could a fully automated car beat a human driven one?
Discussion
Of course they could.
Not on the first go,but with software updates etc then yes. Especially if the computer had a learning programme, where it learnt more about the tracks/cars/opponents.
Don't forget,a computer doesn't suffer from G force nor has a momentary lapse of concentration. Doesn't need a break and can do complex fuel/tyre calculations in an instant.
I think the only thing a computer would struggle with is aggression. It will be programmed to back off traffic, not dive into that 20cm gap like the general public do on every motorway, everyday.
Not on the first go,but with software updates etc then yes. Especially if the computer had a learning programme, where it learnt more about the tracks/cars/opponents.
Don't forget,a computer doesn't suffer from G force nor has a momentary lapse of concentration. Doesn't need a break and can do complex fuel/tyre calculations in an instant.
I think the only thing a computer would struggle with is aggression. It will be programmed to back off traffic, not dive into that 20cm gap like the general public do on every motorway, everyday.
Of course.
If not today - then in a few years.
I remember the chess computer days of "it can never beat a Grandmaster"
Then things improved and improved until Deep Blue came along - and beat Gary Kasparov.
Today - your little phone app running the best engine will wipe the floor with every Grandmaster.
Same as in racing.
You get someone saying "The driver is faster around the track with traction control off than on"
Then you look into it.
And it is over one lap.
You put in a full race duration - and the driver with traction control wipes the floor with the driver without.
It is all about consistency. Bots are damned good at that.
[Edit: had an interesting chat with a race engineer on this over a bottle of plonk or three ]
The weakest component in an F1 car is the driver.
But if you remove the driver - you kill the sport.
But yes - bots will go quicker.
Just give them enough time.
It really won't be long.
( As a kid - I remember the 8" floppy days of holding 32,000 ish characters.
Fast forward 40ish years....
My phone has 1TB in it. )
The days of you or I driving will be gone in my lifetime. And I'm knocking on in years.
If not today - then in a few years.
I remember the chess computer days of "it can never beat a Grandmaster"
Then things improved and improved until Deep Blue came along - and beat Gary Kasparov.
Today - your little phone app running the best engine will wipe the floor with every Grandmaster.
Same as in racing.
You get someone saying "The driver is faster around the track with traction control off than on"
Then you look into it.
And it is over one lap.
You put in a full race duration - and the driver with traction control wipes the floor with the driver without.
It is all about consistency. Bots are damned good at that.
[Edit: had an interesting chat with a race engineer on this over a bottle of plonk or three ]
The weakest component in an F1 car is the driver.
But if you remove the driver - you kill the sport.
But yes - bots will go quicker.
Just give them enough time.
It really won't be long.
( As a kid - I remember the 8" floppy days of holding 32,000 ish characters.
Fast forward 40ish years....
My phone has 1TB in it. )
The days of you or I driving will be gone in my lifetime. And I'm knocking on in years.
Edited by Jenny Tailor on Saturday 18th June 01:24
Jenny Tailor said:
Of course.
If not today - then in a few years.
I remember the chess computer days of "it can never beat a Grandmaster"
Then things improved and improved until Deep Blue came along - and beat Gary Kasparov.
Today - your little phone app running the best engine will wipe the floor with every Grandmaster.
Same as in racing.
You get someone saying "The driver is faster around the track with traction control off than on"
Then you look into it.
And it is over one lap.
You put in a full race duration - and the driver with traction control wipes the floor with the driver without.
It is all about consistency. Bots are damned good at that.
[Edit: had an interesting chat with a race engineer on this over a bottle of plonk or three ]
The weakest component in an F1 car is the driver.
But if you remove the driver - you kill the sport.
But yes - bots will go quicker.
Just give them enough time.
It really won't be long.
( As a kid - I remember the 8" floppy days of holding 32,000 ish characters.
Fast forward 40ish years....
My phone has 1TB in it. )
The days of you or I driving will be gone in my lifetime. And I'm knocking on in years.
IBM cheated when they won the chess match with Deep Blue. If you watch the documentary on it he explains they do something like 'best of three' when playing these tournaments. He beat the computer in the first game but they tweaked the programming before the second game and then he was unable to beat it. The inference being that had he been allowed to play the computer as it was on the first day he would've beaten it every time. If not today - then in a few years.
I remember the chess computer days of "it can never beat a Grandmaster"
Then things improved and improved until Deep Blue came along - and beat Gary Kasparov.
Today - your little phone app running the best engine will wipe the floor with every Grandmaster.
