Jump to content


Photo
- - - - -

driverless cars to hit california ( sorry!)


  • Please log in to reply
410 replies to this topic

#51 gruntguru

gruntguru
  • Member

  • 5,273 posts
  • Joined: January 09

Posted 04 October 2012 - 03:07

I'm trying to think of a polite way to mock this, but I can't think of any. So I'll have to refrain from doing it. However, there are at least 10^43 legal positions in chess. To put it in terms easier to understand, it's at least 10,000,000,000,000,000,000,000,000,000,000,000,000,000,000. It might take a while to look up all of them.

If that is the number of legal positions, it is nevertheless very small compared to the number of ways of arriving at those positions from the start of a game.

Advertisement

#52 Greg Locock

Greg Locock
  • Member

  • 4,510 posts
  • Joined: March 03

Posted 04 October 2012 - 03:33

If that is the number of legal positions, it is nevertheless very small compared to the number of ways of arriving at those positions from the start of a game.

I'm not quite sure what your argument is here, but in the evaluation of a situation on the road, or of the board in a game of chess, history doesn't matter.

#53 johnny yuma

johnny yuma
  • Member

  • 928 posts
  • Joined: May 07

Posted 04 October 2012 - 03:46

Taking evasive action - be it brake, accelerate or turn - to a clear space even off road shouldn't be too hard. As long as you map the surroundings correctly, the AI part is easy. It's a sensory problem. It gets hard when you have to choose between two accidents, though.

Lets hope the mapping is better than APPLES recent navigation efforts !
Oh and the mapping would need constant (continuous?) updating.Price of petrol will probably kill cars before this science fiction tugfest climaxes.

#54 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 04 October 2012 - 03:53

If that is the number of legal positions, it is nevertheless very small compared to the number of ways of arriving at those positions from the start of a game.

Yes, it is. The number of ways at arriving at those positions is about 10^120, easily exceeding the number of atoms in the universe by a factor with quite a few zeros.

In any case, I doubt that algorithms used to develop the best chess player in the world are going to be applicable to this exercise. Chess algorithms are pretty dumb from an AI standpoint, backgammon playing computers IMO are much more interesting (though probably still not very applicable). The only commonality that I see is that there do indeed exist computer algorithms out there that can beat any humans even in endeavors where you clearly can't calculate through every possible situation.

Edited by Dmitriy_Guller, 04 October 2012 - 03:54.


#55 packapoo

packapoo
  • Member

  • 731 posts
  • Joined: May 08

Posted 04 October 2012 - 06:34

Hope it doesn't sink when they hit it.
Then again....................

#56 Bloggsworth

Bloggsworth
  • Member

  • 7,469 posts
  • Joined: April 07

Posted 04 October 2012 - 09:01

I'm trying to think of a polite way to mock this, but I can't think of any. So I'll have to refrain from doing it. However, there are at least 10^43 legal positions in chess. To put it in terms easier to understand, it's at least 10,000,000,000,000,000,000,000,000,000,000,000,000,000,000. It might take a while to look up all of them.


Which is precisely what computers are brilliant at, and why Deeper Blue is so big! Add to which, Kasperov accused IBM of cheating and demanded a rematch - IBM dismantled the computer..
If you are worried about the possible number of chess moves available, how many scenarios would confront a computer driving through the Alps, or on a carnival day in Rio? Incalculable of course, so decision making is required; now computers are brilliant at deciding between known variables, not so hot when confronted by the unknown - Try feeding Jabberwoky into Word and see how many red and green underlines you get - Computers don't do originality, a cross-country drive in England, or anywhere else for that matter, may need it.

I am not decrying computers, they are one of the most useful tools ever invented, but they are tools, they crunch numbers, they choose between 0 & 1, they don't think, they can't think, and if we ask them to we are misunderstanding their purpose. "We" invented a machine that is extraordinarily good at doing one thing, yet we chose to ask them to do something else, something which they can't do; they is not a sentient beings, they cannot look at a hillside and come up with:

I wandered lonely as a cloud
That floats on high o'er vales and hills,
When all at once I saw a crowd,
A host, of golden daffodils;
Beside the lake, beneath the trees,
Fluttering and dancing in the breeze.


