Are there any good systems for doing Racing AI for a game, searched the Asset store and forum and there is not a lot of choice and a lot of older assets I found are no longer available. What does everyone else use for their Racing Game AI? Or is it a case of rolling ones own for best results.
I’m not going to be much help. I attempted a racing game a few years back, and abandoned the project when all the driving AI’s I tested would become unstable at high speed when hitting the turns. I don’t know if the situation has improved, but it sounds like it hasn’t.
NN would probably work fine. If it learned correctly.
2 Likes
Shameless plug:
https://www.youtube.com/watch?v=gEf9V03HWv0
In this case, I fitted the raycast detection tightly to the road layout - and there are no competeing drivers.
Nevertheless, the NN did eventually become pretty good with regard to speed and obstacle avoidance.
3 Likes
Shameless or not, this is very good practical example.
Just to add, learning process can be accelerated, with garbing finished laps from players.
@mbaske this is awesome! 
Thanks!
If someone’s interested in the ml-agents approach - I’ve posted a simplified version of the project at https://github.com/mbaske/ml-simple-driver
Does this work well for different racing lines, obstacles, entering/leaving pits etc?
My educated guess is, you would need train additionally for pit entering / leaving.