Same as in racing.
You get someone saying "The driver is faster around the track with traction control off than on"
Then you look into it.
And it is over one lap.
You put in a full race duration - and the driver with traction control wipes the floor with the driver without.
It is all about consistency. Bots are damned good at that.
[Edit: had an interesting chat with a race engineer on this over a bottle of plonk or three ]
The weakest component in an F1 car is the driver.
But if you remove the driver - you kill the sport.
But yes - bots will go quicker.
Just give them enough time.
It really won't be long.
( As a kid - I remember the 8" floppy days of holding 32,000 ish characters.
Fast forward 40ish years....
My phone has 1TB in it. )
The days of you or I driving will be gone in my lifetime. And I'm knocking on in years.
Edited by Jenny Tailor on Saturday 18th June 01:24
I believe there was a Toyota experiment with automated racing.
My feeling is generally speaking an automated car could not compete with a human unless the challenge is very specific and the machine is specially prepared for the task beforehand. Once underway the challenge must not change for the duration or the event because as soon as something changes it gives the human the upper hand immediately, maybe even to the point that the machine would no longer be able to continue.
Its only a game but the highest level of ai can beat the best of the best gran turismo players. Dont see why you couldnt teach ai to drive a real car as well. Always perfect lines/braking points etc. The tricky bit is making ai work with other people around.
https://www.cnet.com/tech/gaming/sony-gran-turismo...
Skip to like 31 minutes in
https://www.youtube.com/watch?v=qP1gjgtKyYc
https://www.cnet.com/tech/gaming/sony-gran-turismo...
Skip to like 31 minutes in
https://www.youtube.com/watch?v=qP1gjgtKyYc
Neural network/pattern matching AI can do this easily combined with generational/breeding type models.
Learn, breed, pick winners, breed, pick winners etc.
They were doing this in car racing games in the late 90s/early 00s but not sure it took off at the time, probably hard to control (difficulty, refining behaviours etc) for all manner of reasons as NNs are.
The big issue from my perspective is that breeding and teaching these type of AI means bias.
Everyone is trying to create unbiased AI, but bias creeps in and it’s not easily removed without training again from the start.
Ie, how do you train in Schumacher level of X positive attribute, without adding in Schumacher level of push other driver off the track kind of skills?
Then you have the “general AI” need for when a big new variable changes that the AI hasn’t been trained for.
Like front wing being damaged by an unknown amount, or whatever, that changes the config. Can it adapt quickly and intuitively?
Current AI systems might just be really rubbish at that point and play safe and not learn as well from experience like a human can with experience across vast swathes of reality.
Learn, breed, pick winners, breed, pick winners etc.
They were doing this in car racing games in the late 90s/early 00s but not sure it took off at the time, probably hard to control (difficulty, refining behaviours etc) for all manner of reasons as NNs are.
The big issue from my perspective is that breeding and teaching these type of AI means bias.
Everyone is trying to create unbiased AI, but bias creeps in and it’s not easily removed without training again from the start.
Ie, how do you train in Schumacher level of X positive attribute, without adding in Schumacher level of push other driver off the track kind of skills?
Then you have the “general AI” need for when a big new variable changes that the AI hasn’t been trained for.
Like front wing being damaged by an unknown amount, or whatever, that changes the config. Can it adapt quickly and intuitively?
Current AI systems might just be really rubbish at that point and play safe and not learn as well from experience like a human can with experience across vast swathes of reality.
te.Don’t be silly.
There are too many variables.
I watched the Supercars in Darwin at the weekend. Chaz Mostert pitted on lap 9 and put on two rear tyres. He ended up leading with 10 to go when the Safety Car came out. They restarted with 7 to go and he held on for the win - the car behind him had pitted on lap 22 and put on 4 tyres.
A robot couldn’t do that - the algorithms would have told it to pit again for fresh tyres.
There are too many variables.
I watched the Supercars in Darwin at the weekend. Chaz Mostert pitted on lap 9 and put on two rear tyres. He ended up leading with 10 to go when the Safety Car came out. They restarted with 7 to go and he held on for the win - the car behind him had pitted on lap 22 and put on 4 tyres.
A robot couldn’t do that - the algorithms would have told it to pit again for fresh tyres.
Slow said:
Its only a game but the highest level of ai can beat the best of the best gran turismo players. Dont see why you couldnt teach ai to drive a real car as well. Always perfect lines/braking points etc. The tricky bit is making ai work with other people around.
https://www.cnet.com/tech/gaming/sony-gran-turismo...
