20 years

[gaelfx]

A.I. maybe in 20 years, but androids? Just because we have intelligence does not mean we understand how to use it. Artificial Intelligence might work on a computer, and no matter how portable that computer is, it doesn't mean we'll be able to transport it intelligently or even very dynamically, regardless of cost. But then again, they will invent a time machine, ensuring it's use for military and thereby make it the single greatest threat to the existence of man. We do need something new to hunt :D

Re:We WILL have androids in 20 years

[Orange Crush]

Building a human-form robot is easy enough. Refine the basic mechanical parts, get a better battery, and work on a better balance system and you're all set. Controlling it is a bit trickier without an AI, but I think we could feasibly build a robo-maid in the next 20 years. Maybe not to the point of affordable mass production. But certainly clever enough to do the laundry, wash the dishes, and make the occasional grilled cheese sandwitch for researchers.

Re:We WILL have androids in 20 years

[samkass]

No one predicted ... the internet in the 80s as they are now, yet if someone told me in 1989 about Youtube on my hand held device I would have scoffed at it being too Star Trek like.

Not "no one"... at the risk of bringing up stupid falsehoods about who did or didn't claim to have invented what, here's part of the text of a speech Al Gore gave to congress in 1986:

Mr. President, it gives me great pleasure to support the proposed National Science Foundation Authorization Act.

[...]

Both of these amendments seek new information on critical problems of today. The Computer Network Study Act is designed to answer critical questions on the needs of computer telecommunications systems over the next 15 years. For example, what are the future requirements for computers in terms of quantity and quality of data transmission, data security, and software compatibility? What equipment must be developed to take advantage of the high transmission rates offered by fiber optic systems?

Both systems designed to handle the special needs of supercomputers and systems designed to meet the needs of smaller research computers will be evaluated. The emphasis is on research computers, but the users of all computers will benefit from this study. Today, we can bank by computer, shop by computer, and send letters by computer. Only a few companies and individuals use these services, but the number is growing and existing capabilities are limited.

In order to cope with the explosion of computer use in the country, we must look to new ways to advance the state-of-the-art in telecommunications-- new ways to increase the speed and quality of the data transmission. Without these improvements, the telecommunication networks face data bottlenecks like those we face every day on our crowded highways.

The private sector is already aware of the need to evaluate and adopt new technologies. One promising technology is the development of fiver optic systems for voice and data transmission. Eventually we will see a system of fiber optic systems being installed nationwide.

America's highways transport people and materials across the country. Federal freeways connect with state highways which connect in turn with county roads and city streets. To transport data and ideas, we will need a telecommunications highway connecting users coast to coast, state to state, city to city. The study required in this amendment will identify the problems and opportunities the nation will face in establishing that highway.

[...]

Re:We WILL have androids in 20 years

[inca34]

This has little to do with Moore's and a lot to do with the fact that sensors do not follow Moore's law. We were using the same sensor technology as was available 15 years ago with marginal to no improvement on quality or capability.

The software side of this DARPA Urban Challenge should consist of no more than an enormous, but straightforward, state machine that contains all the logic for traffic decisions. Plug that into a simulator and you've got the main software part done.

The problem everyone had out in the field for the qualifiers (I was on one of the teams) was perception. How do you know what you see is an obstacle? And how do you deal with false positives, and more importantly, false negatives? Some people believe in cross referencing sensor data, which is called sensor fusion. It is difficult, to say the least, to characterize every possible obstacle that ought to be considered a true obstacle if it lies on your vehicles path, let alone have a 10^-6 failure rate for improper detection.

The highway lane following has been solved since the 70s, check out R.E. Fentons work on Automated Highways in Transportation Science (1970). We had some "recent" developments in the early 1990s where we got some autonomous vehicles to do the autobahn at 100mph with more modern sensors and vehicles, but really didn't improve that much because the sensors aren't there yet.

Your sensor choice goes something like this:
$75k for a Velodyne 3D laser system
$5k for the SICK 2D (planar) lasers
~$25k for stereo vision cameras (per set)
~$1k for radar
$75K for the Applanix integrated GPS and IMU

The Velodyne is a spinning set of 64 lasers, with 64 photodiodes. Each manually placed so that the photodiodes are aimed precisely where the lasers are pointed. The entire head of the unit spins at ~2Hz and generates 1 million points per second. Most of the teams that bought one mounted it on top of their vehicle. This sensor is great if you have infinite processing power available to crunch the data and turn it into cost maps. It however has some serious problems: it's very expensive, it's not mass manufacturable, the point data for a rock and a shrub are indistinguishable (a weakness of all lasers), some obstacles we're interested in absorb laser or reflect it away from the photodiodes, it has too much information, and it has moving parts.

