Re: robot race

On Thu, 13 Oct 2005 12:15:19 -0500, Randy <joe@xxxxxxxxxxxxxxx> wrote:

>B1ackwater wrote:
>> On Sun, 09 Oct 2005 16:11:41 +0200, Borek
>> <borek@xxxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
>>
>>>http://www.cnn.com/2005/TECH/science/10/08/robot.race.ap/index.html?section=cnn_topstories
>>>
>>>So they did it this time...
>>
>>
>> Yep. None completed the course last year. Now five did - and
>> in pretty much the same amount of time. Next year ...
>
>THEY FLY.

Hmmm ... I'd predict TWO years hence. Even using the same
basic obstacle-avoidance methods, moving them up to genuine
3-D will require a LOT more processor power and some tweaks.

A small self-guiding helicopter, UAV or even blimp WOULD be of
great interest to the military, homeland security and to
certain police and search-&-rescue types. Instead of sending
out two coast-guard planes to seach for survivors of a
sinking - send twenty autonomous flyers that can scan the
surface with robotic precision, never tiring.

>> This is evolving very QUICKLY.
>>
>> No doubt each team used a different approach to the problem(s).
>> Interesting that five of them all performed about the same.
>> I hope there's more information about HOW they programmed the
>> 'bots. Could be that there are multiple paths to this kind of
>> 'intelligence' - all just as good as the other. An analogy
>> perhaps as to what we might expect if we meet intelligences
>> from other worlds - or build our own. Like us - but NOT ...
>
>I understand that a major impediment to success last year was the
>inability of all rovers to scan more than about 10 degrees laterally.
>Dunno how true that is, but the performance difference between years
>*is* almost binary.

Ten degrees just doesn't seem enough - and apparently wasn't.
Still, I don't think all the improvement was strictly a matter
of broader scans ... they must have added some 'eIQ' to the
things as well.

Alas, because of the commercial/military potential, it may be
awhile before we get any juicy *details* about just what makes
these things tick.

>> How long until self-driving civvie automobiles are a practical
>> reality. A lot of civvies couldn't have driven that desert
>> test course as is.
>
>I've been wrong before in my pessimistic prognosticating, but...
>
>- The weather was ideal. Add a downpour, dust cloud, or snowstorm and
>see how it goes, especially when the bad weather shuts down the GPS.

There are ways to get around intermittent GPS signals
(weather-related or jammed). With each valid GPS reading
you've got to survey the landscape, pick landmarks, and
update an internal MAP of the terrain. You then proceed
relative to your last valid position, estimating your
new position based on the map. Usually the GPS signal
will return before you run out of map, usually. In leiu
of GPS, falling back to a simple compass can suffice
for awhile too.

>- The rovers didn't have to worry about running into cars that were
>ahead or behind.
>
>- The rovers didn't have to deal with onrushing cars, headlights,
>motorcycles, popeye cars, etc.

Well, it may be a FEW years before such systems are truely
roadworthy ... I didn't mean to sound OVERLY optimistic.
Real-world traffic is a serious challenge even for 'smart'
creatures like ourselves. The one plus for the 'bots is
that they won't be thinking about anything else EXCEPT
navigating traffic. No cell-phone conversations, no
chicks with big tits in tight sweaters, no worries about
employers, doctors, dentists, spouses, kids etc..

However, while it's *possible* to navigate traffic using
purely reactive methods - IF you've got lightning reflexes
and NASCAR skills - you can do a lot BETTER job if you
employ some 'look-ahead' - prediction - based on the
current and/or remembered situation.

Say there's a figure approaching the highway behind
a hedge - but all that shows is its head and shoulders.
If it's wearing a big plastic helmet or is moving in a
way that is not like walking then it's probably a
motorcyclist about to enter the intersection.

If so, what do you need to prepare for ? Should you
slow down ? Move to the left lane ? He MIGHT pull out
without looking - or being able to see you. Same kind
of stuff goes for running children and cars on the exit
road from parking lots - a simple plan needs to be
worked out BEFORE they actually run out into the road.

People (well most) do this automatically, borderline
subconsciously. A 'bot will have to be taught how to
do it - and the better 'bot will have to be able to
learn so it can steadily improve its performance.

'Reactive' is ultimately limited by the physics of
the vehicle - it can only stop just SO quickly, only
maneuver just SO sharply, only move just SO far
before intersecting the vehicle in the other lane.
Reactive plus predictive ... that's what's needed,
and that's what's going to be so HARD. Even people
mis-predict on a regular basis. The 'bots have to
demonstrate at least equal ability compared to the
average driver. Some of that they can make up
through superior reaction - quicker, more precise -
but SOME predictive abilities will be required too.

>- The road had no white/yellow solid/dashed lines that are often faint
>at best, or occluded by dirt or snow. Nor were the roads multi-lane.
>Nor did they have stoplights, stop signs, yield signs, etc, etc.

I don't need yellow lines to be able to drive on a road.
Neither should the 'bots.

>- The rovers encountered no pedestrians or bicyclists.
>
>- The rovers were allowed to drive over any obstruction.

If they could drive over it, then it wasn't really
an "obstruction". 999-ton boulders, 50-foot-deep pits
in the ground ... THOSE are obstructions.

>The rovers didn't have to obey any rules of the road, other than not
>falling of of it. That's a far cry from Johnny Cab...

Far ... but not impossibly far.

>The DoT has spent about a decade looking into the difficulties of
>building a partly/fully automated roadway transport system. Given the
>need for safety, fully automated cars are probably as hard as strong AI.

There's another approach - the 'car train' - which requires
very little eIQ. Radar is used to entrain a couple dozen
cars a fixed distance from each other. This is low-tech
stuff. Holding a lane ... so far they've had to cheat and
put electronic markers into the road. Haven't entirely
figured out how to deal with obstacles either. This is
where the EIs will have to enter the picture.

>Search for "Intelligent Transportation Systems" or "intelligent
>vehicles" or "driver assistance systems" for more...
>
>http://www.itslessons.its.dot.gov/its/benecost.nsf/images/Reports/$File/Inventory99b.pdf
>http://www.its.dot.gov/index.htm
>http://itsdeployment2.ed.ornl.gov/technology_overview/DAS.asp

With a HUGE population of aging baby-boomers - many of
whom will soon be unable to drive due to medical problems -
a real-world 'Johnny Cab' would be a massive moneymaker.
A supendous moneymaker. Eliminate the human taxi driver
and you eliminate a large expense - making the autononous
cabs vastly cheaper to hire and operate. "Johnny Trams"
could replace existing mass-transit too - especially
on shorter, fixed routes like to shopping malls and
such.

Where there's a lot of money to be made, there are a lot
of people eager to find a way. The results of this DARPA
race demonstrates that the rudiments ARE within our grasp
already. Just takes some time and money to build on that.

.

Relevant Pages

  • Re: robot race
    ... inability of all rovers to scan more than about 10 degrees laterally. ... - The rovers didn't have to worry about running into cars that were ... building a partly/fully automated roadway transport system. ... Randy Crawford http://www.ruf.rice.edu/~rand rand AT rice DOT edu ...
    (comp.ai.alife)
  • Re: Camber adjustment on Rover 600
    ... they've had cars in modding mags anyway) that have modded several ... Rovers. ... They must have known about the camber issue when they did the ...
    (uk.rec.cars.maintenance)
  • Re: OK, time for another fucking what car post. Maybe last fo the year.
    ... > Nice cars, but I don't do Rovers. ... Not unless they were built during the ... Prev by Date: ...
    (uk.rec.cars.modifications)