Amy Hennig Talks More About Struggling With EA's Frostbite Engine

If you want to make a game that is like the Uncharted of Star Wars, you're up against Probably 11+ years of tools that Naughty Dog created for the Uncharted series. I honestly don't know how long it would take to fill the gaps in Frostbite to make a cinematic 3rd-person adventure game, but I imagine EA just wasn't willing to wait for as long as it was taking.
I’m looking at star citizen and how long their engine and tooling process has been. It can be quite significant.
 
In case I was imaging things, looked up videos with somehow similar environments and now i am even more convinced.

In a multiplayer game, the engine is free from many duties, this is the precise reason wy the game can run at a higher framerate. It's not comparable. The single player mode of any Battlefield game is extremely linear with very basic interactions.

The only game comparable to Andromeda running at 60fps on console is MG5 and it's a cross-gen game with empty environments.

All games running at 60fps this gen on modern engines are either mutliplayer FPS or extremely linear games. There's one notable exception and it's Forza 4 but it's only possible on the X, and a racing game is generally less demanding than a third person game.
 
Last edited:
In case I was imaging things, looked up videos with somehow similar environments and now i am even more convinced.
Videos don't do justice, you need to play the games yourself and do close up comparisons. I did. I compared my BF1 to Andromeda and DA:I on PC, and believe me, BF1 has significantly lower poly count on everything, especially characters and geometry. Houses in BF are empty and devoid of life, while in Andromeda and DA they are full of details and objects.
Sure few key models might look good in cutscenes but bf has that too in single player.
No even in Single Player, BF is extremely limited.

I just don't see that assets quality/density and draw distances explaining andromeda performance.
That map in Andromeda is several times bigger than any BF map.
 
Videos don't do justice, you need to play the games yourself and do close up comparisons. I did. I compared my BF1 to Andromeda and DA:I on PC, and believe me, BF1 has significantly lower poly count on everything, especially characters and geometry. Houses in BF are empty and devoid of life, while in Andromeda and DA they are full of details and objects.

No even in Single Player, BF is extremely limited.


That map in Andromeda is several times bigger than any BF map.
..and that's an understatement. A map like Elaaden is literally 100x the surface of a BF1 map. As a matter of fact one of the maps/worlds in Adromeda (can't remember which one) is bigger than the whole map of DA:I.
 
Too bad Dead Space died before they could be a little more adventurous with a meta-AI. It'd have really ramped up the replay factor for NG+, especially given that they had a bit of a teleporter system for AI to travel between the ducts.

I would have much preferred an L4D MP for DS2 instead of that competitive stuff. Sigh.

Maybe Halo can do something like that for the Flood and... sigh. Nevermind.

Doom could use it tbh even if it's less about jump scares now.
 
Yea I'm not even sure if CPU is required to focus more on meta-AI. It's bit sad when I think about it, there's a lot of good discussion here and I do agree that CPU and other parts of engine development contribute more to gameplay than what we've seen so far this generation, which has largely been more or less the same as last gen but with significantly better graphics.

Not sure if you've played any of the new 'Legacy' titles, like Risk or Pandemic Legacy. Or Gloomhaven, but that's like the meta AI that has changed board gaming. Would like to see more development in that area.
 
Oh, innovation has stagnated alright.

When was the last time we saw a game as large as Daggerfall? 15000 cities scattered across a 160000 km2 massive map, populated with 3/4 of a million people that you can talk to and interact with? This was released 20 years ago, no other game has attempted to do such thing. And no, No Man Sky doesn't count.

Last game with honestly good AI? F.E.A.R. 1!

Last game with good total destruction? Red Faction Guerrilla, and maybe Mercenaries 2.

Last game with good massive environment interaction? Crysis Warhead, or Max Payne 3!
 
There's other aspects I've been thinking more about as well.
A lot of games have RPG type mechanics, like progression systems/unlocks.

But I don't see that as being the same thing as meta AI. But those are certainly aspects where innovation has stagnated. I'm surprised there hasn't been a 'legacy' style game on PC yet. I guess the only thing would be EvE Online, that's got some crazy persistence.
 