The SICK 2D planar lasers have more or less the same problems, except there's less data to crunch of course. These lasers also have moving parts internally, which spin a mirror at maybe ~20Hz to get distance data over a 2D plane. Same issues as the Velodyne, except it's manufacturable (has been for 15 years now).

Stereo vision is really hard to do right. When you have roughly a year to develop the platform and the algorithms, I don't expect much, and I didn't see much. This may be the answer in the future for passive detection, but I don't see it working at the moment.

RADAR is the right sensor for this type of work. It gives you distance and speed. If you're clever it also gives you the "cost" of a particular object. Radar is how you can tell the difference between a shrub and rock, or a car and a plastic fence. The real cost in the RADAR is not the sensor, but the $100k guy who knows RADAR well enough to set it up right and get good data out of it.

The Applanix GPS and IMU with 200k RPM laser gyros are not manufacturable and not practical for autonomous vehicles because of the cost. Perhaps the MEMs solutions will catch up and make IMUs cheaper, but in the mean time we're stuck with these systems if you care about your position.

That's my take on it. Improve the sensors and we'll get autonomous vehicles. Buy another Cray, strap on a generator and a multi-ton air conditioner is not the solution. We need to reliably and cheaply generate cost maps that are relevant to the vehicle that's being automated. Once that's been done reliably, we will have autonomous vehicles. Cheers.

Re:We WILL have androids in 20 years

[ucblockhead]

Lots of people predicted cell phones. In Heinlein's futurist essay written in 1950, he predicted in 2000 that everyone would have a wireless phone you could put in a pocket. He revisited this essay a couple times and in the last revisitation in 1980 he referred to the wireless phone prediction as "obviously correct".

Re:We WILL have androids in 20 years

[Smallpond]

Dick Tracy had a 2-way wrist radio in the 40's, which actually looks more convenient than the handheld units of today.

Re:We WILL have androids in 20 years

[jpfed]

Faster and more powerful != More Intelligent. Sorry to nitpick, but yes, it does. Intelligence as a function of speed, power, and strategies remains monotonically increasing with speed and monotonically increasing with power, up to the bounds of the complexity of the problem domain (cf Go and Tic-Tac-Toe). It just so happens that there will be diminishing returns on existing strategies, and finding new strategies will at some point be more cost-effective than making things faster or more powerful.

To put it simply, in order to expect a human-like intelligence in 20 years requires two things we do not yet have: An understanding of human intelligence, and a hardware architecture that is able to implement it. For some definitions of "human-like", sure. But a single architecture capable of creativity in many domains, quick reaction to unforeseen circumstances, and pragmatic reasoning and problem solving doesn't have to do it the same way we do it. As I type this, I realize that people will always expect more; if a machine were to do everything I just listed, then they'd want the machine to have empathy, be able to pick up cultures through immersion, pick up accents, and appreciate humor, etc... These things vary in how central they are to the diffuse definition of "intelligence", but as long as it serves to differentiate humans from computers, raising such objections will make people more comfortable.

Re:We WILL have androids in 20 years

[zero_offset]

"as they are now" is an important qualifier in the grandparent post. Gore's speech spoke to many of the technical requirements, but in terms of the way the 'net is actually used today, and it's extremely broad importance simply wasn't on anyone's radar in those days. I started using the 'net in the 88-89 time frame and e-mail was about as exciting as it got at the time. There was speculation in places like usenet discussions, but the earlier poster is correct -- nobody predicted the net as it is now.

Re:We WILL have androids in 20 years

[cashman73]

Big deal. I'm still waiting for Agent Maxwell Smart's Shoe Phone,... ;-)

Re:We WILL have androids in 20 years

[mangu]

Do you know anyone who understands how any mind works, let alone their own, whether they be computer scientists, psychologists, cognitive scientists, or neuro-biologists?

No, I don't. Yet it's funny what so many people say about Deep Blue: "that's not the way humans reason about chess". Huh? If we have no idea how the human mind works, how can we be so sure that deep inside a human chessmaster's subconscious mind there isn't a search engine looking over all possible game positions?

We do have a rather good idea of how a brain works, there are many lab experiments simulating neural networks in simpler animals, such as snails and cockroaches. What is lacking is a good model of how higher reasoning works, processes such as intuition and consciousness are still a mystery. But that's because we do not have yet the tools for that. We do not have computers capable of simulating a neural network with billions of neurons. Before the microscope was invented, scientists didn't know anything about the structure of cells either, they had no tools for the job.

Of course, it's premature to say that we will discover everything about the structure of the mind when we get the needed hardware, but given the history of the evolution of science in the last five centuries, I'm willing to bet that it won't take long.

