Computer Will Take On Formula 1 Champion

Jacky Baltes writes: "Thought that Deep Thought vs. Kasparov was a big deal. I am part of a research group that attempts to beat the world champion in Formula 1. The goal of the Man v. Machine Challenge is to design and implement a robotic system that can drive a F1 car faster than the current world champion. You can have a look at the progress at the Man v. Machine Challenge Web site . We will had some more technical details about our control system design, data fusion, and car model to the site later. So Michael, hold on to your head. Jacky Baltes"

Comments

but will it..

[geekoid]

.. be able to make a phone call, shave and drink coffee? all while flipping off the guy behind it?

Re:Is this to be *in* a race?

[cathryn]

Well, I think a computer would ultimately have
one advantage, that would be difficult to resolve,
and that would be that it'd be simply unafraid
of death. And, that if you could send a computer
car, barreling through a race, slightly clueless,
but unconcerned about it's own mortality, then
I think the human racers would just have to get
out of the way.

With live drivers, isn't there a slight matter
of 'how much do you want to win' versus, 'how
close are you willing to go to the edge' that
doesn't quite translate when machines are involved.

I'm surprised...

[dark_panda]

... that nobody's made a Microsoft joke about the car literally crashing on Windows 2000.

c'mon, people.

J

We all know what is going to happen...

[Glowing Fish]

The human driver is going to win, but after he wins he is going to stumble out of his car, sweating, and die of exhaustion.

Deep BLUE (not "Deep Thought")

[Apotsy]

First off, the computer that beat Kasparov was called "Deep Blue", not "Deep Thought". Secondly, it was able to beat him because basically, they cheated.

They fed the program Kasparov's entire game history while keeping its game history secret from Kasparov. Normally in competitive chess you are allowed to study your opponents past games in order to learn what tactics they are likely to use. In this case, Kasparov wasn't allowed to do that. The fact that he went ahead with the game anyway was probably due to overconfidence on his part.

The folks at IBM seemed to realize that they won merely because of the setup, and thus when challenged for a rematch by Kasparov, they said they weren't interested, because they had "done everything they set out to do". (Personally, I think they were scared they would loose in a fair match.)

Kasparov has stated publicly that if the "Deep Blue" team actually abided by the rules of competitive chess, he will "tear [Deep Blue] to pieces". They have so far declined.

That said, even if the machine were able to beat the best human player in a fair match, it still would not be that remarkable, because computer chess programs are still limited to the "brute force" approach, where they pick their next move by simply searching as big an area of the total possible game tree as possible. The human mind does it differently, only examining at most a dozen or so possible moves before deciding. Ho the brain can pick such strategic moves without searching a significant portion of the total possible game tree is still one of the great mysteries of cognitive science.

Re:Deep BLUE (not "Deep Thought")

[n xnezn juber]

If you did your homework you would realize Deep Blue was originally called Deep Thought when it was developed at CMU. IBM did work on chess computer and eventually renamed it Deep Blue. So Deep Blue was its name when it beat Kasparov but Deep Thought is really the same thing.

As for Kasparov... you mean Kasparov had to publish all of his private practice sessions and give them to IBM? Nope... let's put it this way... all of Kasparov's public games were known. All of Deep Blue's public games were known. What they did to practice for the event was unknown. Where is the problem? If I have a game with Kasparov and I am unrated, just joined FIDE and have no public games... and win (it's possible!!!) does that mean I didn't deserve to win? Personally I think Kasparov did not win because he tried to out think Deep Blue and its programmers... kind of like reverse-reverse psychology. He did not play like he would with a human opponent (but then most people realize you often have to play differently with a computer).

Now I absolutely agree that Deep Blue is a nearly worthless effort if their only goal was to beat Kasparov. It has no tact and is a brute force approach to a elegant game. The human mind such as Kasparov's is tuned to such precision that research into how the brain learns I believe is many times more important than trying to find the best way to brute force a game. Uh... but then I supposed the brain is itself a sort of brute-force mechanism with 100 billions neurons. Who knows if the development of large scale parallel computation systems like Deep Blue will eventually lead to developments as inredible as the brain.

Re:Deep BLUE (not "Deep Thought")

[Nexx]

Deep Blue was a machine optimised for graph searches. Now, given this, Kasperov claims that there may have been a human component in his match with Deep Blue. According to him, he claims that if a decent human pruned some of the search trees for Deep Blue first, then DB would have no problems performing like a world-class chess player. Something to think about....

--

Not a hope

[Gorobei]

This is so far beyond anything you could train an AI to do within three years that it's not even funny.

Consider chess: you have a vast archive of previous games, a relatively simple domain, the ability to test millions of boards a second, almost free live testing, and almost no financial penalty for mistakes. Contrast F1 racing: no archive, complex domain, almost no simulation ability, real testing costs $1000/hr, the mistake penalty is $100,000.

This is either hopelessly naive or a scam: after three years, you might get an AI around the track at 100MPH. Judging from the website, it's a scam: they talk about all the great value of the webhits and PR, ask for sponsors, etc. There is almost no info on the AI approach, etc.

Looks like nothing but a money sink to me.

suggestion

[Racer X]

The research team should contact the Knight Industries or the Foundation for Law and Government, and hire Bonnie as soon as possible.

(Yes, I read this post and thought, "Jesus Christ, make a Knight Rider reference as soon as possible." May others come and do it better.)

cool

[suitcase]

needs a little r2d2 unit mounted on the top

What if the car kills someone?

[LordNimon]

What if the car enters a real race and something goes wrong, and the end result is that someone dies? Usually, if an accident happens, no one blames the drivers because they all know that they wouldn't intentionally do something like that because it's too risky for everyone.

