Hi, I had make a Race in one player, at So Long with a XRR. I see how the AI burns his front tires in every brake.

I think that it would be hard to program, but it is possible that AI be autolearing? So, if i do a short race and I see my front tires red, i would change for R3 or R4, or increase the pressure, or adjust brake balance to rear, or reduce max brake force...or slow down before, or less brake preassure. I will try a lot of things and I will see what is better, and will improve lap after lap, race after race. I see how make anothers the raceline, AI have to learn about player racing and himself to improve

I think that this is the really Artificial Inteligience

Sorry for my bad english

They already do, don't they?

Nice name, BTW!

They dont are good learning....

I think you are lousing your time coz all here say "This is a ONLINE racing simulator" and is true but some ppl lik play in single player....

Sorry for my bad english.

The AI do already learn.

Run an AI only race for 20 or 30 laps and you'll see the laptimes increase a fair bit. Quit and start again, and they'll still keep those fast laptimes.

If you have a look in the game files, each AI drivers has their own 'learning' file which holds their knowledge about each track/car combination.

Quote from Crashgate3 :

The AI do already learn.

Run an AI only race for 20 or 30 laps and you'll see the laptimes increase a fair bit. Quit and start again, and they'll still keep those fast laptimes.

If you have a look in the game files, each AI drivers has their own 'learning' file which holds their knowledge about each track/car combination.

Decrease*

Do the AI learn from human user data or does it simply have a preset objective to hit 100% accuracy in a learning curve?

Character doesn't apply to these AI, If we could pick out characteristics be they good or bad it would add a lot of interesting variables into the decision making for the AI.

If the answer to my first question is yes, hundreds of thousands of characteristics and good/bad habbits could be taken from human input data. A possibility for over enthusiasm within AI character sets makes me wet my pants(good race).

If the latter to Q#1 is true then we have no character base to work with and would have to make characteristics from scratch by exaggerating the variables in the current AI behavior that make it a noob.

Quote from S14 DRIFT :

Decrease*

'get better' then...

Quote from CSU1 :

Do the AI learn from human user data or does it simply have a preset objective to hit 100% accuracy in a learning curve?

They just randomly experiment with different lines and braking points until they find one that produces a better laptime, then use that one from that point on.

whats the best car/track to use?

I thought Scawen said somewhere in the test patch forum that the AI learning was removed. I remember it used to be in LFS before, definitely, but I thought he removed it

I would search for it, but that's a hella lot of posts to be searching through

There is no more AI learning.
There used to be some learning and you could check your lfs folder for ".knw" files that were updated for each AI after a race (in which the AI participated of course ).
However as I see it the AIs were considering their line was an improvement if the split was better, so it quickly lead to good split --> crash behaviour, reproduced each lap.

The AIs ignored grip variation due to tyre temperature (main issue - they "learnt" a turn with optimal temp and reproduced the line and speed on cold / cooked tyres), were not able to pit, ignored damage and could not cope with clutch so Scawen improved all that at the cost of AI learning. They also never attempted to overtake, but considering the strange moves they pull now I'm not sure it is an improvement

On the other hand AIs are A LOT faster now - try a 10 lap race on WE1 with LX4 or LX6 for example, with pro AIs, and give them a good set...