Re:We WILL have androids in 20 years

[Sinical]

The sensors are coming. One promising sensor is "flash LIDAR" (or LADAR, "light" vs "laser"). Here's a mention from Google's first result (for a Grand Challenge thing in 2005):

True 3D solid-state flash LIDAR devices exist. We've visited Advanced Scientific
Concepts in Santa Barbara, CA, and have seen an eye-safe 128 x 128 pixel solid state
flash 3D LIDAR suitable for outdoor work in operation on an optical bench. The device
consists of two custom chips bonded back to back using ball grid array techniques. The
front chip contains the array of detectors, and the rear chip contains the counters, timers,
and interface logic. The detector chip typically uses indium arsenide technology. Some
versions are front-ended by a photomultiplier cathode, like a night vision device. (The
photomultiplier effect is at the atomic level, and has no integration delay, so it can be
used to front-end a LIDAR detector.) The two-chip approach is a convenience for
prototyping; a volume production unit would probably be a single chip.

from (here.

There are pictures in this one.

The military really wants them. Non-flash LIDAR is being integrated in systems like RAMICS (which uses LIDAR to look for mines underwater (assuming it works)) plus a number of missiles. One example here, the abortive attempt to put a LADAR in a ~7" (180cm) missile seeker with the Loiter Attack Missile (LAM). Lockheed, however, sucks, and couldn't pull it together. I've seen imagery from a competitor's seeker. Not great, but getting there. I think there are efforts underway in larger missiles like Tomahawk and Maverick where there's more power available.

Besides LADAR, there's IR sensors. I mean, Cadillacs and Hummers and stuff use Raytheon's uncooled array: as did that guy who just did the cross-country Cannonball-type sprint in his M5. I dunno the resolution of that one, but there *are* uncooled Focal Plane Arrays out there that are pretty damn big. No 3D info, but some all weather-capability. Presumably pavement looks different from non-pavement at night.

There's millimeter wave radar: ~100-300GHz. There's a *lot* of military interest in this since it gives good visibility in the presence of aerosols and smoke and whatnot (where LIDAR and IR may not work). Not so great in the rain, depending.

There are sensors. They may not be available to universities yet (sadly: I'm sure they are better at finding ingenious uses than big military contractors (I know)), but hopefully within the next 5-10 years. Or maybe we'll get awesome(r) at SAR (Sandia Labs has some great pictures for miniSAR: 4" resolution). I dunno if SAR is practical for ground-based navigation.

And I've used MEMs nav systems: GPS and IMU in a package about the size of a paperback book. There're smaller systems, too: for precision artillery shells (and so they can withstand accelerations of ~15k gees).

I hope that this Challenge will spur development in these areas as teams search for an edge in sensor technology.

Re:We WILL have androids in 20 years

[Anonymous Coward]

It's also a false assumption to believe that a Turing architecture machine will be able to simulate the human brain with whatever specious equivalence used to compare human and computer processing power.

Given a fast enough computer, you can perform any computable process. This is the central tenet of the Church-Turing thesis.

Computable processes includes human intelligence--blobs of biological matter have no special monopoly on intelligence. We may not know how to write a program that exhibits human intelligence, but we know a Turing machine can do it, or at least mimic human intelligence to the point that it'll pass the Turing test.

I'm not so arrogant to believe that it'll necessarily happen as soon as we get sufficient processing power, or even a long time afterwards, but it's firmly in the realm of the possible. After all, we already know that the right collection of molecules and electrical charges can pull off human intelligence. Thinking the brain is special in some way is the same sort of sentimental, mystical thinking that put the Earth at the center of the universe for 4000 years. It's not helpful.

Re:We WILL have androids in 20 years

[Anonymous Coward]

Fancy sensor technology is well and good and all, and I'm sure it'd solve lots of problems if you could create a perfect model of your environment instantaneously. (Though still not a trivial one to solve.)

However, while I'm sure these sorts of approaches are fine for producing a practical vehicle in a, "If we need, within a year, to build a vehicle that can do this, how do we go about it?" sort of way, I thought the whole point of the DARPA Challenges was to encourage thinking outside of the box?

We know human beings can drive in urban environments without immensely sophisticated radar/laser platforms, by replacing sophistication in sensors with massive sophistication in computation (stereo vision, as you mentioned). I think what we were all hoping would come out of these contests would be a relatively cheap hardware solution with some really innovative software, but I suppose that was never realistic.

Still, there's no reason to pooh-pooh possible advances in the software, perhaps combined with clever hardware design. For example, swarms might be a great way to pull this off, or any approach which is reasonably robust to failures, rather than being perfect but brittle. The time limit is an important constraint, but if you had unlimited time, you could probably design a pretty good machine that "felt" out its environment with physical probes for confirmation, rather than trying to scan it remotely X times a second. (Something ants do, in fact.)

