Hello, Why didn't you use something like rayCasts(but on unreal engine) as the inputs? Adittionally, if you want to make them better you can do things that are harder to implement like making them learm from footage of you playing (supervised learning) or make them remember what happened before (something like LSTM).
@MaxMakesGames
Жыл бұрын
Hello :) I did use raycasts as input to let the AI know if an enemy is in sight or not, but it never learned to use that to decide when to aim and shoot and instead just always aimed and shot. Supervised learning is definitely a good idea ! I didn't try it because I wanted to see how they would play on their own and I didn't want them to pickup my playstyle. I've never heard of LSTM, thanks I'll check it out.
@nadavmihov
Жыл бұрын
@@MaxMakesGames Yes it is a downside of supervised learning but you can start training them like that, then take the weights of the best neural network and train them in a simulation where they are judged only by their performance and then they may develop their own playstyle.
@nadavmihov
Жыл бұрын
@@MaxMakesGames You can also simply give the AI inputs from the last 3 frames instead of only the current frame. This way the AI would be able to know where the enemy is headed because you can't know where he is moving or what he is about to do from just 1 given moment.
Пікірлер: 7