I’m attempting to make an AI that learns how to beat advanced obbies with things like wraparounds and other advanced tactics. I know what I’ll have as input parameters and output parameters, but I’m not familiar with neural networks, and any documentation available for Roblox neural networks doesn’t explain it very well to me.
For more information, the input is the camera’s position and orientation, and NPC’s position, orientation, velocity, HumanoidStateType and whether it’s in shift lock or not (which I plan to simulate). The output would be whether it presses the W, A, S, D, Spacebar or Shift keys, whether it scrolls in or out, and how much it rotates its camera. I simply need to figure out how to pass the input through the neural network and get the output.
I would recommend you do a thorough investigation of neural networks before you attempt to do this.
For starters, I don’t think the inputs are gonna allow it to beat anything in an intelligent fashion. You don’t seem to provide it with any information to know anything about what it is trying to achieve, effectively rendering it blind. Unless your plan is to have it learn only one map, in which case it should be fine. If you move a single part the AI would have to re-learn.
I think there are some libraries available for Neural Networks, however I haven’t researched any Roblox libs for this and can not say if they are worth your time or not.
I would probably also try making a simpler AI first. It’s been a while since I dealt with AI, but if I remember correctly Q-learning - Wikipedia is a good approach, even for beginners.