Jump to content


Photo
* * * * - 3 votes

Roborace - Driverless Formula E support series


  • Please log in to reply
946 replies to this topic

#851 prty

prty
  • Member

  • 6,209 posts
  • Joined: April 05

Posted 16 July 2018 - 12:55

Formula student autonomous:

(First lap slow to map the circuit)


Advertisement

#852 7MGTEsup

7MGTEsup
  • Member

  • 2,160 posts
  • Joined: March 11

Posted 16 July 2018 - 12:57

Autonomous racing cars will eventually be much better than humans when driving on the limit. It's essentially a balance control problem, which robots excel at.

 

I have yet to see a robot that can perform tasks (in terms of balance) that a child over the age of 5 can do let alone some of the things that acrobats can do.

 

It took them years to get a robot to catch a ball or to walk up a flight of stairs (ground breaking in terms of robotics but hardly a challenge for your average human being).



#853 7MGTEsup

7MGTEsup
  • Member

  • 2,160 posts
  • Joined: March 11

Posted 16 July 2018 - 13:01

Formula student autonomous:

(First lap slow to map the circuit)

 

Quite impressive and would work on a street track with walls either side, not sure how well it would do on an open circuit. Curious to see what would happen if you moved a few cones by a meter when it wasn't looking.



#854 prty

prty
  • Member

  • 6,209 posts
  • Joined: April 05

Posted 16 July 2018 - 13:14

I have yet to see a robot that can perform tasks (in terms of balance) that a child over the age of 5 can do let alone some of the things that acrobats can do.

It took them years to get a robot to catch a ball or to walk up a flight of stairs (ground breaking in terms of robotics but hardly a challenge for your average human being).

You're thinking about it from the wrong perspective, human shaped robots that balance. It's rather about controlling systems.

And you don't have to look into the future. If you know any 5 year old (or adult!) that can modulate the brakes like ABS does, or that can let's say have a consistent 5% slip ratio like a traction control system has, I'd be interested to know! :)

Ps, in humanoids it's getting closer too

https://www.youtube....h?v=vjSohj-Iclc

Edited by prty, 16 July 2018 - 13:15.


#855 7MGTEsup

7MGTEsup
  • Member

  • 2,160 posts
  • Joined: March 11

Posted 16 July 2018 - 13:25

You're thinking about it from the wrong perspective, human shaped robots that balance. It's rather about controlling systems.

And you don't have to look into the future. If you know any 5 year old (or adult!) that can modulate the brakes like ABS does, or that can let's say have a consistent 5% slip ratio like a traction control system has, I'd be interested to know! :)

Ps, in humanoids it's getting closer too

https://www.youtube....h?v=vjSohj-Iclc

 

A good human driver can brake on the limits of adhesion and control slip very accurately. You're confusing systems put into production cars to prevent Mr/Mrs average from getting into trouble with how competent a top flight racing driver is.

 

I seem to remember them saying drivers would struggle to not burn up the tyres after the traction control ban. Didn't seem to be much of an issue. You're forgetting the human beings greatest trait, being able to adapt.


Edited by 7MGTEsup, 16 July 2018 - 13:27.


#856 prty

prty
  • Member

  • 6,209 posts
  • Joined: April 05

Posted 16 July 2018 - 13:52

A good human driver can brake on the limits of adhesion and control slip very accurately. You're confusing systems put into production cars to prevent Mr/Mrs average from getting into trouble with how competent a top flight racing driver is.

I seem to remember them saying drivers would struggle to not burn up the tyres after the traction control ban. Didn't seem to be much of an issue. You're forgetting the human beings greatest trait, being able to adapt.


Humans can of course control it, but not as well and as consistently, that's the point. That's why it's banned in F1 or why it helps with tyre degradation in WEC.

Edited by prty, 16 July 2018 - 13:53.


#857 HP

HP
  • Member

  • 18,257 posts
  • Joined: October 99

Posted 16 July 2018 - 14:04

Humans can of course control it, but not as well and as consistently, that's the point. That's why it's banned in F1 or why it helps with tyre degradation in WEC.