I think this says more about the state of the modern AI field than anything about sensors vs. software. Modern AI is misnamed, really; as someone who was working on his doctorate in the field once told me, it's more about heuristics than intelligence. You task a roboticist with solving this problem, and they're going to go about it in a way that focuses on handling a million different subcases (that big, fancy state machine you mentioned), rather than trying to come up with anything really innovative. (Not that there's anything wrong with that; you can't turn on genius like water from a tap, and at the end of the day, a solid engineering approach will generally win over waiting for divine inspiration.)

I'd also wonder whether $75K is really "unaffordable." DARPA presumably wants these things to put on military vehicles that may cost several times as much. Of course, we'd like the technology to eventually trickle down to the civilian sector, but presumably, if it's possible at all, refinement over the time is what will get us there, not finding a sensor that does the job right now.

Re:We WILL have androids in 20 years

[Petersson]

Computers themselves are simple expressions of a mere slice of how we understand our own mental processes to work.

Computers are just interrupt-driven number crunchers. That's what they do. They do it very quickly and very effectively.

The letters I see on my screen when I'm writing this text are no real letters, it's just lot of color dots and some digital representation in computer's memory. There's no spoon.

Computers are similar to brain as a steam locomotive is similar to a horse (sorry, I just hate car analogies).

Re:We WILL have androids in 20 years

[marcosdumay]

"The brain is NOT a computer. Computers themselves are simple expressions of a mere slice of how we understand our own mental processes to work."

Funny, last time I saw, computers were machines capable of emulating any mathematical operation... Not based on how we think our brain work, nor limited to it.

Re:We WILL have androids in 20 years

[marcosdumay]

"~$25k for stereo vision cameras (per set)"

Stereo cameras are a kind of sensor that improved a lot on the last 15 years. Today we get better prices, better focus, the entire set is smaller and lighter, uses less energy...

Yet, we can't make a good use of them. That is because there is no computer we can put on a mobile robot able to get all the data a stereo head gives us on real time. We fall back to getting partial data, and optimizing for what we think the robot will face. We lose lots of flexibility doing that, creating situations like not being able to "see" a blue car at blue background, or aerial barriers.

Re:We WILL have androids in 20 years

[inca34]

I agree that the algorithms are by no means a perfected art, yet. However, once they are chosen and run through the appropriate testing gauntlet for acceptance with respect to the requirements of the project, the software becomes firmware and semi-custom hardware. The embedded solutions are way more viable than running the dual quad core intel Xeon boxes with external generators, air conditioning units, Windows XP, etc. It's just a matter of getting the appropriate development time for getting the false positives and false negatives to an acceptable level.

Personally, I don't think stereo vision is the way to go. Get a single cheaper camera with better low light sensitivity and fuse the data with other sensors. That's a much simpler way to weed the false positives out. Use the camera for whatever it can do reliably: some limited object detection and line/lane following. The nice thing about cameras for vision is that it's a passive detection. The rest tend to be active which is vulnerable to jamming. Though cameras are also vulnerable to natural jamming: dust, rain, snow, fog, etc.

It's really a toss up. There is no perfect perception sensor or system that works ALL the time. You're basically faced with being affordable but vulnerable to something, or expensively outfitted with every sensor ever and still being vulnerable to some cases that will be impossible to test beforehand. All I can say is that there won't be any monolithic solutions to this problem. It will be very customized to the individual requirements of the autonomous vehicles.

Re:We WILL have androids in 20 years

[samkass]

From the speech: "Today, we can bank by computer, shop by computer, and send letters by computer. Only a few companies and individuals use these services, but the number is growing and existing capabilities are limited."

I agree that no one could have predicted every detail, but the web was created, in part, with the money Gore drove through Congress to fund exactly this sort of capability. Considering the speech is from 1986, I think it's pretty prescient.

Re:We WILL have androids in 20 years

[marcosdumay]

"...the software becomes firmware and semi-custom hardware."

We are still not capable of gethering all possible info from a stereo head (at several fps), even with totaly custom hardware and the fanciest algorithms available. Well, theoreticaly, we can, but it would suck so much power (besides being expensive) that a mobile robot wouldn't go anywhere.

I disagree that stereo vision isn't the way to go. Jamming is too common on medium-to-small sensors, even from natural sources. Also, a passive sensor could use much less energy than an active one, even more when you are at open field, when you need long distance sensoring.

But on big robots (SUV sized or bigger), radar and sonar are viable alternatives.

Learn to read

[Anonymous Coward]

Actually, 6 robots reached the target during the 6 hour limit. 3 of them (Cornell, MIT and Benjamin Franklin) however took a lot more time than the first 3, so they won't probably win.

