Donate to Remove ads

Got a credit card? use our Credit Card & Finance Calculators

Thanks to monabri,Dod101,Anonymous,Gersemi,robbelg, for Donating to support the site

Musk endeavours

The Big Picture Place
onthemove
2 Lemon pips
Posts: 107
Joined: June 24th, 2017, 4:03 pm
Has thanked: 140 times
Been thanked: 84 times

Re: Musk endeavours

#250978

Postby onthemove » September 11th, 2019, 11:26 am

tjh290633 wrote:
odysseus2000 wrote:Imho the Ap is not far away from being able to navigate UK roads, but to be useable it will also need to learn how to deal with roundabouts and junctions, the former unknown in the US and the latter very different.

If you want to try it out, try sending two autonomously driven cars along our lane, one from each end, and see how they sort themselves out when they meet in the middle. They may well each have a queue of traffic behind them, to complicate matters, and there are always a line of parked cars in the narrowest section.

Manually driven cars are able to negotiate the lane with little problem, until an oversized truck (>6'6" wide) attempts it. They often have to reverse out and that can take some time.

I wouldn't fancy having a driverless taxi to take me home.

TJH


Like others have mentioned, autonomous cars could have a 5g network to resolve these kinds of situations, although personally, I think any autonomous car will need to be able to resolve these situations of their own accord without any recourse to inter-vehicle communication - which I don't see any reason they couldn't.

Autonomous cars will have full 360 degree vision, so driving backwards shouldn't be any kind of technical hurdle.

There are plenty of algorithms in software engineering related to deadlock resolution, etc, so it's not like there aren't plenty of ideas... networks and concurrent systems have been grappling with (the software equivalent of) single track roads for decades now - so it's just a matter of choosing and implementing the algorithm people think is the most appropriate or acceptable (that's not necessarily the most efficient - people don't always instinctively like the theoretically most ideal solution, and may be happier with a suboptimal algorithm for sorting out meeting in the middle of single track lanes)

I think the biggest challenge for autonomous cars will be the ambiguity that often arises between whether a car is parked or dropping someone off, or whether it is stationary because it is queuing.

At all times the autonomous car should at least be 'safe' and won't attempt anything that causes it to collide, but in some of these ambiguous situations there is a high likelihood of the car ending up dithering because its got itself into a situation that it misinterpreted and ended up in the wrong lane or wrong position, or ended up waiting and waiting and waiting behind a car that was actually parked, rather than overtaking it.

The other situation I think they may struggle with is the same issue for human drivers - many junctions only indicate which lane is for which direction using road markings, in lane, and only at the last minute... on a clear road with clear visibility of the lanes, that's fine... but in heavy traffic with the car in front obscuring the lane in front, it's very easy to end up in the wrong lane as you approach a junction.

I guess an autonomous car should be better at continuously monitoring the lane even as a sliding window... unlike a person, the car doesn't need to keep looking up, so it could I guess generate a continuous picture of the road / lane and fully extract any markings from it...

… though often in heavy traffing, you'll already be in the wrong lane / queue before you see the marking telling you that you should be in the adjacent lane, which is now packed with cars queuing.

I think an autonomous car might find it difficult to 'push' into such a busy lane.

These aren't necessarily dangerous problems - in all cases the car's basic safety programming should ensure it doesn't physically try to drive into something else, and should also prevent it shooting out into the path of another moving vehicle... but that's part of the problem... that level of caution will make it difficult to be 'assertive' to get itself out of a situation that it misinterpreted, so may end up stranded for some time if the surrounding vehicles are being driven by, well, let's just say 'certain types'.

I agree with Odysseus2000 regarding road differences.

It's not that the differences in the UK are insurmountable - they're certainly not insurmountable. It's just that the UK is a small island globally with it's own quirky rules (like driving on the left), and an unwillingness to converge with others / determination to keep its own separate rules, so as to protect its 'sovereignty'...

.. that's going to mean developing cars for the UK will always be at least one step behind autonomous cars developed for other markets, assuming manufacturers think the extra investment is justified at all... though I think in most cases, once developed for other markets, the development costs for the UK market will be smaller than completely starting from scratch so the effort will be made... it will just be one or more steps behind, and have an additional cost... developing for the UK quirks won't come for free.