The control technology has been built and invented by humans too, so banning, or making a fuss about the technology is IMO looking fundamentally wrong at it. Tech is humans servant. The question is simply who is responsible for it. The driver or the engineer? That engineers crave to get the proper recognition is understandable, as in people interested in motor sports are mostly fans of a driver or a team. Engineering feats rarely get recognized, they usually get banned. So no wonder that engineers are in particular eager to show of their stuff.

 

I'd only get worried about it, if robots start making humans though.



#858 7MGTEsup

7MGTEsup
  • Member

  • 2,160 posts
  • Joined: March 11

Posted 16 July 2018 - 14:19

I'd only get worried about it, if robots start making humans though.

 

Why would a robot make a whiny emotional flesh bag?



#859 SmallHorsey

SmallHorsey
  • Member

  • 951 posts
  • Joined: July 14

Posted 16 July 2018 - 14:27

Why would a robot make a whiny emotional flesh bag?

Because the three laws!!

Advertisement

#860 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 08 September 2018 - 20:41

The farce continues... 

 

https://www.motorspo...72086/?nrt=3687

 

This is about the lamest news to come out in a while.  Please just scrap this, it has been nothing but an embarrassment every time a new article comes out.



#861 BRG

BRG
  • Member

  • 16,948 posts
  • Joined: September 99

Posted 08 September 2018 - 21:25

:rotfl: :rotfl: :rotfl: :rotfl: :rotfl: :rotfl:



#862 Redback

Redback
  • Member

  • 1,283 posts
  • Joined: June 08

Posted 08 September 2018 - 21:53

The farce continues... 

 

https://www.motorspo...72086/?nrt=3687

 

This is about the lamest news to come out in a while.  Please just scrap this, it has been nothing but an embarrassment every time a new article comes out.

 

Actually, I think this is the most logical step forward for the first year of racing, - the "alpha" year as it's been termed.

 

AI is about machine learning and the best teacher in this instance is going to be a human driver.

 

Software engineers are great at what they do, but most aren't also professional racing drivers, so codifying every aspect of what a human driver actually does is extraordinarily difficult for anyone sitting in front of a screen.

 

Racing drivers sometimes don't even consciously recognise what they do in specific circumstances, so even if they could write code, trying to encapsulate their skills in software would be problematic.

 

A fully-instrumented test bed that records (in real-time) every aspect of a driver's input as well as the car's motion, relative position and the ambient environmental conditions etc, will allow the software developers to "see" the subtle nuances of on-the-limit driving and ensure that the AI is able to learn these skills.  The second part of the race will illustrate how well the AI has acquired (and applied) this knowledge.

 

The first year's "alpha" format is the most logical and the most expeditious means of developing the required code.

 

Certainly, the ill-informed and uneducated will initially laugh, but sceptics and those without imagination also laughed at Wilbur and Orville Wright's first flight of 37 metres.  The smirks of derision quickly faded from their faces.

 

I suspect the same will happen with autonomous racing, - though perhaps even quicker than with the evolution of powered flight.


Edited by Redback, 08 September 2018 - 21:55.


#863 Spillage

Spillage
  • Member

  • 5,910 posts
  • Joined: May 09

Posted 08 September 2018 - 22:31

Yeah, that is pretty seriously naff. Apart from anything else, having a driver teach the car is years old: https://youtu.be/WsnKzK6dX8Q

#864 Kev00

Kev00
  • Member

  • 3,495 posts
  • Joined: July 15

Posted 08 September 2018 - 22:37

Why though? Why don’t they just see how quickly the can make four wheeled robot go around a race track?

What’s so great about simulating and actual human racing driver?

#865 Redback

Redback
  • Member

  • 1,283 posts
  • Joined: June 08

Posted 09 September 2018 - 00:02

Yeah, that is pretty seriously naff. Apart from anything else, having a driver teach the car is years old: https://youtu.be/WsnKzK6dX8Q

 

Not naff, - sensible - and just because the teaching approach has been used before, doesn't mean it shouldn't be used now.  (Unless you can articulate a reason why the concept of "teaching" is now suddenly irrelevant or obsolete.)