The description is completelty wrong.

corporate sham military industrial complex

[Anonymous Coward]

I haven't read whole article but the whole things seems like a sham. Why not have just one course? Whomever finishes one course in the shortest time wins. By having subjective judges decide whom of 3 wins, all it basically says is that whomever has the most political power wins. This contest is about money and whomever gets the contract to build this robot for the military. Naturally the military wants someone connected to them to win and if they can change the race from an impartial race based on time to a subjective race with judges so much the better. Oy vey.

Re:corporate sham military industrial complex

[Anonymous Coward]

because you can just get am m1a2 main battle tank and drive thru the obstacles in a straight line on a single course and win. thats not the objective. some of those vehicles are large 1 ton pickups which can do that to most obstacles. the objective is to see if something can share the road with other drivers.

I believe all 6 finishers were under 6 hours

[leko]

Some of the robots were paused for a long time, and each was clocked individually. There is really no point in speculating as to who the winners are, because in addition to the time, how well the bots obeyed traffic laws as well as just how safe they drove in general are all taken into account. We should find out the scoring soon enough (sometime this morning.)

Better Coverage

[Anonymous Coward]

That article is pretty sparse on detail. The best coverage I found was at

http://www.theregister.co.uk/2007/11/03/darpa_uc_blog/

How Long?

[xLittleP]

How long until one of the sore losers goes into "Destroy All Humans" mode?

Actually on second thought, wouldn't that be the one DARPA would want to have? It's win-win!

You make one fatal flaw

[SmallFurryCreature]

You link computing power with intelligence, clearly this means you now NOTHING about programming. Doom does NOT become F.E.A.R. by simply running it on a faster CPU.

SOMEBODY has to write the program that becomes the AI. It REALLY does not matter that much how fast the underlying hardware is that executes that program, the simple fact is that AI code right now just ain't that smart. Not even if an AI can take weeks to calculate can it come anywhere close to what a human can do in terms of reasoning with the input available.

A smart program that is just very slow would be an amazing breakthrough and if that happened then all we need to do is wait for computers to get faster, but right now the AI code just ain't there. If it was, it would long since have been given a supercomputer to run on.

These robots in the challenge have a simple task that any human can do, "see" the enviroment, and act upon that information. For years this has been attempted and the systems just ain't getting any better despite the fact that computing power has skyrocketed. Simply put, no code exists that can take a video image and turn that reliably, consistently in information that tells the decision making software what the enviroment is like.

For instance, the detection systems have problems with blue colored cars against asfalt. Consider a human being, put a car painted blue against a background painted the exact same color in blue lighting so it totally blends in. WOuldn't fool a human for a second since we would still see the windshields and through it the interior of the car and reason out that there must be car there even if we cannot see the bodywork.

Same with an other obstacle, a barrier hanging in the air, the teams actually complained about this because they thought all the barriers would be on the ground. This shows you why AI programming is so bad, the programmers are morons, the barrier involved is very common at road blocks. The car, designed and programmed to only scan the ground is unable to determine that a barrier might higher up.

Worse, when it hits it, it can't react to it. The cars have to be stopped, not one of the cars, not even that best was able to simply stop, backup and try a different course.

You can throw more GHZ at it, but all that will give you is faster dumbness.

What happens when a computer has the same complexity as the brain? You will have a very fast, braindead piece of machinery. It is the programming that matters.

Your anology to flying is flawed, we knew that things could fly, gliders had been around for ages, all that was needed for a power source that had good enough power to weight ratio. We do NOT have the AI code or any idea how to make it. Compare it to say faster then light travel. We don't know how, so claiming that if only we develop an infinite source of energy we can do it, is flawed.

Re:You make one fatal flaw

[Orange Crush]

I agree with you except on one point:

Your anology to flying is flawed, we knew that things could fly, gliders had been around for ages, all that was needed for a power source that had good enough power to weight ratio. We do NOT have the AI code or any idea how to make it. Compare it to say faster then light travel. We don't know how, so claiming that if only we develop an infinite source of energy we can do it, is flawed.

FTL violates physics as we know it and we've never observed anything indicating it's possible. We *know* sentience is possible and the hardware and code exists. It's right behind our eyeballs. Human-level intelligence can be had in a device smaller than a bowling ball giving off less waste heat than a 100 watt light bulb.

I fully agree that most don't understand the magnitude of the problem and we have a very long way to go, but we know for a fact that it's 100% solvable. Nature already did it.

Re:You make one fatal flaw

[shystershep]

we know for a fact that it's 100% solvable

I'm afraid I have to disagree with your logic. Yes, physics as we know it would be violated by faster than light travel, so we certainly don't know if it can be done at all. Your argument for AI is flawed, though: simply because we know sentience is possible, it does not follow that we know sentience can be created artificially. We know sentience is possible in biological organism, but we do not know if it can be recreated in a machine. Even if your definition of AI includes creating an organism that has sentience, as opposed to the current understanding of AI (machine/software-based), your statement that "it's 100% solvable" does not necessarily follow.

