RE: Robot car closes on driver track times

RE: Robot car closes on driver track times

Monday 5th November 2012

Robot car closes on driver track times

Self-driving Audi TT getting smarter on the race track



A self-driving car is coming close to beating human drivers round a race-track, according to the California university that's developed it.

The car is a standard 265hp Audi TT S dubbed Shelley with of host of extra computing power in its boot. It's the creation of the Centre for Automotive Research at Standard University (CARS, handily), and it has already driven itself up the Pikes Peak course and reached 120mph on track.

The idea is not to create a robot race series, says Professor Chris Gerdes, head of the CARS lab, but to use the info to make cars safer. "If we can figure out how to get Shelley out of trouble on a race track, we can get out of trouble on ice," Gerdes said.

TT takes itself off for some track playtime
TT takes itself off for some track playtime
Shelley, named for Pikes Peak winner and Audi Quattro driving rally hero Michele Mouton, is equipped with a smart GPS that knows where it is to the nearest 2cm. Radar and laser sensors also help position it on the racetrack, with further information coming from the usual array of electronic feedback standard on many cars these days, such as yaw rate detectors and wheel-speed sensors showing tyre grip.

So far drivers have still proved that bit faster, although the gap is closing to within seconds, according to the team. "Human drivers are very, very smooth," Gerdes said. 

He said the car is less good at feeling where its limits lie, while the best drivers know that the quickest way round a corner on the limit might be to use the throttle as well as the steering. Or that going too wide on one corner might better set them up for the next.

The track they've been using is the Thunderhill circuit north of Sacramento but to better understand how humans drive fast they've strapped monitoring systems to a driver at the Rolex Monterey Motorsports Reunion race at Laguna Seca. They also harvested data from the car, a 1966 Ford GT40.

"We need to know what the best drivers do that makes them so successful," Gerdes says. "If we can pair that with the vehicle dynamics data, we can better use the car's capabilities.

The idea of autonomous cars might not be your idea of progress, so take comfort in a line from Gerdes himself back in October. Self driving cars, he said, have been "20 years in the future ever since about 1939".

 

Author
Discussion

E38Ross

Original Poster:

35,068 posts

212 months

Monday 5th November 2012
quotequote all
so we have a thread telling us that it's not far away from driver times, yet no mention of times? ok.....

WCZ

10,523 posts

194 months

Monday 5th November 2012
quotequote all
how long before manufacturers are using robots to get the fastest ring times possible? within 10 years IMO

HustleRussell

24,690 posts

160 months

Monday 5th November 2012
quotequote all
Robots slower than Humans by a margin of 'seconds'. Quite a lot, then...

Cotty

39,533 posts

284 months

Monday 5th November 2012
quotequote all
WCZ said:
how long before manufacturers are using robots to get the fastest ring times possible? within 10 years IMO
How about if they replaced the ring taxi drivers with robots, would you go for a ride?

otolith

56,085 posts

204 months

Monday 5th November 2012
quotequote all
Didn't BMW do this years ago?

http://www.youtube.com/watch?v=nlJ_XGjewdk

McSam

6,753 posts

175 months

Monday 5th November 2012
quotequote all
WCZ said:
how long before manufacturers are using robots to get the fastest ring times possible? within 10 years IMO
Almost no chance whatsoever.

A car and a circuit are far, far more variable than any computer can be programmed for, and no matter how thorough a job is done, it isn't going to come close to a proper, experienced race driver in any of our lifetimes.

To say a computer could ever drive faster than the fastest human misunderstands how much of an art driving a race car is, and the limitations of programming!

Raitzi

640 posts

212 months

Monday 5th November 2012
quotequote all
Make robot drift the whole track, then I believe it is possible. Understeering TT is easier to manage.

Contigo

3,113 posts

209 months

Monday 5th November 2012
quotequote all
Cotty said:
How about if they replaced the ring taxi drivers with robots, would you go for a ride?
Is it any different than getting on a plane where only about 5 minutes of the journey is actually human controlled? Even for health and safety the ring cab would have to have a human at the wheel in case things went wrong with the automation.



405dogvan

5,326 posts

265 months

Monday 5th November 2012
quotequote all
McSam said:
Almost no chance whatsoever.

