该视频显示了Neat Neural Network AI进化了超过60代的19个基因组。 黑船使用手动调整的PID控制箔角,而白船则使用整洁的AI。 健身度是通过假设箔纸使船只走得更快从而覆盖更多空间的方式,在模拟的1个虚拟分钟内使船只达到距离的程度来分配的。 模拟的原始总时间约为18分钟。 仍有许多工作要做,但至少这显示出一些希望。 #独立游戏#rustlang#帆船#制作#ai#整洁#neuralnetwork。
航行模拟器-NEAT神经网络AI进化加速
6 comments
Comments are closed.
I wonder if there is any way to give them some useful information about an upcoming wave crest. I guess that is pretty hard/expensive.
Very cool to see
Veramente impressionante. Ma in questo modo la IA è capace di migliorare """all'infinito"" arrivando al punto di essere praticamente imbattibile, o per qualche motivo può anche regredire?
Seconda cosa, in questo modo la IA continuerebbe ad "imparare" ed evolvere anche una volta arrivata sui nostri PC, o la fase di apprendimento può essere fermata (da in fase di sviluppo quindi)?
Does every boat after every run have the same knowledge as all the others? If yes, then it seems weird to me that there still is so much difference between them. I've got a feeling this NEAT wants to go as close as possible to the limit, but only a bit over the limit gives a nose dive. That in combination with the irregularity of the waves, makes a lot of them nosedive. They are always aiming for the 100%, and from what I understood from your livestream about this, there isn't really a way to get them to aim for 90% for example to limit mistakes, or to make it slower for slower sailors.
Like always though, I am probably wrong as I have 0 knowledge of programming, let alone AI 😛
Neat! 😛
Awesome! How does it work? Did you use reinforcement learning?