PC What Would Be the Best Zombie AI?

[Gyonyoru]

Special zombies should be intelligent, able to find quick mistakes in your base layout and exploit them (Leaving you to make defense-holes on purpose to lure them in first) [Cops, Wights... non ferals]

General zombies should be crazy in number, slow, extremely easy to kill but are attracted to eachother by sound creating wandering hordes.

Ferals should be quick, equally as dimwitted, but take impressive amounts of effort to kill (Usually better through traps or working together with a teammate to dispatch them).

 

[LuckyStar]

Regardless of what they do with the AI we're going to beat and abuse it.

[Oh that doesn't sound right.]

I don't see any possibility of having an open world game or voxel game with an AI strong enough to really stump us.

Both types of games are already using a lot of resources and thus.... there's just not enough left to work with.

In a static world game, that is also very linear and compartmentalized, you can have a very strong AI because you've got the player in a box AND you've got a lot more resources to work with.

Maybe in 20 years when technology allows it we'll get a game like 7DTD that really makes us scratch our heads.

For now... well... there's only so much we can expect.

 

[Zourin]

Survival is a matter of exploitation. Doesn't matter if it's a deer or a zombie.

That said, as a matter of this being a GAME, there are two forces that need to be balanced.

1. Intelligent Offense vs Intelligent Defense. Smart zombies that have discernable intelligence are vulnerable to having that intelligence exploited. This is the current state of A17, and it is in slightly poor shape with the advantage to the zombies due to an out-of-whack gamestage translation (advanced zombies FAR too early).

2. Stupid Offense vs Stupid Defense. Zombies that simply smack their way to you with little regard for obstacles. A17 is NOT balanced at all for this kind of play, with the removal of any practical means of defending a base of operations from a blunt 360-degree offense by zombies capable of punching through reinforced concrete in seconds.

The addition of 'variability' needs to come with some consideration for how 'sprinkling in dumb zombies' requires adjustments that currently do not effectively exist in a17 for a 'defend the base/ground level perimeter. That's what Log Spikes were in previous alphas: static 360 defense that defended against 'random smackers' while your attention was focused on the primary hotspot. They weren't there to stop hordes, they barely did damage by themselves. Log spikes were placed to stop that one dumb zombie from creating a breach, in addition to thinning the mass of hitpoints barrelling to your position.

 

[calebelt]

Complete agreement.

 

[AtomicUs5000]

Regardless of what they do with the AI we're going to beat and abuse it.[Oh that doesn't sound right.]

Having AI that is so smart that there is nothing we can do about it would not make a good game. The idea is to beat the AI, but not because there is some serious flaw in it. I believe the current AI is pretty good and almost good enough... but it still needs an element of randomness. Randomness in the AI would not use any more resources, but would make things more unpredictable. You could take advantage of some weakness in the AI for the most part, but it wouldn't mean ignoring other aspects such as structural integrity. You should have to keep everything in check.

I don't see any possibility of having an open world game or voxel game with an AI strong enough to really stump us.

Both types of games are already using a lot of resources and thus.... there's just not enough left to work with.

In a static world game, that is also very linear and compartmentalized, you can have very strong AI's because you've got the player in a box AND you've got a lot more resources to work with.

Maybe in 20 years when technology allows it we'll get a game like 7DTD that really makes us scratch our heads.

For now... well... there's only so much we can expect.

I agree that there is a lot going on, but the game is far from optimized. I don't develop games for a living, but I am a software engineer, and artificial intelligence was my area of focus in school. Good AI for this game does not need to be a complex set of algorithms, it just needs a few triggers where decisions are made and those decisions are chosen with an element of randomness. This is essentially the initial iteration of what I consider the first real AI in the game. I don't think we have to wait 20 years to get enough improvement to make the AI not be as unintelligent as it is right now. I think there will even be improvements before the official A17.

 

[AtomicUs5000]

Complete agreement.

Absolutely. With this, you don't even need to really make a separate calculation. You establish these branches, but you apply some weight to block durability and some weight to distance. A random choice is made to apply more weight to blocks and so the longer subpath is chosen, or the random choice applies more weight to distance, so the shorter subpath is chosen. By shortening the distance, and forcing a recalc... or by forcing a recalc after a block is broken (I think this is already there) the AI doesn't need a lengthy algorithm... it just needs to run the algorithm several times during the zombie's lifespan. Recalcs do not need to happen immediately. They can be queued so that only one path is established at any given time to avoid stuttering (I don't know for certain, but I believe this is the reason for the stuttering sometimes when a bunch of zombies spawn in.). Additionally, this does not need to be done for every zombie. Of the spawned zombies, only a set few might be responsible for pathfinding while the others follow bread crumbs. If one of these zombies die, leadership is popped from the queue of all zombies. I don't know for certain, but games like They Are Billions, where there are thousands of zombies on the screen, probably follow a similar method. There is no way each of those entities are following their own AI paths when not needed.