They can't look at a sunrise and an old warship and come up with this:

Posted Image

Mind you, I'd be frightened for humanity if they ever could...

Edited by Bloggsworth, 04 October 2012 - 09:04.


#57 Ross Stonefeld

Ross Stonefeld
  • Member

  • 57,279 posts
  • Joined: August 99

Posted 04 October 2012 - 11:26

Using an anecdote of your above average avoidance skills is not a mark against driverless cars. The example you gave where the computer loses will be overwhelmed by the cases where the machine gets it right. Because it's (almost)always paying attention, doesn't panic, isn't under the influence, etc.



#58 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 04 October 2012 - 12:46

Which is precisely what computers are brilliant at, and why Deeper Blue is so big!

You spectacularly fail to appreciate the impossibility of running through 10,000,000,000,000,000,000,000,000,000,000,000,000,000,000 combinations for a computer. The fastest computer in the world currently does 16 quadrillion calculations per second. Assuming that you need one operation to analyze each board position (an impossible feat of programming genius), it would only take you about 20,000,000,000,000,000,000 years to run through all the combinations. Might take a while. My guess is that chess matches with Kasparov didn't take that long, so there must've been a more clever algorithm in place that did not rely on brute force.

#59 Magoo

Magoo
  • Member

  • 2,480 posts
  • Joined: October 10

Posted 04 October 2012 - 14:31

...for oft when on my couch I lie
in vacant or in pensive mood
they flash upon that inward eye
which is the bliss of solitude
and then my heart with pleasure fills
and dances with the daffodils.





Advertisement

#60 munks

munks
  • Member

  • 344 posts
  • Joined: January 03

Posted 04 October 2012 - 14:44

My suggestion is to stop digging on that particular argument Bloggsworth. The "middle game" in chess cannot be computed with lookup tables, and may never be. You simply can't store the lookup table. (That's not to say chess can't be solved ... I believe quantum computers may someday be able to do this, but it would be through a depth-first search where you don't need to store every position. EDIT - thinking through this again, IIRC quantum computers would essentially be able to do this as a breadth-first search, I *think*, but it's through the magic of quantum theory that every position is "stored".)

And you're being remarkably unimaginative about what computers could potentially do. Everybody here agrees that there are going to be unusual situations that a human can still handle better then a computer. But with the right sensors, yes they could simultaneously avoid an accident in front of them without causing one behind them. In fact I would think that would be a basic standard which they should need to pass.

But for those that still want to drive, you can get some of the best of both worlds (in fact, some new cars already do emergency braking for you before your brain has even processed the visuals). And for those that don't want to drive, or that we don't want driving (texting teenagers), I'm pretty sure that the computers can already do a better job overall.

Edited by munks, 04 October 2012 - 14:46.


#61 Bloggsworth

Bloggsworth
  • Member

  • 7,469 posts
  • Joined: April 07

Posted 04 October 2012 - 16:26

You spectacularly fail to appreciate the impossibility of running through 10,000,000,000,000,000,000,000,000,000,000,000,000,000,000 combinations for a computer. The fastest computer in the world currently does 16 quadrillion calculations per second. Assuming that you need one operation to analyze each board position (an impossible feat of programming genius), it would only take you about 20,000,000,000,000,000,000 years to run through all the combinations. Might take a while. My guess is that chess matches with Kasparov didn't take that long, so there must've been a more clever algorithm in place that did not rely on brute force.


So why did IBM dismantle the computer when Kasparov wanted a rematch? Something to hide?

No I don't - Are you saying that navigating an unknown road through the Welsh, or Rocky mountains in variable light, weather and season is less complicated than chess, and will require a computer smaller than an American fridge and weighing less than 2 extra passengers.

Are we to understand that the appurtenances sprouting from the cars on test at the moment will also be on the commercially available models? If so, what an opportunity for vandalism, how easily damaged by a carelessly swung ladder, how easily knocked of by a low hanging branch on a country lane - How long before we hear tell of cars stopped by the wrong kind of snow freezing the rotatey bit on the roof, or platinum particles from catalysers thrown up from the road fogging reception?