 

 

Why though? Why don’t they just see how quickly the can make four wheeled robot go around a race track?

What’s so great about simulating and actual human racing driver?

 

The fact that this has to be explained, reinforces elements of my earlier post.

 

The object of Roborace as an engineering exercise, is to develop AI and autonomous technologies so that what is learned, can be applied to the benefit of the broader community in autonomous transport scenarios.

 

Accelerating this process by whatever means, is critical in that pursuit.

 

It also provides an opportunity for engineers to emerge from the wings and display their skills in an unfettered field of competition.

 

Throwing a bunch of driverless cars onto a track just to see what happens is analogous to putting a bunch of babies in a pen with some chainsaws to see if they can work out how to use them.  Some might eventually learn, but it would take quite a while and there might be casualties...

 

Better to teach.  

 

As to your second point, - human racing drivers currently represent the pinnacle of vehicle control skills and predictive anticipation in a dynamic environment.  Why would you NOT use them as an exemplar from which robots can learn?

 

This shouldn't need to be explained.


Edited by Redback, 09 September 2018 - 00:04.


#866 johnmhinds

johnmhinds
  • Member

  • 7,292 posts
  • Joined: July 09

Posted 09 September 2018 - 00:29

Human reaction times are pathetic compared to how fast computers can do the job.

Teaching a car what a human does and repeating those same actions would just be holding the tech back.

#867 Redback

Redback
  • Member

  • 1,283 posts
  • Joined: June 08

Posted 09 September 2018 - 00:58

Human reaction times are pathetic compared to how fast computers can do the job.

Teaching a car what a human does and repeating those same actions would just be holding the tech back.

 

It's not just about reaction times though is it?  That's a naively simplistic perspective.  It's also about HOW you react to a particular set of circumstances under specific conditions.

 

As I said earlier - "vehicle control skills and predictive anticipation in a dynamic environment".

 

The human brain is effectively a predictive approximation engine.  Almost nothing it does is actually precise because it doesn't have the processing speed to do so and "reasonably accurate approximation" has been a successful approach to survival during the course of human evolution.

 

A computer may react faster and with more precision, but teaching it how to react and to predict both environmental variables as well as the behaviour of other competitors is a lot more complex than simple processing speed.  Indeed the algorithms used will need to forgo precision in many instances.

 

Humans currently do that much better than machines and so they are the best teachers, even if their processing speeds are slower.



#868 RacingGreen

RacingGreen
  • Member

  • 2,503 posts
  • Joined: March 17

Posted 09 September 2018 - 01:09

My favorite autonomous car...

 

f04909b6-8afc-11e6-871c-e4d43e667f97.jpg



#869 FredF1

FredF1
  • Member

  • 2,284 posts
  • Joined: April 00

Posted 09 September 2018 - 07:57

My favorite autonomous car...

 

f04909b6-8afc-11e6-871c-e4d43e667f97.jpg

 

Does this mean that Volkswagen are not going be joining F1? :cry:



#870 pdac

pdac
  • Member

  • 8,252 posts
  • Joined: February 10

Posted 09 September 2018 - 10:30

Human reaction times are pathetic compared to how fast computers can do the job.

Teaching a car what a human does and repeating those same actions would just be holding the tech back.

 

It's 'teaching' the system, though. I would imagine the system isn't going to copy what it's taught, it's going to use the additional knowledge to hone it's own ideas. That's what most humans do when they are taught something (although, there are some that just blindly follow what someone has shown them).



#871 johnmhinds

johnmhinds
  • Member

  • 7,292 posts
  • Joined: July 09

Posted 09 September 2018 - 12:09

A computer can’t learn those kinds of predictive actions by looking at the records of the racing drivers inputs though. You’d need to read the drivers mind to get that kind of info.

And even if you could do that it’s not going to be all that relevant to a car that is using LIDAR to map out its exact position on a track and monitoring the grip levels with so much more precision than a human ever could.

Sending a robocar out onto an unknown track and having it come up with its own unique motions based on what all the sensors are telling it about the things around it is a different task to having a car copy and approximate what humans are already doing.

