Can an AI survive Five Nights at Freddingus'? 

After his life if of crime, Dingus was punished by the god Asmodingus to push a boulder up numerous hills. After finally escaping his divine tormentor, Dingus swore off crime, instead choosing to assist a struggling pizzeria in protecting their property overnight.

They remembered his passion for work from that video with moving the stone up!

Dingus parole notes: "His low level of intelligence is a challenge, but also lends credence to an argument for low risk of recidivism. Parole is approved. Recommendation of the board is that of placing Dingus in a half-way house - there is currently tenant availability at William Afton's home, due to recent incidents resulting in... vacancy. Extra care from Afton's associate, Henry, will surely ensure Dingus' success."

Frfr

How good can he get at that game?

A Dingus for all seasons.

Ding-ception

Almost read this as “Metal-Dingus” and now I want to see them training Dingus to race a metallic copy of itself

@@ScuttleBugReal A metal copy of dingus, this idea kinda reminds me of the time someone made a copy of a certain blue, fast, ring grabbing creature...

​@@LeHasnurg strange, isnt it?

I really should have fit that line in

@@Dingus_Labs for the next time !

Acthually, iths the endinguning of '83 🤓

​@@Dingus_LabsNow we need is five nights at freddingus 2

Yeah unfortunately this video had to be a composite of seperate models because I realised I messed up the freddingus AI

⁠

Maybe if he had it on a time limit, like only getting rewarded once every 5 seconds, or only once after seeing an enemy

Lol the ai must be thinking it’s fnaf 2 withered foxy

Yeah my guess is the reason it kept opening and closing doors on animatronics is because it was farming the reward for closing the door.

I got pretty resistant to it after running like 160+ attempts, but every once in a while it would get me too

Yeah, I'm interested in the reinforcement learning and game design stuff, but high-strung enough that even this parodic version startled me pretty badly a few times. Guess I just don't do well with suspense and sudden movements/noises. Which makes sense -- earlier tonight I spotted a spider about a meter away from myself, just innocently climbing out of a container on my desk, and the noise I made _scared the spider away_ somehow. It dived back into the container and hid, which made me a little embarrassed as I picked up the container and (keeping eyes on the spider the entire time so it couldn't climb up onto me) brought it outside to shake the spider out. Ah well, there's a reason why I avoid most 'horror' games and similar (really, I would classify them more as 'terror', as in the acute startling reactions more than the overall creepyness. FNAF is an extremely straightforward example -- it's _all_ 'jumpscares'. There's a few horror media that are more restricted to horror proper that I can handle better, think Fear and Hunger or other games that try to disturb rather than startle).

Yup. They got me too.

I do like that idea! Definitely requires a bit of effort to code up but would be super fun. This Going on the Dingus potential activity list!

Fun fun! Glad my idea has been potentialized!

A 4 year old would understand FNAF SB more than any other FNAF game.

It doesn't hear, it just knows when the sound is played.

Yeah apologies, I wasn't super clear about rewards and how I handled them. You are correct that I can apply weighted rewards, there is a negative reward for opening the doors when it's unsafe, just like how there's a negative reward for checking other cams when freddingus is at 4b and he eventually seemed to learn that, and the negative reward is higher than the reward for closing the doors when it's dangerous outside (otherwise he'd just spam the doors open and close), but it is interesting how it took a long time for him to stop opening the doors when danger was in front of them compared to how quickly he learned not to check other cams when Freddingus was at 4b.

⁠@@Dingus_Labsperhaps it was just easier to understand that specific scenario was leading to a negative of losing. Opening/closing the door could be leading to a positive or negative or a bigger negative from losing? Just from experience of fnaf I think for a person at least it’s a lot easier to learn say Foxy’s mechanics than any of the others. You know if you look at the camera then you must close the door asap and can open it again right after* Compare that to the others who will appear at the windows or cameras and you have to check those then close the door and when they go is more guesswork than anything else. My two cents is that spamming the door was harder to learn the background mechanics of. For us humans we’ll probably just close the door then wait to check it’s clear and then open it. The AI is having to take many attempts to build that correlation between the mechanics and that spamming is the least optimal strategy.

You could also go for a adversarial network approach. Not sure if whatever machine learning system Dingus is made in supports that, but I'd love to see that if it does.

I have to admit I feel the same, I did program it as per my understanding of foxy behaviour but I'm starting to wonder if I've gotten the behaviour slightly wrong (wouldn't be the first thing), I will have to look into this further. If there's a glitch it would be with the "lock" during cams mechanic and how resuming movement checks work. It's definitely much more aggressive on night 6 so it's respecting the movement check and level mechanics

Characters Broken Freddingus Broken Dongus Broken Dungus Broken Kev Broken Gold-Coated Freddingus Updated Freddingus Updated Dongus Updated Dungus Kev's Failed Update Cuboon Cube Dingus Puppet

they SHOULD come online soon, there can be a lag between when youtube publishes the video and when it finishes creating and making the subtitles available!

@@Dingus_Labs oh, thanks! i guess i arrived a little too early haha

As a certified and professional watcher of pro FNAF Super Custom Night walkthroughs (correct game name?) this guy has the makings of a great streamer.

I believe it would, except maybe with an AI that learned to perfectly play 4/20 mode as certain behaviours become consistent then (like the time an enemy remains at a window), honestly I should just do it and see when I've got some time

I was about to go on a pretty big rant of how this wouldn’t be balanced but it could work is we assume AI’s don’t improve past a certain point unless the enemy gets stronger. Basically, as long as the animatronics don’t get way too strong for the guard to even get any experimentation in between deaths then they should both get stronger in an almost catch up game.

@@Mayflower-Yevdepends on implementation of it, how would the GM controlling the animatronics adapt and/or try to beat the guard? The game itself already has a fair amount of rng and weighted rolls that change each night to spice things up and force the guard to adapt as it progresses.

Yaass

agreed

he Already purple

yes