Jump to content


Photo
- - - - -

AI & F1


  • Please log in to reply
29 replies to this topic

#1 Michael Ferner

Michael Ferner
  • Member

  • 7,559 posts
  • Joined: November 09

Posted 30 March 2026 - 16:26

An outlook:
 
The skill set to drive a 2026 F1 car as fast as possible is something which a properly trained AI algorithm will probably master a lot (and I mean: a LOT!!) better than any human being, and that very soon. From comments I read, I gather that such programs are already in training, and already used to maximise performance on track. I think that even before the end of this year, drivers will be mere passengers in their cars, only there to perform emergency actions - a bit like driving autonomous cars on the road, you can still accelerate and brake, twist the steering wheel, but you won't have to unless some idiot in front of you tries to do something stupid. Of course, the FIA won't have that, so the driver will still be responsible for the usual controls, and perhaps it's better to compare it to the 'beginners mode' in a video game, where the racing line is marked on the track, and a light will go on whenever you have to brake or shift gear, so that you only have to steer as closely as possible to the markings on the track, and avoid running into someone else, or the scenery. Naturally, F1 drivers don't need a painted racing line, but they will be dependent on the "master's voice" as to where to brake, and when to accelerate for best results in harvesting and deployment.
 
You don't believe me? Let's look at it from the other side of the fence. Driving a racing car on the limit, and I mean the limits of physics: adhesion in the corners, braking distance, best compromise for forward bite out of the corners, and so on - is a skill set that demands a lot of what is sometimes called "fuzzy logic", and something at which humans excel: there is simply no such thing as a 'perfect lap', because conditions change constantly, and minutely - the driver has to react to a myriad of fuzzy information coming in (grip level of the road, tyre degradation, engine sound etc) and make thousands of decisions in every millisecond out there, it's so much at once that no one will ever get it absolutely right for any length of time. Instead, it's about getting there als closely as possible, and for as long as possible. A computer, on the other hand, would be able to calculate the exact top speed at which to take the corner, but only if it is fed with EXACT information about ALL the relevant parameters, which is simply way too much information to survey. Humans don't need that, they drive "by the seat of the pants" - fuzzy logic, in other words.
 
I always maintain, that the true litmus test for AI ability is to emulate a racing driver - chess is child's play for a computer, there are only several millions of very well defined parameters, and lots of time. But a racing driver has to compute thousands of VERY unprecise parameters every millisecond, too much for an AI to be really competitive with a human driver. AI thrives on predictability, Humans on fuzziness. I don't expect an AI to make an even only tolerably good racing driver any time soon. But now, we have F1 cars running out of steam on the straights, and drivers having to drive well within the physical limits of their cars as to not waste energy that is badly needed for straightline oomph - ideal conditions for AI to take over, I say!


Advertisement

#2 Ross Stonefeld

Ross Stonefeld
  • Member

  • 71,182 posts
  • Joined: August 99

Posted 30 March 2026 - 17:56

There's already an autonomous racing league. The thirst for non-human sports is so great that no one knows about it or seemingly gives a ****  :lol:



#3 gruntguru

gruntguru
  • Member

  • 7,765 posts
  • Joined: January 09

Posted 04 April 2026 - 01:01

 

The skill set to drive a 2026 F1 car as fast as possible is something which a properly trained AI algorithm will probably master a lot (and I mean: a LOT!!) better than any human being, and that very soon. From comments I read, I gather that such programs are already in training, and already used to maximise performance on track. I think that even before the end of this year, drivers will be mere passengers in their cars, only there to perform emergency actions . . . . 

 

You are not talking about F1 are you? You know such driver aids are banned?



#4 gruntguru

gruntguru
  • Member

  • 7,765 posts
  • Joined: January 09

Posted 04 April 2026 - 01:06

 

. . . . I don't expect an AI to make an even only tolerably good racing driver any time soon. . . . 

 

Whoops - its already happened.

How close is an AI car to F1 driver pace?

10 seconds slower than a human just one year ago and now only half a second behind Daniil Kvyat.



#5 Bloggsworth

Bloggsworth
  • Member

  • 9,580 posts
  • Joined: April 07

Posted 05 April 2026 - 17:03

There's already an autonomous racing league. The thirst for non-human sports is so great that no one knows about it or seemingly gives a ****  :lol:

I know, I saw a film of it and I couldn't stop laughing...