More comments from Hennig here: https://venturebeat.com/2019/02/22/...-making-a-video-game-and-inspiring-newcomers/

"Obviously trying to create a third-person cinematic traversal action game with an engine that was made to make first-person shooters, that was a hurdle. But we knew going in that that was the goal. We were going to put this functionality into Frostbite. A lot of the team was hired to do Battlefield, and so that was a bit of a cultural shift, to make this different kind of game. Normally you cache for the project you’re making rather than trying to — it’s hard to convert the people you have if that’s not their type of game."

I also think that Hennig is a bit biased from working at Naughty Dog and doing third person action games. When ND did the first Uncharted they probably had the most experience in the world doing third person action games (3 Crash games and 3 Jak games).
 
Yea I'm not even sure if CPU is required to focus more on meta-AI. It's bit sad when I think about it, there's a lot of good discussion here and I do agree that CPU and other parts of engine development contribute more to gameplay than what we've seen so far this generation, which has largely been more or less the same as last gen but with significantly better graphics.

Indeed. I’d be somewhat curious about RT being rigged for it because path tracing plus learning to prop things up on there.

Ultimately though, it’s probably a time and resource issue in having predictable behaviours for QA. We shouldn’t forget that games like FEAR or Red Faction or Max alcohol Payne 3 are single player games taking place in a tight corridor of sorts.

I haven’t played Crysis Warhead, so I can’t comment on the sorts of smoke and mirrors in providing the experience.

The Bungie Halo games are a decent example for potential in group behaviour, but again, are largely corridors and also somewhat design restricted due to determinism, which has implications for networked scenarios. Coming up with unique character types is also difficult beyond your usual weapon types.

One of these days we should sit down and talk about the things in Halo as far as setting up a sandbox vs scripted scenarios (e.g. scarab battle exhibiting a bit of both). It’d Certainly be interesting to train sandbox scenarios in an RTS environment and maybe try to scale that down for an FPS?

L4D has the benefit of being a game that only needs to swarm players at unpredictable times with very dumb ranges of actions. Gears horde can be fun with enough variety in game design, but again their range of actions can fall apart under scrutiny.
 

I think part of that is that, much more than in the 80's and 90's and even 2000's, it is far more lucrative for the really good programmers to go into other computing fields. And with the rapid rise of research into AI it has also become in many ways, a sexier field to go into than game programming. That rapid increase in AI R&D also means that gulf in pay between a good AI programmer and a good game programmer is just massive.

That said, I also agree with some of the comments in there, that the reliance on 3rd party game engines likely contributes to complacency with many game programmers.

That doesn't mean innovation doesn't happen or can't happen, but it does mean that far less people are incentivized to be innovative.

Also, the fact that when you hear or read about the game there is this obsessive focus on graphics at the expense of everything else (you can see this a lot even on this forum) that most large studios are sort of forced to put most of their development money and efforts on the graphics at the expense of other systems.

IMO, that last point, I think may be the biggest factor in why the games industry is relatively stagnant (with a few exceptions) in anything other than graphics rendering.

Regards,
SB
 
In a multiplayer game, the engine is free from many duties, this is the precise reason wy the game can run at a higher framerate. It's not comparable. The single player mode of any Battlefield game is extremely linear with very basic interactions.

The only game comparable to Andromeda running at 60fps on console is MG5 and it's a cross-gen game with empty environments.

All games running at 60fps this gen on modern engines are either mutliplayer FPS or extremely linear games. There's one notable exception and it's Forza 4 but it's only possible on the X, and a racing game is generally less demanding than a third person game.
I am sorry but there are not empty and MGS5 is totally a current gen game. Cross-gen game means nothing nowadays when a very demanding game like ARK: Survival Evolved runs on Switch. With time and money almost everything can be scaled down.
 