I think it's somewhat more likely than not that we will eventually develop true AI, but I don't think you can jump from the mere fact that sentience exists to saying that artificially duplicating it is a given.

Re:You make one fatal flaw

[Orange Crush]

I'm trying to avoid the whole "what's natural vs. what's artificial" debate and look at it from a perspective of simply "what's possible". We know sentience is possible, because we're here. We know animals can do advanced optical pattern recognition, navigation, and lots of other "hard AI" problems because they're here. We only understand a fraction of the complex electro-chemical interactions happening within brains to create intelligences. We know a lot of it is analog, and it may not be feasible to map the same processes onto a digital computer. But we do know that our brains contain no magic and no unobtainium. They're incredibly complex machines that we presently understand poorly, but there's no indication we never will

I guess I don't understand your logic that AI might be unsolvable. We have working models to poke and prod and reverse engineer to our heart's content. We may not figure it out in our lifetimes, but given that intelligence has been created before, what could possibly preclude it from being created again?

Re:You make one fatal flaw

[644bd346996]

Wow. If it weren't for the subtlety, I'd think you were an Intelligent Design guy.

Anyways, it is perfectly valid to say that the problem (creating sentience) is 100% solvable. If it weren't, humans couldn't exist. (Your point about "artificially" is meaningful only for a fixed and narrow definition of artificial.) The thing that might not be right is to say that the problem is 100% solvable by humans. We might not be smart enough to create an AI. But I think we are. It seems pretty certain that we'll soon have the computational power to do a brute-force physical simulation of a brain, and I wouldn't be surprised to see it done in real-time in my lifetime. Beyond that, the only challenge it to reduce the simulation to more abstract thought processes that require less computational power. Given the progress that's been made so far, with fairly limited computers, there's no reason to expect that we'll hit an insurmountable brick wall before creating something that can pass the Turing test.

Re:You make one fatal flaw

[Anonymous Coward]

Doom does NOT become F.E.A.R. by simply running it on a faster CPU. Actually, it probably would if you used a scalable raytracing engine instead of a raster graphics engine that has to be practically re-written to add significant new effects.

Re:You make one fatal flaw

[Ironpoint]

"We know sentience is possible in biological organism..."

And what evidence do we have of this? A bunch of biological machines running around saying "I'm sentient" is not good enough for me. No one can explain where sentience comes from or at what point on the tree of life it begins. Most people would agree that bugs and dogs are not sentient but argue that people are without explaining much about their reasoning. The simplest explanation is that people, dogs, and bugs really aren't sentient even though people may believe, talk, and act as if they are. In this case, it should be possible to recreate human behavior perfectly in a machine.

Re:You make one fatal flaw

[IlliniECE]

Comparing this to FTL shows that YOU don't have a very good understanding of the issues. And to call the programmers morons? Ok.. would you mind posting the code here on slashdot that would win this competition just like a human? I'll be waiting...

Re:You make one fatal flaw

[Anonymous Coward]

Actually it does matter how computer processing power continues to increase.

Currenty computers execute a single thread very fast, but the brain executes in parallel which equuates roughly to being several hundreds of thousands of times faster at processing than the fastest current general purpose CPU. In 20 years this may change and brain-like performance may become a reality.

But that does not make a brain...and if you're a dualist it never will ;)

What makes a brain is a fast CPU(s) *and* a program that can model a brain. This is becoming a reality and we'll be soon enough (~20years according to moores law) know whether this is correct....and we don't need to worry about guessing and worrying too much until that time really. Maybe all of the worriers ought to take a decade off!

Re:You make one fatal flaw

[vertinox]

And what evidence do we have of this?

I think he may have confused intelligence with sentience. Neither require each other, but it helps.

As in...

My cat is sentient but he's not intelligent enough to drive my car.
vs
These robots are not sentient but they are intelligent enough to drive my car.

Of course I have no evidence to really prove my cat is sentient other than he appears to be so and I don't feel like cutting him open to double check that he didn't get replaced by a cat android while I was sleeping.

In fact, you could never really prove any other sentient beings sentience (consciousness) other than your own and it is not measurable in a scientific way. Intelligence, on the other hand, can be measured by observing the beings skills in its understanding of cause and effect.

In that regards... My cat is intelligent enough to understand that if he hears a can open, then it means that food is forthcoming.