So we'll be paying a premium for autonomous cars in the UK as a result of our determination to be different and set our own rules, our insistence that we develop our own non-standard road signs different from our neighbours, etc

ReformedCharacter
Lemon Slice
Posts: 961
Joined: November 4th, 2016, 11:12 am
Has thanked: 463 times
Been thanked: 236 times

Re: Musk endeavours

#251011

Postby ReformedCharacter » September 11th, 2019, 12:17 pm

onthemove wrote:Like others have mentioned, autonomous cars could have a 5g network to resolve these kinds of situations, although personally, I think any autonomous car will need to be able to resolve these situations of their own accord without any recourse to inter-vehicle communication - which I don't see any reason they couldn't.

Autonomous cars will have full 360 degree vision, so driving backwards shouldn't be any kind of technical hurdle.

There are plenty of algorithms in software engineering related to deadlock resolution, etc, so it's not like there aren't plenty of ideas... networks and concurrent systems have been grappling with (the software equivalent of) single track roads for decades now - so it's just a matter of choosing and implementing the algorithm people think is the most appropriate or acceptable (that's not necessarily the most efficient - people don't always instinctively like the theoretically most ideal solution, and may be happier with a suboptimal algorithm for sorting out meeting in the middle of single track lanes)


I agree with your comments but I lack your confidence in the resolution of common road problems without inter-vehicle communication. Using the example of the lane problem... My local lane is single track in parts, it has two legs at roughly 90 degrees and is 'blind' in that radar or lidar would be unable to see the approach of another vehicle on the other leg. To complicate matters there are also junctions on each leg and the end of one leg is a T-junction with traffic passing at perhaps 40mph. and there aren't just two vehicles because there are more arriving because it is 'rush hour'. One leg has limited capacity because it terminates in a T-junction... Gridlock! For which the solution is mutual co-operation which requires communication. And that is one of the easier problems.

As to deadlock resolution, which I admit I do not know much about (quick look on Wikipedia) does it not generally require an operating system which can unlock the deadlock, the operating system being a proxy for inter-vehicle communication in this case?

RC

BobbyD
Lemon Quarter
Posts: 4892
Joined: January 22nd, 2017, 2:29 pm
Has thanked: 70 times
Been thanked: 230 times

Re: Musk endeavours

#251021

Postby BobbyD » September 11th, 2019, 12:36 pm

onthemove wrote:Like others have mentioned, autonomous cars could have a 5g network to resolve these kinds of situations, although personally, I think any autonomous car will need to be able to resolve these situations of their own accord without any recourse to inter-vehicle communication - which I don't see any reason they couldn't.

Autonomous cars will have full 360 degree vision, so driving backwards shouldn't be any kind of technical hurdle.


Cars should be able to do the driving and navigating bits without communication, but being aware that there are 16 people queued behind the car coming the other way would be quite useful. Another time this might be quite useful would be 'hey Mr. Tesla, I'm about to change lane because there's a stationary fire truck in this one, try not to pile in the back of it'. Driving is a game of imperfect information, making things which you know known to others and being able to see what they see should make life smoother. 'Oh he wasn't being a ****, some idiot has parked in the passing spot behind him...'.

onthemove
2 Lemon pips
Posts: 107
Joined: June 24th, 2017, 4:03 pm
Has thanked: 140 times
Been thanked: 84 times

Re: Musk endeavours

#251033

Postby onthemove » September 11th, 2019, 12:57 pm

ReformedCharacter wrote:I agree with your comments but I lack your confidence in the resolution of common road problems without inter-vehicle communication. Using the example of the lane problem... My local lane is single track in parts, it has two legs at roughly 90 degrees and is 'blind' in that radar or lidar would be unable to see the approach of another vehicle on the other leg. To complicate matters there are also junctions on each leg and the end of one leg is a T-junction with traffic passing at perhaps 40mph. and there aren't just two vehicles because there are more arriving because it is 'rush hour'. One leg has limited capacity because it terminates in a T-junction... Gridlock! For which the solution is mutual co-operation which requires communication. And that is one of the easier problems.

