Evolution is extremely random

[idfkanymore]

For some reason, it seems like starting the code has a 50/50 chance to either be awesome by Gen. 4, or be absolutely useless for 80+ Gens.

Is it meant to be like this, or is there something wrong?

[Unarelith]

I noticed that too, but it's only because of the simplicity of this AI

I think it's because of two things:

• In the beginning there's a lot of randomly jumping dinosaurs, some of them are super dumb but sometimes they manage to go really far because they're lucky, and the smart ones will follow the lead, becoming dumb af themselves
• Score seems to be way too important, I think the amount of flying birds encountered matters too (one value per bird height), since sometimes you'll go further with a stupid dino just because you didn't encountered any birds

[white91wolf]

Can also confirm this.
Also tried to add some kind of panalty while doing interactions like ducking and jumping but not more then the default reward (and some other logic, that when ducking in air no panalty and so one..). so the model would learn to only jump and duck when really needed - so my intension... but somehow the model wont learn anything :(
i mean you got a lot of reasonable points in your code like the cullSpecies and massExtincion to prevent overfitting. But it seems to me, that the model still is overfitting somehow thats why the model isnt rly learning how to play the game when the first generations are useless.
Also i have a model which can avoid the most obstacles but when the first bird will fly in the dino just jumps right into it. Even after 100 gens the "bird height" input still wont be used..
i think there should always added some random new players/species. can you give me a hint where i can add this?