Edited by Bloggsworth, 04 October 2012 - 16:27.


#62 Ross Stonefeld

Ross Stonefeld
  • Member

  • 57,279 posts
  • Joined: August 99

Posted 04 October 2012 - 21:38

You know Deep Blue was a while ago right? And that they taught it some of Kasparov's games? It wasn't just doing blind calculations.

#63 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 04 October 2012 - 23:05

Yes, that's another good point. Deep Blue was more than 15 years ago. Nowadays there are chess programs running on regular PCs that would've most likely trounced Deep Blue.

#64 gruntguru

gruntguru
  • Member

  • 5,273 posts
  • Joined: January 09

Posted 05 October 2012 - 01:20

...for oft when on my couch I lie
in vacant or in pensive mood
they flash upon that inward eye
which is the bliss of solitude
and then my heart with pleasure fills
and dances with the daffodils.

A computer composed that - right?

#65 gruntguru

gruntguru
  • Member

  • 5,273 posts
  • Joined: January 09

Posted 05 October 2012 - 01:23

I'm not quite sure what your argument is here, but in the evaluation of a situation on the road, or of the board in a game of chess, history doesn't matter.

Some ability to predict/forecast is valuable - in chess and driverless cars.

#66 Kalmake

Kalmake
  • Member

  • 455 posts
  • Joined: November 07

Posted 05 October 2012 - 02:12

Lets hope the mapping is better than APPLES recent navigation efforts !
Oh and the mapping would need constant (continuous?) updating.Price of petrol will probably kill cars before this science fiction tugfest climaxes.


Heh, not that kind of mapping. The car would use it's own cameras/radar in real time to form a map of where all the obstacles etc. are.

#67 gruntguru

gruntguru
  • Member

  • 5,273 posts
  • Joined: January 09

Posted 05 October 2012 - 02:16

Bloggsworth you need to look closer at the Google car. Intro to Google car. (1 year ago) I have no doubt that it is already a safer option than riding with an expert driver (like yourself).

Your assertion that creativity is an important factor in favour of human drivers is dubious for at least the following reasons:
- Creativity takes too long to be useful in accident avoidance. It is reflexes, awareness of surroundings and awareness of vehicle dynamic limitations that matter.
- Unconventional escape routes eg off-road are more accessible to the driverless car which is constantly aware of terrain in all directions and for some distance.
- A driverless car can easily be programmed to be constantly aware of all possible escape routes (forecasting) before an emergency situation even arises.

#68 gruntguru

gruntguru
  • Member

  • 5,273 posts
  • Joined: January 09

Posted 05 October 2012 - 02:20

Price of petrol will probably kill cars before this science fiction tugfest climaxes.

At the current rate of development the technology will be commercialised a lot sooner than you think. I'm guessing less than 10 years.

Oh - and I seem to remember you taking a different view on the life expectancy of the IC automobile on another thread.

#69 pugfan

pugfan
  • Member

  • 174 posts
  • Joined: August 09

Posted 05 October 2012 - 02:40

Bloggsworth you need to look closer at the Google car. Intro to Google car. (1 year ago) I have no doubt that it is already a safer option than riding with an expert driver (like yourself).

Your assertion that creativity is an important factor in favour of human drivers is dubious for at least the following reasons:
- Creativity takes too long to be useful in accident avoidance. It is reflexes, awareness of surroundings and awareness of vehicle dynamic limitations that matter.
- Unconventional escape routes eg off-road are more accessible to the driverless car which is constantly aware of terrain in all directions and for some distance.
- A driverless car can easily be programmed to be constantly aware of all possible escape routes (forecasting) before an emergency situation even arises.


Agree completely. With sufficient computing power, a computer could generate thousands of scenarios per millisecond and evaluate the best course of action, something a human would be completely incapable of.

Driver skill is already augmented anyway with stability control and ABS. These are devices to help the car follow the driver's strategy for safely negotiating the highways. I don't see it as a big step to let the computer come up with the strategy for collision avoidance as well and takeover control if it detects that the driver's strategy is too far from optimal.