Skip to like 31 minutes in
https://www.youtube.com/watch?v=qP1gjgtKyYc
Because the AI in games knows all of the absolute limits:https://www.cnet.com/tech/gaming/sony-gran-turismo...
Skip to like 31 minutes in
https://www.youtube.com/watch?v=qP1gjgtKyYc
Knows the track boundary location and able to use the width of the track to 100%
Knows the ideal racing line and knows it's exact position on it at all times. AI running within a real car would need to process data from GPS, LIDAR, Cameras in real-time to have the same information.
Available tyre grip dependent on temperature and conditions
Will utilise ABS and TC to the full (part of the above)
Sim Physics are not 100% true to real life. They're close but not real.
Also, AI in games could easily be configured to not adhere to the limits defined for a human user.
I'd be interested to see how "Sophy Rouge" would have done if she hadn't started on Pole.
Edited by ridds on Monday 20th June 01:53
M.F.D said:
I can see the argument over 1 flying lap, but when you bring in changing weather conditions, tyre management etc into the equation then I would imagine we are a long way off.
Computing and A I can handle billions of variables in a few seconds, simple. Getting the algorithms right, a bit tricky. Ten years away.Easier than an AI sense of humour.
ridds said:
Slow said:
Its only a game but the highest level of ai can beat the best of the best gran turismo players. Dont see why you couldnt teach ai to drive a real car as well. Always perfect lines/braking points etc. The tricky bit is making ai work with other people around.
https://www.cnet.com/tech/gaming/sony-gran-turismo...
Skip to like 31 minutes in
https://www.youtube.com/watch?v=qP1gjgtKyYc
Because the AI in games knows all of the absolute limits:https://www.cnet.com/tech/gaming/sony-gran-turismo...
Skip to like 31 minutes in
https://www.youtube.com/watch?v=qP1gjgtKyYc
Knows the track boundary location and able to use the width of the track to 100%
Knows the ideal racing line and knows it's exact position on it at all times. AI running within a real car would need to process data from GPS, LIDAR, Cameras in real-time to have the same information.
Available tyre grip dependent on temperature and conditions
Will utilise ABS and TC to the full (part of the above)
Sim Physics are not 100% true to real life. They're close but not real.
Also, AI in games could easily be configured to not adhere to the limits defined for a human user.
I'd be interested to see how "Sophy Rouge" would have done if she hadn't started on Pole.
Edited by ridds on Monday 20th June 01:53
Slow said:
Its only a game but the highest level of ai can beat the best of the best gran turismo players. Dont see why you couldnt teach ai to drive a real car as well. Always perfect lines/braking points etc. The tricky bit is making ai work with other people around.
https://www.cnet.com/tech/gaming/sony-gran-turismo...
Skip to like 31 minutes in
https://www.youtube.com/watch?v=qP1gjgtKyYc
If the AI were interfaced to the hand held controller with the same handicap (or advantage, depending on your view) of the 4 fingers and opposable thumb, had to interpret gameplay using output from the screen observing the graphics and converting it into reactions which can be output to the controller I think you'd find it would really really struggle. If you call it AI which is operating within the program (i.e. what we used to call single player mode, or playing the computer) then it's as somebody else said... it knows all of the parameters so would be very hard to beat. https://www.cnet.com/tech/gaming/sony-gran-turismo...
Skip to like 31 minutes in
https://www.youtube.com/watch?v=qP1gjgtKyYc
Mr Whippy said:
Neural network/pattern matching AI can do this easily combined with generational/breeding type models.
Then you have the “general AI” need for when a big new variable changes that the AI hasn’t been trained for.
Like front wing being damaged by an unknown amount, or whatever, that changes the config. Can it adapt quickly and intuitively?
Current AI systems might just be really rubbish at that point and play safe and not learn as well from experience like a human can with experience across vast swathes of reality.
THIS. Then you have the “general AI” need for when a big new variable changes that the AI hasn’t been trained for.
Like front wing being damaged by an unknown amount, or whatever, that changes the config. Can it adapt quickly and intuitively?
Current AI systems might just be really rubbish at that point and play safe and not learn as well from experience like a human can with experience across vast swathes of reality.
There's that saying... in battle your plans last until first contact with the enemy. i.e. if you put AI in a challenge against a human on the same track at the same time and it was an agressive Schumacher driving style rather than a careful Coulthard one... could it cope with that?
Forums | General Gassing | Top of Page | What's New | My Stuff