As to deadlock resolution, which I admit I do not know much about (quick look on Wikipedia) does it not generally require an operating system which can unlock the deadlock, the operating system being a proxy for inter-vehicle communication in this case?

RC


Even with inter-vehicle communication, the same resolution issue arises. Specifically, which vehicle is the master? If a group of vehicles come into proximity such that they can communicate, you still have the problem of establishing a single coordinator from them. And there are plenty of ad-hoc network protocols that have been designed to deal with just such situations (nodes being added to a network without any being explicitly designated the master prior to being connected)

And the algorithms associated with that are not too dissimilar to how you'd resolve the traffic dead lock in the first place.

There are lots of different algorithms with different features, quirks, etc, but without some form of designated master from the off, all of them require some degree of participants willing to relent and relinquish to (an)other participant(s).

OK, I'd agree that's better to do in computer memory than by shuffling cars around on a road, but doing it in computer memory will only work if all participants are involved in the electronic negotiation.

Unfortunately, unless you completely ban all human drivers and mandate that all vehicles must use the same network communication, you're going to have participants in the traffic dead lock that will not be participants in the electronic dead lock resolution negotiation.

The key thing in dead lock resolution is that one (or more) or the participants has to relinquish. Whether through being told by an agreed arbitrator, or by 'algorithm'.

A noddy way of dead lock resolution without an arbitrator in computers is for each participant to voluntarily and unilaterally relinquish their 'lock', then wait a random time before attempting to regain the lock.

In traffic terms, the equivalent of relinquishing a 'lock' would be reversing along the road on which the dead lock has occurred, back to a point where the oncoming traffic would be able to pass. And then wait there for a random time.

Without an arbitrator, it's possible that an oncoming vehicle might do the same - reverse up to a point where two way traffic is possible, then wait before retrying to drive along the road.

With random waits either side, eventually one vehicle will make it through while the other one is waiting, thereby (eventually) resolving the dead lock allowing the other to pass.

In reality, a better algorithm is the moment that one vehicle sees the other relinquishing (reversing), it then takes the opportunity to move forwards.

Just like humans do when they meet each other on a single track road. As soon as one relinquishes and starts reversing, the other car doesn't normally then do the same! Usually on seeing an oncoming car start reversing, you'd gratefully make use of it and move forwards.

And it just needs to be the rule that if you encounter a car on a single track road reversing towards you, then you also need to reverse - whether human driven or autonomous.

In practise this is just the same as what people do... only a real ****head would see that the car in front of them is reversing on such a road, and act a complete ****k and stand their ground preventing them from reversing further. Most reasonable people would realise they are reversing for a reason, and do the same!

So the rules are not really any different from how humans deal with single track roads already.... the occasional idiots excepted who will both determinedly stand their ground, refuse to reverse, ensuring no-one gets anywhere.

This fairly simply algorithm is compatible with (largely is) how humans already drive on single track roads, easily understandable by humans, and doesn't require an inter-vehicle communication that all participants must be involved in. And it is based on quite simple rules that an autonomous vehicle would be easily able to implement as well.

So perfectly reasonable in a mix of human and autonomous vehicles without intervehicle communication.

onthemove
2 Lemon pips
Posts: 107
Joined: June 24th, 2017, 4:03 pm
Has thanked: 140 times
Been thanked: 84 times

Re: Musk endeavours

#251036

Postby onthemove » September 11th, 2019, 1:07 pm

BobbyD wrote:Cars should be able to do the driving and navigating bits without communication, but being aware that there are 16 people queued behind the car coming the other way would be quite useful. Another time this might be quite useful would be 'hey Mr. Tesla, I'm about to change lane because there's a stationary fire truck in this one, try not to pile in the back of it'.


Absolutely - if the information is there, then yes, use it.

But, I think we agree, no autonomous car should be allowed on the roads if it is unable to fully and properly function should such extra information (not independently gleaned from its own sensors) become unavailable for any reason.

Networks go down, the car in front might be human driven without such a network, or a car with such a network might develop a fault, etc.

Or worse case, if the network comes under a DoS attack, then the driver / occupant / control system should have the option to completely shut off external connections and still have the car continue to drive autonomously relying entirely on its own independent sensing.