Look at it from the flipside, we're quite happy to let the average punter (and I do mean average) onto the road in ~2 tonnes of metal with very limited training in situational awareness and collision avoidance and that is somehow much better?

Edited by pugfan, 05 October 2012 - 02:47.


#70 Kalmake

Kalmake
  • Member

  • 455 posts
  • Joined: November 07

Posted 05 October 2012 - 03:01

Kasparov played poorly in the match after the game where he decided IBM was cheating. Had he kept his head straight, he was still the favorite to win. Having completed their marketing exercise IBM had no reason to agree to a rematch.

Actually Deep Blue calculated positions faster than your desktop computer today. Basically it just had so many chips and was purpose built for the task. But modern chess programs are smarter in evaluating which lines are worth calculating. They get deeper with much less work. Home computer beating world champion happened several years ago.

#71 bigleagueslider

bigleagueslider
  • Member

  • 855 posts
  • Joined: March 11

Posted 06 October 2012 - 04:33

The idea of autonomously piloted automobiles on public highways itself does not make me nervous. But the fact that the drooling morons in the California state government are overseeing the project scares the bejeezus out of me. These same idiots oversee the California public school system, which has a science/math high school graduation proficiency level below the 7th grade, and a drop-out rate close to 50%.

Of course, all commercial aircraft mostly fly autonomously, and does that present a problem?

#72 Felix

Felix
  • Member

  • 707 posts
  • Joined: December 03

Posted 06 October 2012 - 07:31

http://plus.autospor...nnected-future/



#73 mariner

mariner
  • Member

  • 1,373 posts
  • Joined: January 07

Posted 06 October 2012 - 08:07

There is one huge potential payoff of autonomous ( driverless) cars which the GM Shanghai tech. centre has already identified some time ago - reducing global warming.

If you want a auto. car to be safe it has to be able to avoid 99.9999etc % of collisions i.e what most of this discussion has been about.

If you know it can't hit anything there is absolutely no need for any passive safety features.

So vehicle weight can be massively reduced with big savings on fuel ( of any type ) so global warming is cut.

That is the basis of the city car proposed and demo'ed by GM Shanghai a while back. It s for city streets and only where 100% are auto. but its small simply because it needs no crumple zones etc. With more and more of the developing world moving to supercities it may be the only way " car" sales can be kept up in the future.

To anybody who says " you can't get rid of the passive safety rules for auto cars" the answer is simple " airliners don't have ejector seats , they rely on not having a crash at all".

BTW the concept of driverless cars is far from new , various mfrs have demo'ed them for years . The big difference is the autonomous part. Whether such cars will be allowed to be truly autonomous is an interesting question. I can see an almost relentless logic of gov't that they must be subject to central tracking and control. That would help ease traffic flow. It would also ( probably and sadly) allow govts to decide where and when you can go in your car.

Maybe it will end up like airspace - split into controlled and uncontrolled road space.

#74 mariner

mariner
  • Member

  • 1,373 posts
  • Joined: January 07

Posted 06 October 2012 - 09:55

On a lighter note the wonders of internet logic

http://www.youtube.c...t=LPUyujUXL3_vc

The average driver has one accident evry 165K miles and one Google car has done 300K without an accident so its now safer than the average driver.

On the subjsct of data accuracy I love this analysis of a "science" story which ran in several newspapers around the world

http://www.bbc.co.uk...gazine-19755695

#75 Bloggsworth

Bloggsworth
  • Member

  • 7,469 posts
  • Joined: April 07

Posted 06 October 2012 - 13:50

On a lighter note the wonders of internet logic

http://www.youtube.c...t=LPUyujUXL3_vc

The average driver has one accident evry 165K miles and one Google car has done 300K without an accident so its now safer than the average driver.

On the subjsct of data accuracy I love this analysis of a "science" story which ran in several newspapers around the world

http://www.bbc.co.uk...gazine-19755695


From a sample of one you project future accident rates! I know someone who has been driving for nearly 50 years without hitting anything, my wife, therefore people can drive for 50 years...

Driverless cars will work very well as long as all vehicles on the road are driverless, the danger will come from the mixing of autonomous and not.

