A bubble doesn't make AI more performant, on the contrary. ED have avoided a bubble system by design choice and they are not going to ever implement one. It would be their instant death on the military market.

Having not played a lot of flight sims in my early gaming career, could someone explain/describe what bubbles are in AI design or implementation? This has been a really interesting read on campaign design but you guys lost me there!

Perhaps a few words on why the military market wouldn't find it an acceptable design or compromise as well?

Thanks in advance

Bubbles ... think of a sphere existing around your aircraft.

Inside that bubble the AI operates individual units (turn, fight, drop bombs, etc) where each air unit, ground unit or whatever exists in the virtual space.

Outside of that bubble the AI operates abstractly. Example: Mig-29 flight versus F-16 flight ... what do the stats say that outcome should look like .. well lets skip the blood and guts fighting and just make the result 'happen'. No units exist and math is used to abstract out the fight.

Math still exists in both but inside the bubble it is more detailed and outside way, way less detailed. That lets a whole war exist while only having the units that can directly impact the player be operated in detail. That saves a ton of computer AI cycles.

Regarding to what to what Fridge said, I want to add an another practical example of what could happen in the "AI Bubble architecture" (this actually happens in BoBII):

- Imagine for example a flight of lets say 4 AI aircraft flying in formation. When those aircraft are flying inside the "Bubble" each aircraft has its own AI and act like 4 different entities as expected.
But when and if those same 4 aircraft are flying in formation outside the "Bubble" than instead of behaving as 4 different entities (each with its own AI) they will act as a 1 (one) unique entity, so you'll only have 1 AI controlling 4 aircraft (those 4 aircraft will fly exactly the same way). If the player could see this, this would look awkward since it would appear that those 4 aircraft were connected or "glued" by some king of invisible strings/cables (or whatever). But since this happens outside the player's bubble (which should always be outside the player's viewdistance and/or outside the player's aircraft sensors), the player would never see such thing happening.

The "Bubble" solution has the obvious advantage in the situation above of 4 AI aircraft flying in formation outside the player's bubble that it would take 4 times less computing power (cycles) to make calculations for those same 4 AI aircraft (when flying outside the "Bubble" that is).
Now multiply this with dozens or even hundreds of other AI air, ground and naval formation and their associated units which must exist in order to make a virtual battlefield believable and you could certainly make a "wild guess" on the tremendous computer power that could be saved with such an approach as this "AI Bubble".

Thanks Fridge, got it.

If I understand correctly, it is exactly like the reality that I live in..only things that directly impact me and my situation are important, and everything else is immaterial or inconsequential. Got it.

Thanks Ricnunes. It would seem to me that the leap forward we need is developers fully taking advantage of multiprocessors. That way events can take place and be tracked in the gamespace without the player needing to be there or the need for bubbles, with far less of a performance hit.

Concur, TankerWade.

The most famous bubble has been Falcon 4.0 because of its dynamic campaign and the fact that there used to be an actual bubble slider in the config menu that could cause crashes if set too high. But bubbles have been in dynamic campaigns since there have been environments with more objects than can be tracked at once. Even if you take a F-22 ADF/TAW mission and convert it to multiplayer, you will find a fairly small bubble around the host aircraft.

Falcon 4 was written between the Pentium II and III generations; with quadcore now (and even hyperthreading i7s) the AI should be relegated to another core. If it's too difficult to code, then I imagine an easier solution being to have a "core bubble" where AI outside the player bubble would be computed in high fidelity on other cores and moved to the primary core upon bubble entry.

Exactly. That's your umwelt. :-)

Another interesting direction might include some sort of grand campaign directed in some part by humans instead of AI. I mean, with Combined Arms we have a little bit of influence (or a lot? - I dunno..I have it but don't use it much) on how and where units go. It isn't hard to imagine an AWACS type module or plug-in to CA that would allow a human to control the flow and respond to the enemy movements (whether they be by AI or by another human "General"). If I'm not mistaken, ARMA3 uses some sort of on the fly mission tweaking (Zeus). I wouldn't want to be able to just create challenges out of thin air though. My mantra is always all the units on the map, so the "General" would have to use those resources within the scope of that. That all might be going more toward MP dynamic campaign, but I would imagine something similar could be developed for SP. Of course, I'd like it to be an opt in/opt out type of thing because I don't really *want* to play General or have to direct the movements of an entire campaign. It'd be just like me to put up a platoon of HMMWVs against a platoon of T-72s...

BeachAV8R

I've always wanted to play a FalconAF campaign with 5-10 people each given a squadron and a division to control. The reason is the only good quality control is human intelligence but the workload for doing it all by 1 person is too much.