#6 Michael Ferner

Michael Ferner
  • Member

  • 7,559 posts
  • Joined: November 09

Posted 13 April 2026 - 09:31

Whoops - its already happened.

How close is an AI car to F1 driver pace?

10 seconds slower than a human just one year ago and now only half a second behind Daniil Kvyat.

 

Verstappen is shivering in fear.



#7 7MGTEsup

7MGTEsup
  • Member

  • 3,147 posts
  • Joined: March 11

Posted 17 June 2026 - 08:53

If they are going to make a robot racing series they need to make the cars completely insane with huge power and huge downforce to at least make it interesting.



#8 Greg Locock

Greg Locock
  • Member

  • 6,552 posts
  • Joined: March 03

Posted 17 June 2026 - 21:20

One big unknowable is the friction at each contact patch. Incidentally I read that 90% of F1 laps are open loop, the driver is not using feedback, just using memorised inputs. That may have been written by Stirling Moss.



#9 jcbc3

jcbc3
  • RC Forum Host

  • 15,654 posts
  • Joined: November 04

Posted 18 June 2026 - 08:01

It happened in chess. Computers are superior to humans. But we still go see the humans.

 

In the late 19th century the pianola was invented. Could play and reproduce the sound of any pianist just by paper rolls and some electricity. We still go to concerts.



#10 cbo

cbo
  • Member

  • 1,076 posts
  • Joined: September 21

Posted 18 June 2026 - 19:30

If they are going to make a robot racing series they need to make the cars completely insane with huge power and huge downforce to at least make it interesting.


If you remove the driver, you gain some space and if you ditch the now redundant safety features on the car you get more space as well as less weight and more freedom
of design.

An F1 car with the current engine and downforce redesigned as a driverless vehicle would probably be a bit of a monster.

#11 Henri Greuter

Henri Greuter
  • Member

  • 14,915 posts
  • Joined: June 02

Posted 23 June 2026 - 12:03

If they are going to make a robot racing series they need to make the cars completely insane with huge power and huge downforce to at least make it interesting.

 

 

 

A safe manner to see the 250 mph barrier broken at Indy as long as spectators are kept inside the track and not permitted in the grandstands



#12 Sterzo

Sterzo
  • Member

  • 7,890 posts
  • Joined: September 11

Posted 24 June 2026 - 11:56

A safe manner to see the 250 mph barrier broken at Indy as long as spectators are kept inside the track and not permitted in the grandstands

Why have spectators when you could have cheering robots in every seat?



#13 Bikr7549

Bikr7549
  • Member

  • 399 posts
  • Joined: May 16

Posted 24 June 2026 - 13:11

Why have spectators when you could have cheering robots in every seat?

Why have a track, cars or the robots? Just use a simulator for all.



#14 Bikr7549

Bikr7549
  • Member

  • 399 posts
  • Joined: May 16

Posted 25 June 2026 - 16:53

Why have a track, cars or the robots? Just use a simulator for all.

This was intended as a joke, a sour one perhaps but not a serious one and is mostly coming from disillusionment with rule complexities/limitations and spec-ish type series. There are some advantages to this route tho, and it could open up all kinds of technological ideas into the arena, sort of a modern Can Am with very loose rules. WIthout the huge money aspect of F1 it would likely (at the beginning anyway) just attract serious hobbyist players, or maybe F1 teams would find it interesting as a 'weekend' sideline project to unofficially test new ideas, done on the sim peoples own time . I am not a programmer or sim person so I am completely unaware of what it would take to do this, but an open playground would be cool, but of course there would be no real playground to go to, and not seeing/hearing/smelling the cars it is hard to imagine any kind of spectator or following interest. 



#15 Greg Locock

Greg Locock
  • Member

  • 6,552 posts
  • Joined: March 03

Posted 25 June 2026 - 18:44

You'd need to have  a standard vehicle model, otherwise you just wind in a huge tire friction number.

 

The teams would provide AI drivers and strateg, and maybe a setup for the car.



#16 MatsNorway

MatsNorway
  • Member

  • 2,867 posts
  • Joined: December 09

Posted 30 June 2026 - 07:29

A engineering challenge with a spec car sounds rather daft. Computer driven cars alone is not going to be very interesting. They need to have some additional appeal. Mad power would help.

 

And given complete lack of driver safety needed it would be rather cheap to make with the right ruleset. Just ban carbon fiber anything and you will still be able to automate a gocart with a turboed motorcycle engine. 