BobbyD
Lemon Quarter
Posts: 4892
Joined: January 22nd, 2017, 2:29 pm
Has thanked: 70 times
Been thanked: 230 times

Re: Musk endeavours

#251045

Postby BobbyD » September 11th, 2019, 1:29 pm

onthemove wrote:
BobbyD wrote:Cars should be able to do the driving and navigating bits without communication, but being aware that there are 16 people queued behind the car coming the other way would be quite useful. Another time this might be quite useful would be 'hey Mr. Tesla, I'm about to change lane because there's a stationary fire truck in this one, try not to pile in the back of it'.


Absolutely - if the information is there, then yes, use it.

But, I think we agree, no autonomous car should be allowed on the roads if it is unable to fully and properly function should such extra information (not independently gleaned from its own sensors) become unavailable for any reason.


I'm not sure we disagree...

That information raises the bar of what passes for properly function. There should be a base competency below which no system should be allowed on the road and that competency is inherently personal. At the very least in any situation a car should be capable of safely removing itself from the road, and safely covering itself, other vehicles and pedestrians. However through the application of simple standards the experience of using this technology can be massively improved, and even made safer to a degree not possible when each car has only one perspective to inform its decisions. At that point the wilfully dumb system which refuses to engage cooperatively, but passes the personal properly function test, might be deemed a disruptor, and therefor not to properly function on a wider scale. The driver who causes opprobrium without actually breaking any laws probably isn't too hard to imagine in human form.

dspp
Lemon Quarter
Posts: 3743
Joined: November 4th, 2016, 10:53 am
Has thanked: 2803 times
Been thanked: 883 times

Re: Musk endeavours

#251057

Postby dspp » September 11th, 2019, 2:08 pm

BobbyD wrote:
onthemove wrote:
BobbyD wrote:Cars should be able to do the driving and navigating bits without communication, but being aware that there are 16 people queued behind the car coming the other way would be quite useful. Another time this might be quite useful would be 'hey Mr. Tesla, I'm about to change lane because there's a stationary fire truck in this one, try not to pile in the back of it'.


Absolutely - if the information is there, then yes, use it.

But, I think we agree, no autonomous car should be allowed on the roads if it is unable to fully and properly function should such extra information (not independently gleaned from its own sensors) become unavailable for any reason.


I'm not sure we disagree...

That information raises the bar of what passes for properly function. There should be a base competency below which no system should be allowed on the road and that competency is inherently personal. At the very least in any situation a car should be capable of safely removing itself from the road, and safely covering itself, other vehicles and pedestrians. However through the application of simple standards the experience of using this technology can be massively improved, and even made safer to a degree not possible when each car has only one perspective to inform its decisions. At that point the wilfully dumb system which refuses to engage cooperatively, but passes the personal properly function test, might be deemed a disruptor, and therefor not to properly function on a wider scale. The driver who causes opprobrium without actually breaking any laws probably isn't too hard to imagine in human form.


When these standards are finally agreed & harmonised, I can well imagine that a lot of humans will fail them. This will create an interesting set of tensions. Will we call them sub-humans ?

regards, dspp

ReformedCharacter
Lemon Slice
Posts: 961
Joined: November 4th, 2016, 11:12 am
Has thanked: 463 times
Been thanked: 236 times

Re: Musk endeavours

#251077

Postby ReformedCharacter » September 11th, 2019, 2:47 pm

dspp wrote:
When these standards are finally agreed & harmonised, I can well imagine that a lot of humans will fail them. This will create an interesting set of tensions. Will we call them sub-humans ?

regards, dspp

I think it is likely that human driven vehicles will be severely restricted in the future for the greater good of the 'autonomous' traffic system because they won't mix optimally or safely. Perhaps roads will be hired at a price for those who enjoy driving 'classic' cars on special occasions. I look forward to it personally because I find the often aggressive and incompetent actions of human driven vehicles unpleasant and sometimes hazardous but I'm unlikely to live long enough to see it.