As for computers sequencing scenarios faster than a human, they can only do it for scenarios pre-programmed into their memory banks; the human brain can deduce things given far less information than a computer - I give you the television show "Name that tune" on which contestants regularly name a tune correctly given only a 2 or 3 notes, and do it far, far, more often than could be put down to chance. How, because they can make correlations and likelyhoods which no computer could given that there are 1,000s of tunes which start with the same 2 notes in the same key; the contestant will be asked to name a tune that it is possible they know, something which is of no use to a computer, as the computer knows them all equally well and has no means of weighting the possible one choice out of hundreds - Program the weighting? So the programmer is a heavy-metal fan and weights his preferences, no no, we can't have that, give the job to a Motown fan, but what if the answer is New York New York? This is the sort of task an autonomous car will have to deal with in mixed traffic and on roads it has no baseline knowledge of.

When I was nobbut a lad I had a friend who thought his car-driving uncle was a genius, he had to be didn't he, he remembered everywhere in England where you were supposed to change gear.

#76 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 06 October 2012 - 19:57

As has been the pattern throughout this thread, you're failing to take into account three important things about computers:

1) Computer AI logic has long since evolved from "if-then" instructions.
2) Computers don't have to approach problems the same way humans do. As a silly example, for computers to drive a car, you don't need to create a robot with hands, legs, and eyes that is programmed to push the pedals and turn the steering wheel with sufficiently quick reaction time.
3) Computer hardware and computer science are still evolving at a breakneck pace. What doesn't seem possible today may be the new reality 10 years from now.

#77 Bloggsworth

Bloggsworth
  • Member

  • 7,469 posts
  • Joined: April 07

Posted 06 October 2012 - 21:06

As has been the pattern throughout this thread, you're failing to take into account three important things about computers:

1) Computer AI logic has long since evolved from "if-then" instructions.
2) Computers don't have to approach problems the same way humans do. As a silly example, for computers to drive a car, you don't need to create a robot with hands, legs, and eyes that is programmed to push the pedals and turn the steering wheel with sufficiently quick reaction time.
3) Computer hardware and computer science are still evolving at a breakneck pace. What doesn't seem possible today may be the new reality 10 years from now.



But humans aren't - I'm just saying, you mix them at your peril.

#78 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 06 October 2012 - 21:24

I'm not sure how that addresses anything that I've said.

#79 Ross Stonefeld

Ross Stonefeld
  • Member

  • 57,279 posts
  • Joined: August 99

Posted 07 October 2012 - 09:02

To anybody who says " you can't get rid of the passive safety rules for auto cars" the answer is simple " airliners don't have ejector seats , they rely on not having a crash at all".


Yet the military does. It comes down to the value they place on the occupants.

Advertisement

#80 Bloggsworth

Bloggsworth
  • Member

  • 7,469 posts
  • Joined: April 07

Posted 07 October 2012 - 09:15

Yet the military does. It comes down to the value they place on the occupants.


And the fact that when someone is shooting at them it might be wise to leave the scene...

#81 Bloggsworth

Bloggsworth
  • Member

  • 7,469 posts
  • Joined: April 07

Posted 07 October 2012 - 09:20

I'm not sure how that addresses anything that I've said.



Computers may well have a different approach, but different humans have different approaches at different times and in different situations. Trains run on well defined tracks, no possibility that they can take a different route, yet irrational people still manage to wander into their path; you cannot consider the apparent efficiency of the computer in isolation, it is part of a world full of irrational people, semi-rational animals and falling objects with no pretence of rationality at all...

#82 carlt

carlt
  • Member

  • 1,053 posts
  • Joined: June 09

Posted 07 October 2012 - 10:20

I'm absolutely certain a lot of the cars in our school car park are already 'driverless'

#83 mariner

mariner
  • Member

  • 1,373 posts
  • Joined: January 07

Posted 07 October 2012 - 11:17

I think a common theme is emerging - driverless cars can't predict and avoid irrational human driving behaviour but otherwise are pretty good so the end logic is inevitable

- human driven cars wil be banned from zones where auto cars operate - that way the crash avoidance and interactive manoveouring power of the computer cars is maixmised


-" you can't drive here - its too dangerous to have human's steering cars".