Re:You make one fatal flaw

[fain0v]

You may know something about programming, but you don't seem to know anything about biology. We are all molecular machines made of carbon, nitrogen, oxygen, hydrogen, sulfer, etc. I can at least understand people that make the argument that it will take us hundreds of years to develop AI. But to argue that only "biological" systems are capable of it is absurd.

Re:You make one fatal flaw

[shystershep]

Try reading my post again. I was commenting on a previous post that claimed that, since we know sentience/intelligence is possible, then we can be 100% sure that we can replicate it mechanically. My point was that the logic does not extend to a different kind of system, i.e., just because we know intelligence is possible with one type of system (biological) it does not necessarily follow that it is possible with a mechanical system.

I specifically said that I thought machined-based AI would be possible; all I was trying to do was point out what I saw as a logical fallacy in the parent post, not rate the relative merits of biological over machine-based AI systems. I don't know enough about either to make that comparison, but I do know language and logic. If there is absolutely no difference between how biological organisms and machines 'think,' then my point fails. I admit I don't know biology well enough to say -- but it is my understanding that nobody currently understands how the brain works.

Re:You make one fatal flaw

[Anonymous Coward]

Only thing is... How many programmers out there have made software that writes itself? That seems to be what happens in the learning process to some extent in species that don't rely too heavily on instincts. Intelligent living things seem to be able to add new variables and fill in the blanks as exposure to new sources of data grows. This might work for some neural architecture. But how many good examples mimic biology by having processing and memory distributed across the nodes? And then perhaps for efficiencies not quite fully understood, have regions of nodes more responsive to particular data types. It might be possible to virtualize this in a more conventional computer architecture, but it still seems a ways off.

And even if that can be pulled off, then someone gets the lengthy task of teaching the AI what they want to do by repeatedly exposing it to various scenarios. Part of this will probably involve pushing a few pre-made variables to weigh behavior towards a desired action. And then on top of that, maybe throwing in some other tricks to save/reward a desirable behavior while deleting/punishing unwanted behaviors. This might take longer than the traditional programming method, but could give more adaptability in the long run.

another reply

[SmallFurryCreature]

You mention chess. Alright, Deep Blue. Lets challenge deep blue. Half way through the game, we switch the board, introduce a new rule. Jumble the pieces up and tell it to pick them up and put them in the correct place again.

NOT a challenge for a human being. Deep Blue? Will fail totally, unable to even understand the commands.

AI is NOT the same thing as doing a simple task over and over again really fast. Laser range finders are nothing new, ACTING upon that information, THAT is the challenge. Especially when that information is not constant and reliable.

Kasparov showed that when he switched styles constantly, Deep Blue was unable to cope. That Kasparov went on to beat Deep Blue is often forgotten. It however showed very clearly that Deep Blue had been setup by HUMANS to beat Kasparov, when he became another player by changing style, the computer could not cope, it had no AI to deal with this.

It reminds me of futurama and robot blurns ball. Putting a howitzer on the field does NOT prove robots are good pitchers. IF Deep Blue could be put in front of a `checkers board and pick that game up in seconds, like a human could, then switch to TicTacToe and then play some poker ALL without human input, then I would be impressed.

Re:another reply

[Anonymous Coward]

IF Deep Blue could be put in front of a `checkers board and pick that game up in seconds, like a human could, then switch to TicTacToe and then play some poker ALL without human input, then I would be impressed.

The test proposed cannot be passed by many humans, why should Deep Blue be subjected to it ?

I can play TicTacToe. Put me in front of a checkers board and I won't do a move. I haven't played checkers in ~30years, don't even remember the exact rules, no matter how simple they are. I don't play poker, don't know the rules (I do play other card games).

Those who can switch between those games must have learned all of them previously, yet you deny that learning to Deep Blue. Even providing the rules to all would not be enough, one has to train to become even mildly successful at most games.

Re:another reply

[Dixie_Flatline]

It's also worth noting that Chess and Checkers and all games of that type which claim to be in the field of 'AI' are really just extensions in the field of deep search. Deep Blue doesn't REASON about chess, it consults a giant play book. If it's outside of the playbook, it actually just runs through millions of possibilities, as deep as it can. Even still, there was always a team of master chess players keeping watch over Deep Blue's actions.

Deep Blue has never had anything to do with AI, despite hype to the contrary. It has chips specially designed to search chess moves faster. I'm no more impressed by this than by a person that can look up names in a phone book, and failing that, call 411.

At least this urban challenge is trying to get the systems to react to a changing and complex environment. This stuff is ten times the AI research that Deep Blue was.

Well, it's a nice start I suppose ...

[ScrewMaster]

Wake me when a robot can drive me to work down I294 in rush hour during construction season. I'd buy one, because then I could take a nap while the robot drove me to work.

Re:Well, it's a nice start I suppose ...

