Hello,

New bots could be created for each track specifically.

Read AlphaGo Zero: Learning from scratch.

Marino

I don't think new bots is necessarily needed. They follow a pre-set race line which is available on every track so theoretically you can just trace a new line based on a very fast human driver. Look at the WR records, they are pretty insane as it is.

But you bring up something that machine learning can do which is adapting to different setups and learning their limits. Considering the Ai can run a variety of setups provided by the player and some setups allow for later braking or turn-in etc.

But I think the Ai is challenging currently. Not everyone may agree, but in that case just handicap yourself with intake restriction or a lower performance setup.

they are too slow in corners

It would be a nice science project to test LFS engine and Google's DeepMind, and also a good marketing for LFS.

Now there are three levels of AI to choose: beginner, intermediate and pro. With neural networks you would have an individual AI for each car and for each race track. There are 20 cars and 8 tracks. This means 160 different boots.

Quote from Marino108 :

It would be a nice science project to test LFS engine and Google's DeepMind, and also a good marketing for LFS.

Now there are three levels of AI to choose: beginner, intermediate and pro. With neural networks you would have an individual AI for each car and for each race track. There are 20 cars and 8 tracks. This means 160 different boots.

There are more than 8 tracks. You have to factor in every different configuration which is probably nearer 50 which means 800 different AI memories. Btw I think it's perfectly possible to create an AI driving machine using Outgauge and emulated inputs if anyone is up to try this on their own..

Quote from pärtan :

...Btw I think it's perfectly possible to create an AI driving machine using Outgauge and emulated inputs if anyone is up to try this on their own..

The project is in hiatus at the moment, but: https://www.lfs.net/forum/thre ... acing-Simulations-Project

Quote from pärtan :

You have to factor in every different configuration which is probably nearer 50 which means 800 different AI memories.

More - before Westhill was updated,there were 940 official combos. Then Westhill got few more configurations,also Blackwood has one additional after update. So it's close to 1100 these days.

By the way I also agree to improve LiveForSpeed AI, because they are even necessary for training and apprehension of many people who are training and prepare for an online test (who is doing it for the first time).
I think AIs are very important and should also be wagered