But a computer has no such fear. It makes decisions based on programming. So let's say that it cuts too close to another car for whatever reason, and in the collision the driver of the other car dies. Is the programmer liable? After all, he is the one who effectively made the decision to cut that close to the other car. But it doesn't affect him negatively because he was never at risk. So someone is going to sue him, saying that he was careless because he was never at risk.

I wouldn't touch this project with a 10-foot pole. If this car ever drives on a real track, it's going to end badly.
--

What *are* the proposed rules, anyway.

[dbrower]

Conditions and vehicles vary so much, that with a fair set of rules, I'd encourage Shuey to take it on. Heck, I'd even let Zonta take it on. But what would fair rules be?

Same FIA Formula 1 car for both operators

Computer must mechanically operate the same controls used by the human;

Computer must fit in the same space as the human, including power source.

Computer can have no electrical connection to vehicle for power or sensors; all its sensors must be self contained, and have no physical extension beyond what is allowed the human (no camera through the floor to follow the line :-)

Computer can have as many hands, legs, arms and eyes as it likes

Computer gets human equivalent sensors only - visible light vision and accelerometers; no radar, sonar, or active illumination allowed.

Computer controlled car must meet same weight requirements as the human/car combination

I won't demand race/traffic interactions. Solo qualification laps will suffice. Even under these conditions, I'll take the human for ten years easily, and probably twenty years.

I don't think it's an "AI" problem as much as a robotics, sensor, and machine vision problem. I don't think there's been so much progress in the last 20 years that it's feasible.

Though I sure Frank Williams and Patrick Head would sign it up as soon as it was available.

-dB

Re:Is this to be *in* a race?

[K8Fan]

With live drivers, isn't there a slight matter of 'how much do you want to win' versus, 'how close are you willing to go to the edge' that doesn't quite translate when machines are involved.

It's a measure of how far we've come that we can actually approach discussing the real-world application of Asimov's "Three Laws of Robotics".

Isaac Asimov's "Three Laws of Robotics"

1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey orders given it by human beings except where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

Clearly, this is a case of the First Law overriding the Third Law. The reasonable thing would be to not try to win the race. (OK, so sue me. I was a SF geek long before I ever touched a computer.)

Re:This smells fishy...

[Trygve]

Absolutely!

I can't believe nobody else is really seeing this. Read through their site (don't worry, there isn't much actual information ...), it's all about *money*! It's about drawing in investors, it's about "free" publicity. It quotes on their main page about how ACM estimates the Deep Thought v. Kasparov match garnered publicity equivalent to ~$125m advertising dollars. Their main page's 2nd link is "Marketing Opportunity."

They have a flow chart for the project, and the biggest component is a bit *thick* arrow pointing at themselves labeled as '$', for crying out loud! Included in that flow chart is a *separate* company that will be doing the actual technical work.

Oh, and what's that company again? "RDD is a research, design and development company." I wonder, did they already have a company by that name, or did they have to think about it for a whole 30 seconds to come up with it? Oh wait, I forgot, these are management types, they probably spent a few weeks in meetings just to determine the consulting company to hire for suggestions.

And as Lish points out, even the semi technical parts are all fluff. If you'll notice, all of the pages where you might find technical information on this are the shortest pages on their site.

What about all of the other R&D going towards self driving cars? It's been going on for years. Every now & then you'll see some more about it in a Popular Science/Mechanics, Discover, and/or Scientific American. Self driving cars have been done countless times, they usually need something special in the road to keep track of and/or a human driver to follow. They mention they've already got positioning equipment all around the track, allowing them to position the car to within 1cm. Okay, that's a huge advantage over other self driving car projects, feasible only because it's in a closed track environment. But what about that other driver, that's still a huge feat to overcome, not to mention driving conditions and other "non-linear" elements. They don't address any of that. All they say is "this is what we want to do, imagine how much money it could bring in."

This site wasn't written for those interested in R&D, the advancement of robotics, or AI, or even F1 racing, this site was written to garner investor interest for a project that I don't think even they expect to be finished.

Until I see real evidence of ground breaking, well funded R&D, I consider this nothing more than a hoax, a deceptive ploy for money.

Don't get too excited, guys.

A bit to strict?

[guran]

Same FIA Formula 1 car for both operators
Of course.

Computer must mechanically operate the same controls used by the human;
Why? The extra mechanics involved adds no real difficulty to the computers task. I agree that the computer should not have any extra control over the car, but what advantage does the computer get by issuing the set_brake_level(50); command insted of extend_left_foot(50); ?

Computer must fit in the same space as the human, including power source.
Yes, we want identical cars, but I'd rather state "Computer must be contained in the car. No remote control." That said I'd be equally impressed by a remote control setup as long as all sensors were in the car

Computer can have no electrical connection to vehicle for power or sensors; all its sensors must be self contained, and have no physical extension beyond what is allowed the human (no camera through the floor to follow the line :-)
Same as for the controls. Wether the computer has a direct feed from the cars sensors or points a camera at the dials and does some image processing is not important. What matters is that the computer must not have access to more information about vehicle status than the human.

Computer can have as many hands, legs, arms and eyes as it likes
Yes, but as I said: I prefer a display of AI not robitics.

Computer gets human equivalent sensors only - visible light vision and accelerometers; no radar, sonar, or active illumination allowed.

Hart to tell what is equivalent. A human has stereoscopic vision to measure distances. I think that would be a hard task to match.

Computer controlled car must meet same weight requirements as the human/car combination
Definitely. The question will rather be: Should the human car add ballast to make up for a heavy computer? I don't think it will be very lightweight...

*IF* this ever comes to be, I'd like to scale it down to an AI problem as much as possible, not a robotics/sensor contest. "Can a computer drive a car at 300 km/h?" is a much more interesting question than "Can a robot controlled by that computer operate the car?"