Download

Blueprint Nativization with Behavior Trees/Tasks and A.I Sensing Possible?

I know that blueprint nativization is fairly newish in Unreal Engine, but I was curious to see who has nativized some of their blueprints and what they were able to nativize while accomplishing a successful build of their game. I attempted to nativize a few blueprints and my project would not compile. I searched the web and found very little info about experiences with nativization. I know c++ really shines in these frequent maths and updates…

I am getting concerned because all of the A.I is in blueprint unfortunately. I tweaked most of the behavior updates and added more delay to their sensing, but blueprints are still struggling. The game thread is maxing a 6700k with around 15 bots(in an fps prototype). I know the answer is probably convert all of this to regular c++ code but I wanted to see if anyone has successfully nativized behavior tree tasks and A.I. sensing updates required for a believable behavior.

Also if anyone knows how complex a shooter A.I. should be let me know. I have studied several AAA shooter games and their A.I. seems pretty basic; is that just necessary because of cpu load?
Thanks for any help,

Sure your AI BT isn’t too complex?
I’ve learned that a few tasks is needed to get some awesome results.
It’s like telling a guard to not let anyone in and that’s it and it will do some crazy ■■■ s**t if left to it’s own devices. :smiley:

Are you sure that the calls are as simple as possible and you not calling a crazy function every tick?

Let me know as I have coded some pretty cool AI.

Well I have only worked with ue4 a few months and my c++ is not so good yet; I understand the main uses of things,

It is just basic a.i. using pawn sensing, shooting, behavior tree tasks for checking player distances, strafing etc, cover firing, pathnodes,

I tried moving stuff from on tick to different update intervals(0.05, 0.1, 0.5 secs), still limited by the single gamethread…Maybe should instance animations somehow?

I want to expand a.i. further using goal oriented decisions, probability weights, but seems like already having to limit a.i. instead to squeeze gamethread performance isnt right…

I would cherish any advice honestly bc i think game a.i. has not advanced much since 2007, not to discredit anyone who made these still great games, I just want to make it the smartest I can,

​​​​