Hi everyone!
I'm a student from the "Politecnico di Milano" university, and I'm preparing my thesis. I decided to work in the field of Artificial Intelligence, playing around with Neural Networks and Genetic Algorithms, so I choose to develop an AI controller for VDrift with the ability to drive through different tracks all by itself.
This will resemble the work done with TORCS, resulted in an international competition (
http://cig.dei.polimi.it/)
Naturally, this implies training a specific neural net (I'll use the NEAT library) with chosen inputs. Now, I know from a previous post that all the information about the car can be accessed and modified through the CAR class. I've also downloaded the last revision from the SVN repository and studied the AI class structure.
At this point a series of questions popped in my mind:
1. Where actually I do implement my bot? Does all the code go along (or replace) the one in ai.cpp file?
2. How the AI methods is called during the game?
3. Is it possible for VDrift to run in text mode only, and thus simulating thousands of laps (with only the AI car) in an acceptable time frame, in order to train the neural net? This is a critical step.
Thanks in advance for the help! I think that all the work may be integrated into VDrift.
Best regards,
Antonio Caiazzo