RC

tjh290633
Lemon Quarter
Posts: 3610
Joined: November 4th, 2016, 11:20 am
Has thanked: 238 times
Been thanked: 1200 times

Re: Musk endeavours

#251106

Postby tjh290633 » September 11th, 2019, 4:11 pm

ReformedCharacter wrote:
dspp wrote:
When these standards are finally agreed & harmonised, I can well imagine that a lot of humans will fail them. This will create an interesting set of tensions. Will we call them sub-humans ?

regards, dspp

I think it is likely that human driven vehicles will be severely restricted in the future for the greater good of the 'autonomous' traffic system because they won't mix optimally or safely. Perhaps roads will be hired at a price for those who enjoy driving 'classic' cars on special occasions. I look forward to it personally because I find the often aggressive and incompetent actions of human driven vehicles unpleasant and sometimes hazardous but I'm unlikely to live long enough to see it.

RC

It looks more likely to me that autonomous vehicles will be restricted to certain roads, from which normal vehicles will be excluded. The evidence so far is that they are dangerous missiles, which can go rogue at the slightest disruption, like a bird dropping on a sensor.

TJH

ReformedCharacter
Lemon Slice
Posts: 961
Joined: November 4th, 2016, 11:12 am
Has thanked: 463 times
Been thanked: 236 times

Re: Musk endeavours

#251174

Postby ReformedCharacter » September 11th, 2019, 7:36 pm

tjh290633 wrote:It looks more likely to me that autonomous vehicles will be restricted to certain roads, from which normal vehicles will be excluded. The evidence so far is that they are dangerous missiles, which can go rogue at the slightest disruption, like a bird dropping on a sensor.

TJH

In the near term perhaps, but then autonomous vehicles will become the 'normal' vehicles and non-autonomous vehicles will be regarded as we regard the horse drawn carriage today, as relics of the past. I quite agree about today's attempts at autonomous vehicles though, definitely not fit for purpose yet and probably for some time to come.

There are considerable potential savings of energy (and pollution) with a smarter transport system and that is likely to be a major force for change IMO.

RC

odysseus2000
Lemon Quarter
Posts: 1616
Joined: November 8th, 2016, 11:33 pm
Has thanked: 393 times
Been thanked: 199 times

Re: Musk endeavours

#251221

Postby odysseus2000 » September 11th, 2019, 9:44 pm

ReformedCharacter wrote:
tjh290633 wrote:It looks more likely to me that autonomous vehicles will be restricted to certain roads, from which normal vehicles will be excluded. The evidence so far is that they are dangerous missiles, which can go rogue at the slightest disruption, like a bird dropping on a sensor.

TJH

In the near term perhaps, but then autonomous vehicles will become the 'normal' vehicles and non-autonomous vehicles will be regarded as we regard the horse drawn carriage today, as relics of the past. I quite agree about today's attempts at autonomous vehicles though, definitely not fit for purpose yet and probably for some time to come.

There are considerable potential savings of energy (and pollution) with a smarter transport system and that is likely to be a major force for change IMO.

RC


Anyone who recalls the long trek to computers becoming world chess champions, remembers the great efforts put in by IBM and others and how the human world champions managed to beat them for many years and many experts said it would be a very long time, maybe never before computers would be world champions.

Now most smart phones running chess software are powerful enough to beat all human challengers. The same happened with Go, but from the first victory to computers being so much better than humans that humans no longer play computers at Go took months.

One can argue because of the dynamics of road travel and the plethora of unlikely events that can occur, that computers will never be as good as humans and it is true that the early efforts and to date humans are still more flexible and more competent to asses dangers, but I am not convinced that that superiority will last much longer.

When you consider that a 14 year old is often physically capable, but not legally allowed, to drive there is not much experience needed to gain a driving licence and that all happens within the limits of human vision, reaction times and other aspects of the human emotional condition. Computers have much better vision, reaction times many orders of magnitude faster and do not have emotional inputs to deal with.

It seems to me unlikely that computers will not be able to drive cars in next few years at better levels than average human driving.

I may be wrong and I welcome input that gives reasons as to why computers will not be able to drive cars better than humans.