[Sparr0]

10 years. Tops. In 20 I expect it to be illegal for a human to operate a vehicle on a major roadway. 50% of all accidents are caused by drunk drivers, but 99% of all accidents are caused by human drivers.

Re:Well, it's a nice start I suppose ...

[ScrewMaster]

Seriously, that would be, well ... cool. Waaay cool. And if all those vehicles operated on a mesh network, communicating with themselves and with remote data sources (weather services, etc.) traffic jams could be a thing of the past, and we'd all get where we're going more safely, a lot faster, and on less fuel. If some blockage occurred that would otherwise cause traffic to pile up, cars could automagically route themselves around the problem. "Sir, we are talking an alternate route to the airport because a bridge collapsed five miles ahead." Sounds idyllic ... I just hope that America is enough of an industrial nation in ten years to be able to afford the technology (or even to have need of it.)

Re:Well, it's a nice start I suppose ...

[Anonymous Coward]

You are rather optimistic.

The general population (that is, dumbasses) will be SO OUTRAGED if a law like that actually got proposed. "I got a right to drive my car, goddamn it!" It doesn't benefit any corporations, so nobody except for small activist groups will be pushing for it.

It would never fly.

First thing that came to mind...

[CAIMLAS]

Autobots, transform?

urban taxi system

[sgt scrub]

I don't know why the contestants are spending all that cash and beefing up the AI on these machines. A tape recorder that mumbles incoherent obscenities could pass as a NY cab driver.

D'oh

[evilviper]

Three autonomous vehicles crossed the finish line within the 6-hour time limit here at the DARPA Urban Challenge in Victorville, CA.

Why is NOW the first time I've heard where it was being held? I looked through the past /. articles, and NONE of the pages linked bothered to put in the two magic words to tell readers where it was being held. You had to navigate DARPA's site to find that info, so basically, only those people already determined to go would find out where it was happening.

Had I at all known it was being held so close, I would certainly have gone and watched.

I eat crow

[CrazyJim1]

I thought these things weren't going to happen for at least a decade(the desert challenge that is).

Obligatory

[MHPanruka]

I for one welcome our three autonomous navigation robotic overlords.

Agree on most counts, but...

[bdwoolman]

No voice recognition? Have you ever tried Dragon System's Nat Speak 9.0? Seriously. I am just a very happy user not some shill. It is astounding. However, I am also very disappointed by the future. I mean here we are and we are nowhere near where I thought we would be by now. No mars, no moon presence, no cool robots. Okay. The internet, but the telegraph was far more of a jump in it's day than the net was in it's. However IMHO you can check speech to text off the list.

Re:A factoid about the Urban Challenge

[zero_offset]

Should we be reading anything into this?

Yeah: Urban Challenge II will be cool as hell.

This just in

[DollyTheSheep]

Urban Challenge Event winners announced!

1st Place - Tartan Racing, Pittsburgh, PA

2nd Place - Stanford Racing Team, Stanford, CA

3rd Place - Victor Tango, Blacksburg, VA

Re:This just in

[statemachine]

And here's the link to DARPA's website.
http://www.darpa.mil/grandchallenge/

The winners have been announced

[SnowZero]

On the official website.

1. Tartan Racing (Carnegie Mellon)
2. Stanford Racing Team (Stanford)
3. Victor Tango (Virginia Tech)

Re:The winners have been announced

[qeorqe]

I have submitted a story. It has a few more details about the race. It does not have the details of the elapsed times or the time corrections.

videos of the event

[OtherFarm]

darpa urban challenge videos ordered by date http://video.google.com/videosearch?q=darpa+urban+challenge&num=10&so=1&start=0 We were prepared on communistrobot.com

RADAR isn't perfect

[Anonymous Coward]

> RADAR is the right sensor for this type of work.

Just my point of view: maybe a sophisticated type of radar would be good, but in real-world scenarios, the kind of radar I know about is too easily jammed or confused by other radars and interference sources. With the shorter wavelengths, lasers are a lot more directional and interference is easier to reject. Obviously passive optics presents big challenges, but I suspect *eventually* (rather long-term) they will be overcome.

Wondering which OS the winners runs?

[black_penguin]

Fill in the blanks.

1) Tartan -
2) Junior - GNU/Linux; Fedora Distro.
3) Victor Tango -

Re:Wondering which OS the winners runs?

[Anonymous Coward]

Victor Tango has 2 dual cpu quad core servers. One mainly used for planning running fedora linux, and the other mainly running perception software on windows.

Just simulate the physical brain and body

[coult]

We don't necessarily have to understand human intelligence in order to duplicate it. Given a fast enough computer one could simulate all the individual cells in the human brain and body with high fidelity, thereby creating a "virtual" human who has human intelligence (and flaws, too).