#872 PayasYouRace

PayasYouRace
  • RC Forum Host

  • 20,146 posts
  • Joined: January 10

Posted 09 September 2018 - 12:16

The idea isn't that the car will just copy and approximate what the human driver is doing. It'll be using the information it gains from being driven by the human driver, in terms of what the limits are like and what it can be capable of and using that to improve its own algorithms.

 

It makes perfect sense, as is essentially like having a Pro-Am class in GT racing, where the Am watches what the Pro does, and then attempts to use that knowledge to drive themselves.



#873 johnmhinds

johnmhinds
  • Member

  • 7,292 posts
  • Joined: July 09

Posted 09 September 2018 - 12:28

But the cars sensors are going to be so much better at finding those limits than a human is.

 

There is little point in telling the robocar that a human driver can take a certain corner at 75mph in his car when the robocar can constantly sense everything that is going on around it and knows its own limit can be pushed to 125mph based on all the variables it has.



#874 pdac

pdac
  • Member

  • 8,252 posts
  • Joined: February 10

Posted 09 September 2018 - 12:45

A computer can’t learn those kinds of predictive actions by looking at the records of the racing drivers inputs though. You’d need to read the drivers mind to get that kind of info.

And even if you could do that it’s not going to be all that relevant to a car that is using LIDAR to map out its exact position on a track and monitoring the grip levels with so much more precision than a human ever could.

Sending a robocar out onto an unknown track and having it come up with its own unique motions based on what all the sensors are telling it about the things around it is a different task to having a car copy and approximate what humans are already doing.

 

That's exactly what I expect it's trying to achieve. My understanding about these kind of learning systems is that the human will take an action. The system will store away the action along with the context. Then the system will keep comparing the current situation with the stored information, looking at how closely the current situation matches others where a human action is known. With more human actions stored, it can refine its understanding of how to react given a particular set of circumstances. Eventually it will become quite good at identifying how well the current situation matches with the known actions and can take the right action by itself. The more human input, the more the learning machine can get into the mind of the human.



#875 Kalmake

Kalmake
  • Member

  • 3,420 posts
  • Joined: November 07

Posted 09 September 2018 - 12:54

Really they just want the human in there for safety. Yes, they will want to learn each track, but this could be done using one human driver and sharing the data.



#876 ANF

ANF
  • Member

  • 10,198 posts
  • Joined: April 12

Posted 09 September 2018 - 13:39

But the cars sensors are going to be so much better at finding those limits than a human is.
 
There is little point in telling the robocar that a human driver can take a certain corner at 75mph in his car when the robocar can constantly sense everything that is going on around it and knows its own limit can be pushed to 125mph based on all the variables it has.

Currently it's the other way around: the self-driving cornering speed is laughable.



#877 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 09 September 2018 - 13:45

True machine learning doesn't mimic humans.  Rules are established, and then the AI is set loose to figure it out on its own, coming up with completely unique approaches (such as AlphaGo Zero).  What they should do (and should have been doing for a long time now already, instead of still fumbling around with the idiotic DevBot) is to unleash a swarm of scale models in a controlled environment and let them race each other to learn the skill of defeating each other.  Change the environment and conditions up, and eventually scale it up.  This would be 1000 times more interesting than this embarrassing farce that is currently going on.


Edited by maximilian, 09 September 2018 - 13:57.


#878 prty

prty
  • Member

  • 6,209 posts
  • Joined: April 05

Posted 09 September 2018 - 18:21

Currently it's the other way around: the self-driving cornering speed is laughable.

 

Lack of mid-corner balance and control is for sure no issue for robots. If it's not there yet in Roborace, it's because they didn't reach the development of that part at this point in time.

BMW had for example a self-drifting car more than four years ago:

 



#879 BRG

BRG
  • Member

  • 16,948 posts
  • Joined: September 99

Posted 09 September 2018 - 21:38

Evidently the threat to humanity from Skynet has been rather exaggerated.



Advertisement

#880 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 09 September 2018 - 23:11

Evidently the threat to humanity from Skynet has been rather exaggerated.