Regards,

BobbyD
Lemon Quarter
Posts: 4892
Joined: January 22nd, 2017, 2:29 pm
Has thanked: 70 times
Been thanked: 230 times

Re: Musk endeavours

#251230

Postby BobbyD » September 11th, 2019, 10:28 pm

ReformedCharacter wrote:
tjh290633 wrote:It looks more likely to me that autonomous vehicles will be restricted to certain roads, from which normal vehicles will be excluded. The evidence so far is that they are dangerous missiles, which can go rogue at the slightest disruption, like a bird dropping on a sensor.

TJH

In the near term perhaps, but then autonomous vehicles will become the 'normal' vehicles and non-autonomous vehicles will be regarded as we regard the horse drawn carriage today, as relics of the past. I quite agree about today's attempts at autonomous vehicles though, definitely not fit for purpose yet and probably for some time to come.

There are considerable potential savings of energy (and pollution) with a smarter transport system and that is likely to be a major force for change IMO.

RC


I would caution against judging current autonomous tech fby looking at Tesla's precocious cruise control.

BobbyD
Lemon Quarter
Posts: 4892
Joined: January 22nd, 2017, 2:29 pm
Has thanked: 70 times
Been thanked: 230 times

Re: Musk endeavours

#251231

Postby BobbyD » September 11th, 2019, 10:34 pm

odysseus2000 wrote:
When you consider that a 14 year old is often physically capable, but not legally allowed, to drive there is not much experience needed to gain a driving licence and that all happens within the limits of human vision, reaction times and other aspects of the human emotional condition. Computers have much better vision, reaction times many orders of magnitude faster and do not have emotional inputs to deal with.

It seems to me unlikely that computers will not be able to drive cars in next few years at better levels than average human driving.
,


Ody, you would do well to learn more about the brain before testing to compare it to silicon and AI. It's all already been covered in this thread.

TUK020
Lemon Slice
Posts: 733
Joined: November 5th, 2016, 7:41 am
Has thanked: 151 times
Been thanked: 331 times

Re: Musk endeavours

#251274

Postby TUK020 » September 12th, 2019, 7:28 am

VW has unveiled a new version of its all-electric e-up! city car at this year’s Frankfurt Motor Show, with a claimed range of 162 miles

...The revised Volkswagen e-up! will go on sale in January 2020, with prices starting from around £23,000 (before the Government’s £3,500 plug-in car grant has been applied). First deliveries are expected to arrive during spring.


How will they pronounce this in Yorkshire?

odysseus2000
Lemon Quarter
Posts: 1616
Joined: November 8th, 2016, 11:33 pm
Has thanked: 393 times
Been thanked: 199 times

Re: Musk endeavours

#251279

Postby odysseus2000 » September 12th, 2019, 7:45 am

BobbyD wrote:
odysseus2000 wrote:
When you consider that a 14 year old is often physically capable, but not legally allowed, to drive there is not much experience needed to gain a driving licence and that all happens within the limits of human vision, reaction times and other aspects of the human emotional condition. Computers have much better vision, reaction times many orders of magnitude faster and do not have emotional inputs to deal with.

It seems to me unlikely that computers will not be able to drive cars in next few years at better levels than average human driving.
,


Ody, you would do well to learn more about the brain before testing to compare it to silicon and AI. It's all already been covered in this thread.


Arguments used to predict that computers could not become invincible in chess & Go, were often about how much more complex is the human brain than a machine, but those arguments were wrong.

There may indeed be reasons why a computer can not drive better than a human, as we have covered before on this thread, but no one knows & we now have more & more of the world's scientific & human capital focused on AI.

Robotic driving may happen far sooner than many experts believe and just because it hasn't yet happened does not mean that we can ignore it & assume it never will.

Regards,

odysseus2000
Lemon Quarter
Posts: 1616
Joined: November 8th, 2016, 11:33 pm
Has thanked: 393 times
Been thanked: 199 times

Re: Musk endeavours

#251356

Postby odysseus2000 » September 12th, 2019, 12:37 pm

S sets fastest lap time (not official) for 4 door passenger car at Laguna with new plaid power train:

https://www.techspot.com/news/81860-tes ... laims.html

Regards,

ReformedCharacter
Lemon Slice
Posts: 961
Joined: November 4th, 2016, 11:12 am
Has thanked: 463 times
Been thanked: 236 times

Re: Musk endeavours

#251359

Postby ReformedCharacter » September 12th, 2019, 12:55 pm

odysseus2000 wrote:
Arguments used to predict that computers could not become invincible in chess & Go, were often about how much more complex is the human brain than a machine, but those arguments were wrong.

