I’m currently trying to learn Neural Networks, I’ve successfully made an AI NPC that can learn how to go through a maze, however, now I’m trying to make an AI NPC that can go through a basic parkour.
I figured out how to do most of it, but I still don’t understand how I can find the right moment to make the NPC jump using the output.
As i understood it’s AI that learns how to do things, when it loose he gets “punished” (that means he did it wrong, and he need to try other way to achieve good result). So it learns and every time it does better result it getting awarded, so he will try to use that way, but it will still try any other ways, and if it won’t give something useful - he will try last succesful ways. So it will be learning and do better and better every time, like something alive. So, it takes so much attempts, i would say thousands attempts to learn how to do something, but then it looks alive. Like it doesn’t use exact way it used before, beacuse humans can’t do that too. Also as i know they don’t know where they need to go and so on, they just getting awarded if they walked into the right way, and punished if not, so they like humans, they see - then use this information and makes results where they should go, what they should do, and then code just checks if it did right, and then choose award or punishment (or both, like if it walk into wall but it got closer, it gets punished for walking into the wall and awarded for being closer so it can get more information how to do it in future)
But i’m not sure it exact what i said, this is just how i understood.
This is not really like, a solution, but a suggestion. Neural networks are very computational so creating an NPC that uses Machine Learning will use a lot of memory in your game rather than hard coding AI.