 

[Verne]

I don't like changes, TFP ruined this game for me. Zombies used to be stupid and should remain stupid. Whole point of zombie game is facing stupid enemies in big numbers. What we have now is some kind of entity with wallhack that pretends to be a zombie.

Underground base was an exploit? Turn zombie off and end effect is the same however this way you ruin immersion and whole point of playing survival game.

Oh right, you can also spend whole horde night on bicycle.

 

[Viktoriusiii]

Don'T forget that they will do EVERYTHING to avoid spikes.

Like seriously. They dug through 25k of blockhealth last hordenight, just to avoid like 3 spikes and 3 barbed wire. THAT THEY DIDNT EVEN SEE BECAUSE IT WAS BEHIND A CORNER!!!

 

[Kalarro]

1. Isnt as smart as to find exact best path to you by magic

2. Is smart enough to not get caught in an infinite loop anywhere.

A16 was nice regarding that they attacked like real zombies. Straight to you, trying to bash anything in their ways. But, they were dumb in regards to getting stuck in infinite loops often.

A17 new AI was a feature I longed for, thinking they will be smarter as to not get stuck in an infinite loop. Instead, they now find the best path and all bash teh same block, making cool defenses of previous builds not useful. AND at the same time they get stuck in infinite loops even more.

They fixed it the wrong way IMO. They needed to fix the "dumbness" of beeing an AI, not of beeing a zombie. But they fixed the dumbness of beeing a zombie and made the AI dumbness even worse. Not sure if you get what I mean.

 

[AtomicUs5000]

I don't like changes, TFP ruined this game for me. Zombies used to be stupid and should remain stupid. Whole point of zombie game is facing stupid enemies in big numbers. What we have now is some kind of entity with wallhack that pretends to be a zombie. Underground base was an exploit? Turn zombie off and end effect is the same however this way you ruin immersion and whole point of playing survival game.

Oh right, you can also spend whole horde night on bicycle.

But how do you solve it?

 

[calebelt]

Don'T forget that they will do EVERYTHING to avoid spikes.Like seriously. They dug through 25k of blockhealth last hordenight, just to avoid like 3 spikes and 3 barbed wire. THAT THEY DIDNT EVEN SEE BECAUSE IT WAS BEHIND A CORNER!!!

Each trap, thorn and barbed wire must have its own value. Generator accidents can choose what traps he will ignore every event.

 

[ArcticPrism]

The biggest problem I had with the old zombies was the lack of pathfinding. I'm happy that they can reach the player a lot better than before now. The other issue I had was that if they couldn't reach the player they'd just stand around under them. Being more consistent about breaking down nearby blocks if they can't reach the player would be nice

I don't think zombies should know how to avoid traps except for maybe some special zombies.

 

[LuckyStar]

Having AI that is so smart that there is nothing we can do about it would not make a good game. The idea is to beat the AI, but not because there is some serious flaw in it. I believe the current AI is pretty good and almost good enough... but it still needs an element of randomness. Randomness in the AI would not use any more resources, but would make things more unpredictable. You could take advantage of some weakness in the AI for the most part, but it wouldn't mean ignoring other aspects such as structural integrity. You should have to keep everything in check.

Oh by no means would I want to see an unbeatable AI.

I agree with you that it would be no fun.

I'm simply talking about something that would really have us scratching our heads.

Also you point on randomness is something TFP seem to agree on as more than one Dev has posted that they are adding a randomness element to the AI.

Looking forward to seeing how it plays out.

I agree that there is a lot going on, but the game is far from optimized. I don't develop games for a living, but I am a software engineer, and artificial intelligence was my area of focus in school. Good AI for this game does not need to be a complex set of algorithms, it just needs a few triggers where decisions are made and those decisions are chosen with an element of randomness. This is essentially the initial iteration of what I consider the first real AI in the game. I don't think we have to wait 20 years to get enough improvement to make the AI not be as unintelligent as it is right now. I think there will even be improvements before the official A17.

I respect your opinion and knowledge. Sounds like you've also spent time with AI's.

Here's where our feelings on the matter differ.

I see a stark difference between what people "say" is possible and reality.

Too many times I've seen AI's that just never performed as intended and too much of this unicorn-esque hopefulness rather than reality based logic.

I don't agree with you that AI as it is now in games like 7DTD can perform well.

They DO in fact need complex algorithms because we're dealing with near infinite possibilities in an open world/sandbox game.

I just can't see what's required running on the hardware of today.

Look I'm not saying you're wrong.

I think what you are saying applies to static world games or games that have a much more guided and limited player area.

It's okay... we don't have to agree.

Truthfully I hope you're right and I'm dead wrong.

I'd really love to see the AI really give us a go!

I just don't have my hopes up.