A car and a circuit are far, far more variable than any computer can be programmed for, and no matter how thorough a job is done, it isn't going to come close to a proper, experienced race driver in any of our lifetimes.

To say a computer could ever drive faster than the fastest human misunderstands how much of an art driving a race car is, and the limitations of programming!
'I think there is a world market for about five computers.' T Watson - Founder of IBM

You're THAT wrong...

All you need is money - the technology already exists but you're going to trash a few cars along the way and it will take time for the computers to 'learn' the fastest way around the track (and their mistakes will be expensive and noisy).

Thing is - once you've done this, you'll have a program which can drive the track in any conditions - could learn changes to the track - could analyse the surface and allsorts - all WAY beyond the abilities of a human being.

The only reason it's not happening is

a - it costs a lot
b - I'm not sure what you gain by it - nothing new need be invented.

Edited by 405dogvan on Monday 5th November 11:35

405dogvan

5,326 posts

265 months

Monday 5th November 2012
quotequote all
Contigo said:
Is it any different than getting on a plane where only about 5 minutes of the journey is actually human controlled? Even for health and safety the ring cab would have to have a human at the wheel in case things went wrong with the automation.
Most airline passengers have no idea that 90% of their flight is computer controlled - in fact 100% of it is with most modern jets as the onboard systems won't let the pilot do much/will actively stop them doing stuff etc. etc.

Pilots are slowly becoming 'comfort' devices to make you feel good - train drivers have been this for decades (I'm not saying you don't need a human to run a train - but they don't need to be IN the train! smile

Dr Z

3,396 posts

171 months

Monday 5th November 2012
quotequote all
article said:
...the creation of the Centre for Automotive Research at Standard University (CARS, handily)...
So where is this Standard University then? Google can't find it, therefore it must be in Mars. wink

Mr Whippy

29,029 posts

241 months

Monday 5th November 2012
quotequote all
WCZ said:
how long before manufacturers are using robots to get the fastest ring times possible? within 10 years IMO
What's the point?

You may as well just run a simulation and get the result in seconds biggrin

Human drivers might naturally look at a shaded area in winter and think ice, slow down, will the AI driver do that? Leaves on the road reducing grip?

Iirc there was a game using some kind of survival of the fittest algorithms, it was a PS1 based Ferrari 360 racing game. They gave the system the inputs (I guess speed/rpm/position on track and some look-ahead data) and car controls and let the cars go, and the ones that even drove a few metres were then re-bred and evolved and slowly they had cars that would drive around the course.

That was well over a decade ago, so I can only imagine that adding those inputs and outputs to a real car would result in a car that got faster and faster as it learnt how to interpret the i/o data to go faster basically. Racing is pretty easy really. Understeer/oversteer thresholds and xyz g sensing, yaw sensing, it won't take long to build up a table that shows what the car can and can't do, and then you just let it learn from statistical results from doing lots and lots of laps.




The problem with all these computers is they are in ideal cases so far though.

They can either do the performance bit, or the safe driving around obstacle courses bit, but I've yet to see a blend of the two where they hoon it through a situation where a hazard may present itself and the performance in that hazard is acceptable.



As good as computers are they still have a huge way to go to being as good as humans to assess a situation and make a good decision based on lots of factors.
Car control, sure, 'seeing' everything, yeah pretty good too, but making the choice on *what* to do?

How do they decide what to do, run over the kid or the OAP? Given the choice it'd probably run them both over in confusion not realising there was no winning move biggrin


I'd only trust a computer driven car absolutely when they have driven on every road on the planet at least 100 times and have a 100% safe record. Until then they are just as fallible as a good human driving waiting for that rare occasion to do something wrong.

Dave

Edited by Mr Whippy on Monday 5th November 11:50

McSam

6,753 posts

175 months

Monday 5th November 2012
quotequote all
405dogvan said:
McSam said:
Almost no chance whatsoever.

A car and a circuit are far, far more variable than any computer can be programmed for, and no matter how thorough a job is done, it isn't going to come close to a proper, experienced race driver in any of our lifetimes.