There may indeed be reasons why a computer can not drive better than a human, as we have covered before on this thread, but no one knows & we now have more & more of the world's scientific & human capital focused on AI.

Robotic driving may happen far sooner than many experts believe and just because it hasn't yet happened does not mean that we can ignore it & assume it never will.

Regards,

I don't recall anyone worth listening to suggesting that computers would never become better than humans at Chess or Go only disagreements about how long it would take. IIRC the ability of a computer to beat the best Go players happened sooner than predicted.

I happen to think that 'better than human' autonomy will take quite a few years to achieve, probably longer than you do. I also think that the task will be unnecessarily difficult without relatively inexpensive and relatively technologically achievable additions such as inter-vehicle communications and smart road 'furniture' that can relay information to vehicles. To try to achieve autonomy without those things seems to make that objective harder than it need be.

RC

odysseus2000
Lemon Quarter
Posts: 1616
Joined: November 8th, 2016, 11:33 pm
Has thanked: 393 times
Been thanked: 199 times

Re: Musk endeavours

#251366

Postby odysseus2000 » September 12th, 2019, 1:23 pm

ReformedCharacter wrote:
odysseus2000 wrote:
Arguments used to predict that computers could not become invincible in chess & Go, were often about how much more complex is the human brain than a machine, but those arguments were wrong.

There may indeed be reasons why a computer can not drive better than a human, as we have covered before on this thread, but no one knows & we now have more & more of the world's scientific & human capital focused on AI.

Robotic driving may happen far sooner than many experts believe and just because it hasn't yet happened does not mean that we can ignore it & assume it never will.

Regards,

I don't recall anyone worth listening to suggesting that computers would never become better than humans at Chess or Go only disagreements about how long it would take. IIRC the ability of a computer to beat the best Go players happened sooner than predicted.

I happen to think that 'better than human' autonomy will take quite a few years to achieve, probably longer than you do. I also think that the task will be unnecessarily difficult without relatively inexpensive and relatively technologically achievable additions such as inter-vehicle communications and smart road 'furniture' that can relay information to vehicles. To try to achieve autonomy without those things seems to make that objective harder than it need be.

RC


I used to lecture the digital computer option in Oxford before computers became world chess champions & there were then respectable experts including senior Don's there who claimed that computers would never be able to beat humans. One could have some sympathy with such views in that the initial methods of calculating very large numbers of possible moves & then trying to optimise needed very large & powerful computers. The mistake was not to realise that more efficient neural net processors would be created that over came many of these troubles. The primary reason for this being that a lot of the early neural net theory was wrong.

Much of the current limitations in robotic driving come from pattern recognition failures. If the robotic systems can become better at this almost all the other aspects have already been solved and the evidence I am seeing indicates that they are becoming better at pattern recognition in terms of things like face id.

I may be completely wrong & yes vehicle to vehicle systems might make the whole business better but a young human soon learns how to ride a bike & does this via pattern recognition & brain limb coordination. Computers do not seem so far away from this. The more complex stuff about child to child & child to adult actions is currently much more challenging for computers imho.

Regards,

BobbyD
Lemon Quarter
Posts: 4892
Joined: January 22nd, 2017, 2:29 pm
Has thanked: 70 times
Been thanked: 230 times

Re: Musk endeavours

#251394

Postby BobbyD » September 12th, 2019, 2:57 pm

odysseus2000 wrote:
BobbyD wrote:
odysseus2000 wrote:
When you consider that a 14 year old is often physically capable, but not legally allowed, to drive there is not much experience needed to gain a driving licence and that all happens within the limits of human vision, reaction times and other aspects of the human emotional condition. Computers have much better vision, reaction times many orders of magnitude faster and do not have emotional inputs to deal with.

It seems to me unlikely that computers will not be able to drive cars in next few years at better levels than average human driving.
,


Ody, you would do well to learn more about the brain before testing to compare it to silicon and AI. It's all already been covered in this thread.


Arguments used to predict that computers could not become invincible in chess & Go, were often about how much more complex is the human brain than a machine, but those arguments were wrong.