#84 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 07 October 2012 - 14:55

Yet the military does. It comes down to the value they place on the occupants.

The military planes are expected to fail in flight a bit more frequently, for some reason.

#85 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 07 October 2012 - 15:00

Computers may well have a different approach, but different humans have different approaches at different times and in different situations. Trains run on well defined tracks, no possibility that they can take a different route, yet irrational people still manage to wander into their path; you cannot consider the apparent efficiency of the computer in isolation, it is part of a world full of irrational people, semi-rational animals and falling objects with no pretence of rationality at all...

And what makes you think that computer cars can't evade humans? It seems like with quick reaction times and much more expansive awareness of surroundings, they should be able to do it much better than humans. Sure, if a human sets out on a task to crash into a computer car, he would probably succeed, but this is getting silly.

#86 Ross Stonefeld

Ross Stonefeld
  • Member

  • 57,279 posts
  • Joined: August 99

Posted 07 October 2012 - 15:36

The military planes are expected to fail in flight a bit more frequently, for some reason.



But that's not why airliners don't have ejection seats/parachutes/magic umbrellas. Nor is it a reason to support a theory that driverless road cars will need fewer safety features.

#87 Bloggsworth

Bloggsworth
  • Member

  • 7,469 posts
  • Joined: April 07

Posted 07 October 2012 - 16:05

And what makes you think that computer cars can't evade humans? It seems like with quick reaction times and much more expansive awareness of surroundings, they should be able to do it much better than humans. Sure, if a human sets out on a task to crash into a computer car, he would probably succeed, but this is getting silly.


Computers will be brilliant at evading pedestrians........... if the pedestrians behave in exactly the way that the computer expects, but then, panicky humans frequently do stupid things, like trying to outrun the car rather than stepping sideways onto the pavement. Experienced drivers can read the body language, can see that it is an old woman with her shopping who is rooted to the spot, can tell whether it is an athletic 16 year-old who's going to make a run for it. Will the computer see moving feet beneath a car and deduce that as it can see that there is no corresponding head there may well be a child about to run out into the road, or see a bouncing ball and understand that it too may be followed by a small child? Will it, when driving down a country lane in high winds see a tree at the side of the road split and the detaching part start falling towards the road and stop, or will it not register as anything other than the general movement around it? Will it, in the Alps, see a landslide starting high above and immediately stop?

Nobody can offer proof that these cars are so much better, they just say that because a few Google cars under experimental conditions have been safe all autonomous cars will be safe, well that's an extrapolation too far for me, proof of concept needs 10,000 of the things in the US, Europe, the Sahara region, the plains and hills of India, the high veldt for a few years before any final conclusions can be drawn. Pilot error accounts for 49 percent of all aviation accidents, and approximately 83 percent of all private aircraft accidents, and planes operate in a far less fraught environment than cars. Unfortunately I doubt I'll be alive when there are significant number of these devices on the road, but a small point, these vehicles are being developed in the US, where nobody walks...

And then there's the environment, all the mining for the metals used in the circuits and the batteries, the extra carbon emissions from the inefficient Chinese power-stations polluting our skies, because don't kid yourself that the elecronics and batteries will be made anywhere with a conscience. There was a report this last week demonstrating that electric cars were more damaging to the planet than conventional IC cars, and they will only become less so if the whole world has really green means of producing electricity - Like that's going to happen in parts of asia any time soon.

None of this is as simple as "Computers are more efficient than people at some things..."

Edited by Bloggsworth, 07 October 2012 - 16:30.