But it is too lazy in design to force the player(s) to do all the gruntwork. Even in Starcraft the AI does some things by itself. I really enjoyed playing Company of Heroes because there was a lot of depth in 2-4 squads flanking and such. The problem is the game is 20+ squads all needing constant attention. It's a great game except I need about 10 other people to help me play it. So instead all the precious potential is wasted on "box-select, right click attack" because the workload on the player is too high.

Control is best scalable, allowing the player to nudge and micro if they want to without requiring it. For many months I've been trying to find out how to play BMS without fragging every sortie myself. The fully auto ATM is low work but low quality. Fully manual is the opposite. I've been trying to figure out how to walk the middle ground to get acceptable ATO out of minimal work. It's a struggle.

We have used the Zeus feature with varying levels of success in our SimHQ Arma 3 sessions.

-We started with just an open mission (a template, if you will) and a Zeus player "dropped" objectives, Opfor, and assets, to the map for the rest of the players to "solve." That was fun for awhile, and was a good learning method for those who wished to learn the Zeus command set. But as Beach alluded to, that gameplay felt random and disconnected, like we were chasing a carrot or just fedexing death around to different locales.

-We have also had our bouts with Zeus players punking us during a mission, sneaking up on and infiltrating our squads or sniping us from a camping spot. But we are a good community and we talked it out and that has gone away.

-We have found the best use of the feature is to give a bit of direction to the forces that are arrayed against the players. Getting the AI unstuck from a bad script or changing it's dispositions to prevent a particular mission from playing out with a low excitement-per-minute (EPM) factor. We all have busy schedules and wondering why we aren't being shot at is no ones idea of a good time.

-Its also useful to scale a mission for larger than expected player turnouts. In that case it is nice to be able to use a mission that wasn't set up for a large number of players but otherwise has a nice setup. The Zeus can also drop mission assets to the players if the going is rougher than anticipated, which can feel like HQ hasn't totally forgotten/not give a #%&*$# about us as it usually does.

So yes - I would like to see a more robust set of tools like this in DCS. I know Eno and Wrecking Crew have legendarily fought with convoy AI and the bridges they hate to cross. And Gamemaster AWACS directing enemy fighter sweeps would be very cool. When used right it's a good tool and I am glad Arma has it.

Tis a shame 'bout the AI convoys stacking up. Soo many missions gather dust for just this reason, including my five-mission !S'paign campaign set.


Thanks for the mention, TankerWade!

Wasn't there talk of an AWACS for F4 back in the day? The F-15E was the only announced addon for it, but there were plans discussed for A-10s, AH-64s, and an AWACS doing C&C as far as I recall.




The Jedi Master

I could be wrong, but I believe that F4AF had it where if you joined an AWACS you would get the campaign display but with vector graphics and map outlines instead of the full color view of the campaign map.

I'm curious why bubbles are less performant than full AI for every actor. Would you mind elaborating?

This is an example of how you can't make a product that is best for customer X when the real customer is customer Y.

A bubble makes the AI better because it allocates scarce resources efficiently. If 100% of your AI CPU budget goes to 100 units they get 1% each. With a good bubble system 10 units can get 9% and the AI that don't matter get 0.9%. The AI experience is improved by biasing fidelity to where it's needed.

This is obvious with 3D vertexes in LODs. Everyone agreed that LODs were the smart way to allocate rendering power. Why not treat AI the same way?

Which would be easier to implement - a bubble system or dumping AI onto it's own core?

Nate

When you are rendering a LOD, you are only rendering for one particular screen (the end user, or client). When you compute an AI, what is far away for one client may be in the Cheerios of another client. In order to implement a bubble in multiplayer, you would need to have DCS apply the most resource-intensive AI for all clients based on the proximity of each closest client to the AI.

Let me fix that for you with regard to LOD.

When you compute LOD, what is far away for one client may be in the Cheerios of another client.

Example: Arma3 or DCS where a unit far away from me has a low LOD while it is next to my multiplayer friend who sees it in high LOD.

Same would go for AI. It's just an object. Now ... imagine how lag looks to two players flying in a virtual world. One player sees the other player where he/she was a few milliseconds ago (or seconds, whatever). Imagine the frustration of shooting at him/her where they were instead of where they are. Example: Two players get too close and one rams another ... but on the other screen, that ramming didn't occur.

You think that's a problem. Well that kind of condition with AI coding is magnified greatly. Two AI threads interacting with one another leads to race conditions or AI responding to something that does not exist in 'reality'. Welcome to the problem of multicore AI interacting with each other and the virtual world in a constantly changing state. I difficult problem.

That was actually my point, albeit the comparison to LOD was inarticulate.

What I was saying was that if two clients see LODs of different details based on their own display, that's no big deal. However, the AI needs to be consistently applied across all users to prevent what you describe. That's where AI and LOD diverge.