There may indeed be reasons why a computer can not drive better than a human, as we have covered before on this thread, but no one knows & we now have more & more of the world's scientific & human capital focused on AI.

Robotic driving may happen far sooner than many experts believe and just because it hasn't yet happened does not mean that we can ignore it & assume it never will.

Regards,


None of that has anything to do with your ability to make a comparison between something and the human brain, which you clearly don't have a great understanding of. Your logic on the potential of AD doesn't make your comparisons to meatware systems valid.

BobbyD wrote:The human brain is the most complex known thing in the universe, it contains about 86 billion neurons each of which is connected to an average of somewhere north of 1000 other neurons, it is massively parallel and contains areas of deep specialisation whilst retaining levels of functional plasticity which allow it to come back from devastating injury, and while it is at it it creates the entire universe as you know it. The processor in a Tesla can manipulate 1's and 0's quickly, it's a pretty decent calculator.


BobbyD wrote:I don't want to unnecessarily recycle old arguments, but meatware and silicon solutions aren't directly comparable because of the fundamental difference in the processing units. That isn't to say that AD's can't outperform humans, but those AD's are not going to be human drivers rendered in silicon and algorithms, making direct comparisons misleading.


...and do we really have to cover the difference between a game with a defined playing area, defined pieces with defined properties and absolute rules and the Scorch Hill junction during school run again? Chess and Go both have massive numbers of permutations, in the same way that if you properly shuffle a deck of cards it is likely that the cards in your deck are in an order that has never before existed in the universe. But each of those permutations obeys rules. Black can't have 8 pawns and 3 bishops, even if all the bishops are on black squares. White can't place his stone on top of a black stone which is already on the intersection he wants to occupy. If in a chess match you turn the corner to discover a pram rolling down the e file while a football bounces out from behind the rook on c5, and a knight casually steps on to the board at h2 you raise your hand and inform the referee.

Chess and Go are both games of perfect information. Both players not only know how many pieces are in play, what those pieces are, whose those pieces are, and how they are allowed to behave they know every possible permutation of pieces which will be possible for the entire game from this point. If you gave them enough time, and we are talking fantastic amounts of time here, each player could logically derive every possible board state allowed from the present position. Chess and Go are inherently calculable.

This makes them incredibly bad comparisons for driving a car in the real world, which is a game of very limited information, unlimited players with unlimited properties, and rules, the most solid of which are optional on a good day, and even then subservient to greater principles like not running over the old lady lying in the middle of the road even if you have to make a prohibited lane change to avoid her.

...and please bare in mind before you set off on the potential of AD that I am one of the strongest advocates for AD on this board. These are not arguments against your conclusion, they are perceived weaknesses in the justifications you use to support it.

odysseus2000
Lemon Quarter
Posts: 1616
Joined: November 8th, 2016, 11:33 pm
Has thanked: 393 times
Been thanked: 199 times

Re: Musk endeavours

#251425

Postby odysseus2000 » September 12th, 2019, 4:08 pm

BobbyD
This makes them incredibly bad comparisons for driving a car in the real world, which is a game of very limited information, unlimited players with unlimited properties, and rules, the most solid of which are optional on a good day, and even then subservient to greater principles like not running over the old lady lying in the middle of the road even if you have to make a prohibited lane change to avoid her.

...and please bare in mind before you set off on the potential of AD that I am one of the strongest advocates for AD on this board. These are not arguments against your conclusion, they are perceived weaknesses in the justifications you use to support it.


Your assumptions that robotic driving is a game of limited information, unlimited players and unlimited properties is an interesting one.

Cars, people, most things that matter in the real world obey Fermion statistics.

In the driving world all things are necessarily limited by available space which is limited by the known dimensions of humans, roads, cars and their performance parameters. There are not unlimited possibilities, there is a limited set of possibilities.

AI has to discard the impossible and determine what is the most likely probable from what is left.

In essence it has to use the methodology of Sherlock:

When ever you have eliminated the impossible, what ever is left, no matter how improbable, is the truth.

Regards,


Return to “Macro and Global Topics”

Who is online

Users browsing this forum: No registered users and 1 guest