 

Apparently they created Roborace to lull us into a false sense of security as a decoy.  Couldn't be further from the truth, we are quickly approaching that "singularity" threshold, if we haven't already.  Which makes Roborace seem even the more ridiculous, soooo far behind what current AI is up to.

 

Btw since now the series won'd be "driverless", I propose we change the thread title to:

 

Roborace: Pointless Formula E Support Series



#881 pdac

pdac
  • Member

  • 8,252 posts
  • Joined: February 10

Posted 09 September 2018 - 23:18

Apparently they created Roborace to lull us into a false sense of security as a decoy.  Couldn't be further from the truth, we are quickly approaching that "singularity" threshold, if we haven't already.  Which makes Roborace seem even the more ridiculous, soooo far behind what current AI is up to.

 

Btw since now the series won'd be "driverless", I propose we change the thread title to:

 

Roborace: Pointless Formula E Support Series

 

Meanwhile, those involved in Roborace have invented jobs for themselves so that they can play around and suck up someone else's money.



#882 Redback

Redback
  • Member

  • 1,283 posts
  • Joined: June 08

Posted 10 September 2018 - 01:48

True machine learning doesn't mimic humans.  Rules are established, and then the AI is set loose to figure it out on its own, coming up with completely unique approaches (such as AlphaGo Zero).  What they should do (and should have been doing for a long time now already, instead of still fumbling around with the idiotic DevBot) is to unleash a swarm of scale models in a controlled environment and let them race each other to learn the skill of defeating each other.  Change the environment and conditions up, and eventually scale it up.  This would be 1000 times more interesting than this embarrassing farce that is currently going on.

 

It's important to understand that a machine with 4 dedicated TensorFlow Application Specific Integrated Circuits (ASICs) playing GO and taking half a second for each move is not directly comparable with what's needed to steer a car around a race track in real time at racing speeds.  Whist game-play in GO is very complex, the basic rules are very simple and the TCU's inputs at any one time are just the position of its opponent's pieces and its own pieces.   The same cannot be said for autonomous vehicle racing.

 

Furthermore, AlphaGo Zero took 4 million trial games (admittedly against itself) to "teach itself how to win.  Not sure the RoboRace team or the spectating public have time for 4 million laps of each circuit.  Even .1% of that would be impractical.

 

In addition, before the 4 million trial games, the baseline node weights of the "neural net" had to be established and this was done with training.

 

Finally, these TPUs (and the servers in which they reside) are sort of expensive and pretty big.  I doubt RoboRace has the budget for very expensive custom ASICs at present which is why they've partnered with Nvidia and will be using their Pegasus system (as a replacement for the earlier PX-2).



#883 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 10 September 2018 - 02:18

Meh, you could easily run 4 million trial simulations on virtual tracks for the AI to learn the basics, then set them free on a scale physical track.  Basic rules are very simple:  stay on track, go as fast as possible, don't damage yourself or your opponent, defeat as many opponents as possible.  That's it.

 

I suspect the problem is indeed funding, but why go thru the exercise of introducing this to a global audience, promising the sky and acting like this is the cutting edge of autonomous mobility, and then making yourself look ridiculous, whatever the reason?  Do it properly, or go home.

 

Especially adding human drivers back in is just an insult to the intelligence of the audience.


Edited by maximilian, 10 September 2018 - 02:28.


#884 Redback

Redback
  • Member

  • 1,283 posts
  • Joined: June 08

Posted 10 September 2018 - 10:41

Meh, you could easily run 4 million trial simulations on virtual tracks for the AI to learn the basics, then set them free on a scale physical track.  Basic rules are very simple:  stay on track, go as fast as possible, don't damage yourself or your opponent, defeat as many opponents as possible.  That's it.

 

I suspect the problem is indeed funding, but why go thru the exercise of introducing this to a global audience, promising the sky and acting like this is the cutting edge of autonomous mobility, and then making yourself look ridiculous, whatever the reason?  Do it properly, or go home.

 

Especially adding human drivers back in is just an insult to the intelligence of the audience.

 

OK.  If you really believe it's as simple as that (I don't think you do) then you're well out of your depth.

 

If you know it's not that simple (as I suspect) then you're just trolling...