#88 Tony Matthews

Tony Matthews
  • Member

  • 17,499 posts
  • Joined: September 08

Posted 07 October 2012 - 17:11

Computers will be brilliant at evading pedestrians........... if the pedestrians behave in exactly the way that the computer expects, but then, panicky humans frequently do stupid things, like trying to outrun the car rather than stepping sideways onto the pavement. Experienced drivers can read the body language, can see that it is an old woman with her shopping who is rooted to the spot, can tell whether it is an athletic 16 year-old who's going to make a run for it. Will the computer see moving feet beneath a car and deduce that as it can see that there is no corresponding head there may well be a child about to run out into the road, or see a bouncing ball and understand that it too may be followed by a small child? Will it, when driving down a country lane in high winds see a tree at the side of the road split and the detaching part start falling towards the road and stop, or will it not register as anything other than the general movement around it? Will it, in the Alps, see a landslide starting high above and immediately stop?

All these things are perfectly possible now, and with the rate that computing power and sensor sensitivity is increasing, things can only get better. 'Experienced' drivers are not necessarily good drivers, or always concentrating, or looking in the right direction. Even allowing for excellent peripheral vision, it is impossible to see what is happening for 360° horizontally, and 180° vertically at the same time, all the time.

#89 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 07 October 2012 - 17:14

So what if humans are unpredictable? We need predictability because our reaction times are slow. The faster your reaction times, the more you can cope with unpredictability.

#90 Magoo

Magoo
  • Member

  • 2,480 posts
  • Joined: October 10

Posted 07 October 2012 - 18:31

A major problem with human drivers is lack of focus. Much of the time we drive far below our actual ability.


New study proves: your driving stinks | Mac's Motor City Garage

#91 Greg Locock

Greg Locock
  • Member

  • 4,510 posts
  • Joined: March 03

Posted 07 October 2012 - 21:40

The "Name that tune" argument is specious. If thousands of tunes start with the same two notes and the computer selects a different answer to a human, then it has chosen a correct but different answer. if you mean that a human being can recognise a likely candidate from, for example, the very first riff of "Have You Seen Your Mother", yes I agree that computers will be unlikely to manage that. On the other hand, http://en.wikipedia....wiki/Soundhound has capabilities that are equally impressive, although I'd admit it couldn't recognise the opera I was listening to last night (neither could I).

Edited by Greg Locock, 07 October 2012 - 22:05.


#92 Ross Stonefeld

Ross Stonefeld
  • Member

  • 57,279 posts
  • Joined: August 99

Posted 07 October 2012 - 22:00

Computers will be brilliant at evading pedestrians........... if the pedestrians behave in exactly the way that the computer expects, but then, panicky humans frequently do stupid things, like trying to outrun the car rather than stepping sideways onto the pavement. Experienced drivers can read the body language, can see that it is an old woman with her shopping who is rooted to the spot, can tell whether it is an athletic 16 year-old who's going to make a run for it. Will the computer see moving feet beneath a car and deduce that as it can see that there is no corresponding head there may well be a child about to run out into the road, or see a bouncing ball and understand that it too may be followed by a small child? Will it, when driving down a country lane in high winds see a tree at the side of the road split and the detaching part start falling towards the road and stop, or will it not register as anything other than the general movement around it? Will it, in the Alps, see a landslide starting high above and immediately stop?

Nobody can offer proof that these cars are so much better, they just say that because a few Google cars under experimental conditions have been safe all autonomous cars will be safe, well that's an extrapolation too far for me, proof of concept needs 10,000 of the things in the US, Europe, the Sahara region, the plains and hills of India, the high veldt for a few years before any final conclusions can be drawn. Pilot error accounts for 49 percent of all aviation accidents, and approximately 83 percent of all private aircraft accidents, and planes operate in a far less fraught environment than cars. Unfortunately I doubt I'll be alive when there are significant number of these devices on the road, but a small point, these vehicles are being developed in the US, where nobody walks...

And then there's the environment, all the mining for the metals used in the circuits and the batteries, the extra carbon emissions from the inefficient Chinese power-stations polluting our skies, because don't kid yourself that the elecronics and batteries will be made anywhere with a conscience. There was a report this last week demonstrating that electric cars were more damaging to the planet than conventional IC cars, and they will only become less so if the whole world has really green means of producing electricity - Like that's going to happen in parts of asia any time soon.

None of this is as simple as "Computers are more efficient than people at some things..."


All these examples are extreme rather than commonplace. 90% of people won't notice these things. The computer is to help the idiots not the clairvoyants like yourself.