To say a computer could ever drive faster than the fastest human misunderstands how much of an art driving a race car is, and the limitations of programming!
'I think there is a world market for about five computers.' T Watson - Founder of IBM

You're THAT wrong...

All you need is money - the technology already exists but you're going to trash a few cars along the way and it will take time for the computers to 'learn' the fastest way around the track (and their mistakes will be expensive and noisy).

Thing is - once you've done this, you'll have a program which can drive the track in any conditions - could learn changes to the track - could analyse the surface and allsorts - all WAY beyond the abilities of a human being.

The only reason it's not happening is

a - it costs a lot
b - I'm not sure what you gain by it - nothing new need be invented.

Edited by 405dogvan on Monday 5th November 11:35
Not exactly what I meant - and you have to realise that it's nowhere near as simple as learning the "fastest" way around the track. Of course you could program the computer to match a particular line and even, if you did a really good job with technology that is still in its infancy, get it to be able to match that line at nearly the maximum speed allowed by the conditions at that time.

But that completely disregards that the conditions change the fastest line around the circuit, and that's where the art lies. The prediction, not only by looking at the condition of the surface at an exact point but also by the feeling through his arse, that a race driver can make is far beyond the abilities of computing, and it means that the driver will be able to find the fastest way far more easily and more intuitively than the computer. Iteration methods, even really good ones, are such that a large change in conditions could take thousands of iterations for the computer to work its way back to the very fastest line. An experienced driver will have it in five laps.

And even assuming the programming is possible, all this could be done with only one car, and would need complete reprogramming to use anything else, unlike a real driver.

There's almost no question of ever making a "better" race driver than a person.

ETA - Clarifying I'm talking only about race drivers.

Edited by McSam on Monday 5th November 11:52

otolith

56,085 posts

204 months

Monday 5th November 2012
quotequote all
Mr Whippy said:
Human drivers might naturally look at a shaded area in winter and think ice, slow down, will the AI driver do that? Leaves on the road reducing grip?
It will if you program it to - and it has as many extra senses as you can devise sensors for, including those belonging to the car.

GranCab

2,902 posts

146 months

Monday 5th November 2012
quotequote all
http://me.stanford.edu/groups/design/automotive/

STANFORD Uni - not Standard ... teacher

GhostDriver

878 posts

192 months

Monday 5th November 2012
quotequote all
Must be like watching Damon Hill all over again

NISaxoVTR

268 posts

169 months

Monday 5th November 2012
quotequote all
Who let these robots out of the kitchen, it used to be they were going to do our laundry and cooking not hooning around in our cars scratchchin

R66STU

273 posts

176 months

Monday 5th November 2012
quotequote all
How does the car adapt to tyre wear ? and what about darkness ? I assume it can go flat out in the pitch black without the headlights on.. so maybe night time laps would be better than a real driver ?? and in the rain there would be no need for the wipers !!




offshorematt2

864 posts

216 months

Monday 5th November 2012
quotequote all
Mr Whippy said:
As good as computers are they still have a huge way to go to being as good as humans to assess a situation and make a good decision based on lots of factors.
Car control, sure, 'seeing' everything, yeah pretty good too, but making the choice on *what* to do?
From a lot of the drivers I see on the road, it wouldn't be take much to be able to assess and react to a situation better than your average commuter though...

Mr Whippy said:
How do they decide what to do, run over the kid or the OAP? Given the choice it'd probably run them both over in confusion not realising there was no winning move biggrin
Love the example - they should put that in the theory test for learners. That would keep the pass rates down... biggrin

DannyScene

6,624 posts

155 months

Monday 5th November 2012
quotequote all
405dogvan said:
Contigo said:
Is it any different than getting on a plane where only about 5 minutes of the journey is actually human controlled? Even for health and safety the ring cab would have to have a human at the wheel in case things went wrong with the automation.
Most airline passengers have no idea that 90% of their flight is computer controlled - in fact 100% of it is with most modern jets as the onboard systems won't let the pilot do much/will actively stop them doing stuff etc. etc.

Pilots are slowly becoming 'comfort' devices to make you feel good - train drivers have been this for decades (I'm not saying you don't need a human to run a train - but they don't need to be IN the train! smile
Trains are always something i wondered about, just how neccesary are the drivers?