#17 Ross Stonefeld

Ross Stonefeld
  • Member

  • 71,182 posts
  • Joined: August 99

Posted 30 June 2026 - 14:43

I don't think people are going to care about non-human competition. Turbo F1 cars of the 80s were crazy because someone had to drive them.



#18 MatsNorway

MatsNorway
  • Member

  • 2,867 posts
  • Joined: December 09

Posted 02 July 2026 - 22:38

Some watch greyhounds race. I would probably watch automated cars over that if they got the speed and the violence they can have.



#19 MichaelHickey

MichaelHickey
  • Member

  • 7,051 posts
  • Joined: July 01

Posted 03 July 2026 - 00:41

People watch Battlebots. 



Advertisement

#20 Ross Stonefeld

Ross Stonefeld
  • Member

  • 71,182 posts
  • Joined: August 99

Posted 03 July 2026 - 14:06

Imagine the viewing figures on the human version  :eek:   :lol:



#21 Greg Locock

Greg Locock
  • Member

  • 6,552 posts
  • Joined: March 03

Posted 03 July 2026 - 20:15

Somewhere between gladiators and jousting I guess.



#22 gruntguru

gruntguru
  • Member

  • 7,765 posts
  • Joined: January 09

Posted 03 July 2026 - 22:33

One big unknowable is the friction at each contact patch. Incidentally I read that 90% of F1 laps are open loop, the driver is not using feedback, just using memorised inputs. That may have been written by Stirling Moss.

I imagine an AI driven car would have a more rapid response time than a human - testing friction and feeding back as you go. After the initial lap it will also make predictions for mu at every point based on previous laps.



#23 Ross Stonefeld

Ross Stonefeld
  • Member

  • 71,182 posts
  • Joined: August 99

Posted 04 July 2026 - 20:17

 That's what a human brain does, just a little slower. I think this underlines what AI is, and more importantly, not. It's a predictive model in a closed loop. 



#24 gruntguru

gruntguru
  • Member

  • 7,765 posts
  • Joined: January 09

Posted 06 July 2026 - 01:12

 That's what a human brain does, just a little slower. I think this underlines what AI is, and more importantly, not. It's a predictive model in a closed loop. 

You mean "faster"?



#25 Ross Stonefeld

Ross Stonefeld
  • Member

  • 71,182 posts
  • Joined: August 99

Posted 06 July 2026 - 14:23

No the human brain will do those live calculations much slower than the most basic computer. But it will do the same grip assessment. The load through the steering wheel rather than data from the suspension load cells. And less accurately. But it has the advantage of anticipating. 



#26 GreenMachine

GreenMachine
  • Member

  • 3,109 posts
  • Joined: March 04

Posted 06 July 2026 - 23:18

I don't think anticipation will be a problem.

 

More difficult will be the intuition/interpretation, the merging of data with memory and judgement.



#27 mariner

mariner
  • Member

  • 2,460 posts
  • Joined: January 07

Posted 07 July 2026 - 20:11

It would help to read the description by Peter Wright in his book of the G-G traces on a Lotus 79 at Paul Rickard, comparing the traces of  Mario Andretti, Jackie Stewart and Ronnie Peterson.

 

Andretti was fast and smooth lapping close to but not over the outer grip limit. Stewart was a bit slower but so, so smooth - he was just trying out the car not being a hero.

 

The interesting one was Ronnie P. Constantly over then under the theoretical limit , frantic corrections (even when not needed  it seemed) .  Peter Wright was mystified but acknowledged he was so fast.

 

The point is I wonder if AI could ever understand how Ronnie drove as not even one of the world's top race enginers couldn't figure it ot



#28 Greg Locock

Greg Locock
  • Member

  • 6,552 posts
  • Joined: March 03

Posted 08 July 2026 - 00:20

Bear in mind the theoretical G-G diagram is based on steady state tire tests, if you can somehow exploit the tire dynamics by rapid steering inputs then maybe it is faster. 



#29 LittleChris

LittleChris
  • Member

  • 4,611 posts
  • Joined: April 01

Posted 08 July 2026 - 12:25

I thought Jackie tried the 78 rather than the 79 ?

#30 gruntguru

gruntguru
  • Member

  • 7,765 posts
  • Joined: January 09

Posted 16 July 2026 - 03:26

No the human brain will do those live calculations much slower than the most basic computer. . . 

 

Like I said - "faster".    :)