 

Either way, - the conversation over.



#885 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 10 September 2018 - 10:45

OK.  If you really believe it's as simple as that (I don't think you do) then you're well out of your depth.

 

If you know it's not that simple (as I suspect) then you're just trolling...

 

Either way, - the conversation over.

 

Not for me   ;)



#886 kumo7

kumo7
  • Member

  • 2,005 posts
  • Joined: May 15

Posted 10 September 2018 - 12:41

Ha, as long as racing track is concerned, it is a easy win for computer riven car.

Teaching machine how tog o fast is a silly idea as computer leaned from human to chess and it is nearly unbeatable.

we know also that it is not interesting to look at.

 

Rallying will be more difficult, but as Computer will remember all corners perfectly it will be not interesting.

 

The true try is on the actual road where guys drive 40 - 60 km/h between the humans, bicycle, Motors, aged man women, children horse cats mouse dogs and rubbish dying round sometime birds and insects.

Funny man-machine, it is exactly the other way round. 



#887 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 10 September 2018 - 13:24

Well, the "appealing" part of it is to have AI compete with other AI to see how far these machines are able to push the boundaries of physics in order to arrive at the finish line before the other robocars, and how they might implement strategies or moves that no human driver might ever attempt or think of.  Again, this is why I think they should just set a swarm of these things free on a scaled control track and keep running them until interesting things begin to happen.  



#888 phrank

phrank
  • Member

  • 1,200 posts
  • Joined: May 01

Posted 10 September 2018 - 13:30

I like the concept, I love how the car look, but maybe they just should start to race at an oval first?



#889 Jordan44

Jordan44
  • Member

  • 8,199 posts
  • Joined: May 15

Posted 10 September 2018 - 13:33

Ha, as long as racing track is concerned, it is a easy win for computer riven car.

Teaching machine how tog o fast is a silly idea as computer leaned from human to chess and it is nearly unbeatable.

we know also that it is not interesting to look at.

 

Rallying will be more difficult, but as Computer will remember all corners perfectly it will be not interesting.

 

The true try is on the actual road where guys drive 40 - 60 km/h between the humans, bicycle, Motors, aged man women, children horse cats mouse dogs and rubbish dying round sometime birds and insects.

Funny man-machine, it is exactly the other way round. 

 

This. The idea that race tracks will provide useful technology transfer to road vehicles is far wide of the mark in my opinion.

 

This is just an entertainment exercise, there's nothing they'll learn that the big players wont already know.



#890 Tenmantaylor

Tenmantaylor
  • Member

  • 12,496 posts
  • Joined: July 01

Posted 10 September 2018 - 13:59

Evidently the threat to humanity from Skynet has been rather exaggerated.

 

Nah just the timing and embodiments. If we carry on as we are (infinite technology/capitalism loop) we'll end up creating technology that's more valuable than human life. The number of people that are valuable in a capitalist society will dwindle until we are redundant/outnumbered. What's that, only humans can buy things? Computers have been making purchases on our behalf for decades and contributing to capitalism. Once AI becomes sufficiently advanced one of these situations might actually come to be:

 

https://www.newscien...s-ai-intellect/

 

https://futurism.com...happen-by-2045/

 

Anyway. Back to Roborace. The number of profitable televised/webivised motorsports can probably counted on 2 hands. Can one without people driving b popular? I doubt it. For now.



#891 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 10 September 2018 - 14:03

 

Anyway. Back to Roborace. The number of profitable televised/webivised motorsports can probably counted on 2 hands. Can one without people driving b popular? I doubt it. For now.

 

Raises an interesting question:  would AI find it interesting to watch AI race against AI?  If millions of computers tune in to watch, it will be VERY popular!  :wave:


Edited by maximilian, 10 September 2018 - 14:04.


#892 BRG

BRG
  • Member

  • 16,948 posts
  • Joined: September 99

Posted 10 September 2018 - 15:26

This rather begs the question of why, if all those bright people beavering away for many months have had to chuck in the towel over driverless Roboracers, should we have any confidence in all the claims from the likes of Uber, Google, Amazon et al that they will be delivering autonomous road vehicles any time soon?  Surely a developing AV for a closed track must be infinitely easier than for your local High Street?