#93 Tenmantaylor

Tenmantaylor
  • Member

  • 8,317 posts
  • Joined: July 01

Posted 07 October 2012 - 22:30

A driverless car wouldn't have kicked Webber's back door in like Grosjean today.

#94 pugfan

pugfan
  • Member

  • 174 posts
  • Joined: August 09

Posted 07 October 2012 - 23:51

Computers will be brilliant at evading pedestrians........... if the pedestrians behave in exactly the way that the computer expects


I think you fundamentally misunderstand how computers can be used. The computer need not necessarily expect the pedestrian to behave in any given way, in fact it probably doesn't even need to determine that it's a pedestrian. Hundreds of times a second it can evaluate the pedestrians position and trajectory as well as every other object in it's field of view and determine the best course of action to avoid everything.

With reference to the start of the weekends F1 race, there is simply no way a human, no matter how skilled can do the same thing.

#95 bigleagueslider

bigleagueslider
  • Member

  • 855 posts
  • Joined: March 11

Posted 08 October 2012 - 00:45

I think we will see fairly rapid implementation of crash avoidance features on new cars, but I doubt we'll ever see a truly autonomous car on public highways. Governments are too averse to potential financial liability from accidents. As long as there is a human "driver" in the vehicle, regardless of how little actual control they exert, the government will have someone to assign blame to for any accident.

#96 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 08 October 2012 - 00:57

I think we will see fairly rapid implementation of crash avoidance features on new cars, but I doubt we'll ever see a truly autonomous car on public highways. Governments are too averse to potential financial liability from accidents. As long as there is a human "driver" in the vehicle, regardless of how little actual control they exert, the government will have someone to assign blame to for any accident.

I wouldn't be quite so cynical or pessimistic. Apart from safety, another benefit of driverless cars is that they would vastly increase the capacity of highways. Imagine Talladega draft trains on highways with remote chance of accidents. That's a lot of paving dollars, gas, and time saved. And, ultimately, the government itself defines who incurs liability for what by passing laws.

#97 bigleagueslider

bigleagueslider
  • Member

  • 855 posts
  • Joined: March 11

Posted 08 October 2012 - 02:15

I wouldn't be quite so cynical or pessimistic. Apart from safety, another benefit of driverless cars is that they would vastly increase the capacity of highways. Imagine Talladega draft trains on highways with remote chance of accidents. That's a lot of paving dollars, gas, and time saved. And, ultimately, the government itself defines who incurs liability for what by passing laws.


Another issue with a true driverless car on public highways would not be the driverless cars themselves, it would be the massive number of older vehicles without such technology sharing the same highway. In the US alone, there are still millions of registered vehicles at least 20 years old.


#98 desmo

desmo
  • Tech Forum Host

  • 12,987 posts
  • Joined: January 00

Posted 08 October 2012 - 02:36

It would be a fine excuse to force a turnover of old rolling stock. Stimulus maximus.

I hate the idea by the way, but the transition to truly self-piloting cars will probably have to mandate highways (or at the very least lanes) where only such vehicles are allowed.

#99 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,043 posts
  • Joined: July 01

Posted 08 October 2012 - 03:32

Yes, definitely. Having dumb cars trundling around in an autopilot lane would at the very least defeat the whole concept.

Advertisement

#100 mariner

mariner
  • Member

  • 1,373 posts
  • Joined: January 07

Posted 08 October 2012 - 07:53

This is partly an excuse to post some great Argentinian saloon car racing video ( thanks repco von brabham) but as Google is mega, mega rich how about it fits its driverless gear onto each of these cars and shows a similar race can be done by computers.

http://www.youtube.c...feature=related

Its not a sarcastic or flippant suggestion - raciing has improved the breed in the past and if Google could pull it of it would be great publicity for the concept.

Thre is plenty of space in a saloon to fit all the gear ( remember the first in-car TV from Australian sedans).

The question of course is how will one car beat another one?? Well, fewer kids are keen on cars it seems but let teams of digigeek teens loose on the software to "tune " it to win but not crash and you may have a whole new generation of race competitors - and no worried about injuries parents.