And with the rapid rise of research into AI it has also become in many ways, a sexier field to go into than game programming. That rapid increase in AI R&D also means that gulf in pay between a good AI programmer and a good game programmer is just massive.
let me step back. I'm not talking necessarily about technical wizardry in AI. But perhaps expecting new gameplay paradigms that may or may not be tied to more R&D, but perhaps just the degrees of how you influence the game is different. I think @DSoup could talk about what Bethseda does in that regards and that really just stems from their Elder scrolls series if I recall correctly.

But like things like Fable, even those concepts of 'time' changing and how what you did affected the world. I guess I was hoping to see more of that. The technology is there, but we didn't advance the game design concepts to focus on it as much.

A lot of it has been graphics and the reason for that being the big factor is because this generation more than last generation, has been about narrative and story in gaming.

And on that note, this generation of gaming is probably where as players we've made the least number of decisions in a game and definitely the least amount of thinking involved. With the exception of battle royale, and MOBAs, games have streamlined hard.

And with regard to Bioware, this is killing them. Their strength was about decision making. More decisions was better. And right now their titles have very little of that.
 
Last edited:
can you explain me what meta AI means? thanks
So the way the AI adapts to your style in MGSV is meta AI.

Or how in Shadow of Mordor how you torture and kill and convert and send messages, those types of things affect the Uruk leaders differently.

I guess it’s how your actions make an impact to say the whole scenario, as opposed to something very tactically.

So halo AI is very tactical, but predictably the same. It’s great tactical AI but it never changes regardless of circumstance. Thus it’s not meta.
 
Too bad Dead Space died before they could be a little more adventurous with a meta-AI. It'd have really ramped up the replay factor for NG+, especially given that they had a bit of a teleporter system for AI to travel between the ducts.

I would have much preferred an L4D MP for DS2 instead of that competitive stuff. Sigh.

Maybe Halo can do something like that for the Flood and... sigh. Nevermind.

Doom could use it tbh even if it's less about jump scares now.
What's a Meta-AI?
Is it a form of adjustable AI to player's actions?

edit: I see it was answered earlier
 
Last edited:
So the way the AI adapts to your style in MGSV is meta AI.

Or how in Shadow of Mordor how you torture and kill and convert and send messages, those types of things affect the Uruk leaders differently.

I guess it’s how your actions make an impact to say the whole scenario, as opposed to something very tactically.

So halo AI is very tactical, but predictably the same. It’s great tactical AI but it never changes regardless of circumstance. Thus it’s not meta.

Thanks for the reply but I still don't get it and the MGSV/Halo comparison is not clear to me.

Is it a form of director AI like in Left 4 Dead?
 
They mean high-level, overarching game behaviour changes. Typically AI is immediate and tactical, choosing to hide or attack, where to position, etc. What's being described as meta AI is a change in behaviour on a wider level. So, for example, how NPCs respond to you based on your actions. Or in a shooter, how the enemies could adjust their tactics based on how you act.

Adaptive AI is rare. A game plays pretty much the same for everyone who plays it. A good meta AI will give different players different experiences as it changes the game based on their play.

I also don't think 'meta AI' is at all a good term. It's AI, just on a different scope. I guess it could be meta AI if the high-level decision changes aren't trying to be intelligent decision-making and are just tweaking AI parameters to adjust the AI behaviour.

Edit: As to why we don't have better 'meta' AI, I reckon it's because:
  1. It's a bugger to test and maintain. In hundreds and hundreds of hours of testing at the studio, aberrant behaviour may never appear, but among millions of players the AI could break and end up doing annoying or weird stuff. Things like killing progress-essential NPCs or breaking the game's economy don't go down well.
  2. You don't sell any more games. People love graphics and physics, so interactivity such as GTA and LoZ:BotW, gets everyone excited for propagating fire etc. while that stuff is a lot more deterministic and easier to design for.
Until someone invents in a drop-in SuperSmartz AI library* that can fit in with any game comfortably, providing intelligent behaviour without screwing it up, I imagine it's something all devs will give a wide berth as being more trouble than its worth.

*AlBran's AIBrain
 
Last edited:
Back
Top