#893 Kalmake

Kalmake
  • Member

  • 3,420 posts
  • Joined: November 07

Posted 10 September 2018 - 15:55

This rather begs the question of why, if all those bright people beavering away for many months have had to chuck in the towel over driverless Roboracers, should we have any confidence in all the claims from the likes of Uber, Google, Amazon et al that they will be delivering autonomous road vehicles any time soon?  Surely a developing AV for a closed track must be infinitely easier than for your local High Street?

AV has been developed for a closed track years ago and lapped close to pro human level (by imitating humans).

 

This is not a massive company. This is a small number of people. Google could probably do in a week what these guys have spent three years on.

 

First season was supposed to be 2016-2017. I don't know who keeps giving them money and why.



#894 johnmhinds

johnmhinds
  • Member

  • 7,292 posts
  • Joined: July 09

Posted 10 September 2018 - 15:58

What has the Roborace budget been so far? A few million a year?

Silicon Valley companies like Google have been spending billions on driverless systems, so it’s not surprising that they’re well ahead of the competition.

#895 kumo7

kumo7
  • Member

  • 2,005 posts
  • Joined: May 15

Posted 11 September 2018 - 03:26

This. The idea that race tracks will provide useful technology transfer to road vehicles is far wide of the mark in my opinion.

 

This is just an entertainment exercise, there's nothing they'll learn that the big players wont already know.

 

From the video's liked on this thread it is obvious that car maker does know quite a bit already and they will not jump in besides making a AD exercise.

So to what extreme it can go, or who wants to pay that much to bring a robot hit its limit, is yet a question, don't you think?



#896 Jordan44

Jordan44
  • Member

  • 8,199 posts
  • Joined: May 15

Posted 11 September 2018 - 09:49

This rather begs the question of why, if all those bright people beavering away for many months have had to chuck in the towel over driverless Roboracers, should we have any confidence in all the claims from the likes of Uber, Google, Amazon et al that they will be delivering autonomous road vehicles any time soon?  Surely a developing AV for a closed track must be infinitely easier than for your local High Street?

 

There are already autonomous vehicles commercially available, Tesla currently being the most advanced.



#897 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 11 September 2018 - 11:00

This rather begs the question of why, if all those bright people beavering away for many months have had to chuck in the towel over driverless Roboracers, should we have any confidence in all the claims from the likes of Uber, Google, Amazon et al that they will be delivering autonomous road vehicles any time soon?  Surely a developing AV for a closed track must be infinitely easier than for your local High Street?

 

Exactly, which further illustrates just how painfully pathetic Roborace is.



#898 phrank

phrank
  • Member

  • 1,200 posts
  • Joined: May 01

Posted 11 September 2018 - 13:52

There are already autonomous vehicles commercially available, Tesla currently being the most advanced.

A Tesla is hardly autonomous. Creating a vehicle drives at peak performance seems to me a but more difficult than creating a car that just follows the road at limited speed. (Which in itself of course is already incredible)



#899 Jordan44

Jordan44
  • Member

  • 8,199 posts
  • Joined: May 15

Posted 11 September 2018 - 14:00

A Tesla is hardly autonomous. Creating a vehicle drives at peak performance seems to me a but more difficult than creating a car that just follows the road at limited speed. (Which in itself of course is already incredible)

 

Tesla is level 2 autonomy. Some cars are releasing level 3 (full-self driving) very soon.

 

Depends how you define difficulty. The Roborace teams likely wont have to face ethical issues (unless they create a sentient driving machine  :eek: ) and similar stuff that comes with road vehicles, which is why I believe there will be no useful technology transfer from this series to commercial use.


Edited by Jordan44, 11 September 2018 - 14:07.


Advertisement

#900 maximilian

maximilian
  • Member

  • 4,693 posts
  • Joined: February 10

Posted 11 September 2018 - 14:12

Racing 20 cars against each other on a closed, controlled circuit is a LOT easier task than navigating all the intricacies of urban and rural traffic on the open roads in direct interaction with thousands of other participants...