But one would have to put a delay in the blindness, or we'd wind up with some really stupid AI.

Here's the scenario:

One enters a cloud at 90 degrees to the AI plane at, oh, say, 200 meters distance. Let's say it's a smallish cloud, too.

Now the human would tend to go high and look to the opposite side of the cloud for the enemy to emerge, anticipating the path of the opponent. We'd circle and be very keen to find out where an enemy went after entering a cloud.

AI, however, stink at short term memory - indeed, they have none at all.

So what we might see is "out of sight, out of mind" and the AI who was engaging at a very close range suddenly slowing down and returning to his next waypoint as if he had never encountered an enemy at all.

The problem with AI in clouds, IMHO, is that they never lose the exactness of one's position. What would be really sexy would be the AI losing that refined position. Instead of being plane sized object in a detectable sphere (the "clear skies" condition), what if that sphere and the size of one's plane shrank, with the position in the sphere offset (ideally, randomly).

While the above scenario is not perfect I would personally prefer it if the AI simply forgot about you when you flew into a cloud rather than pursue you as if the cloud didn't exist. Here is the current scenario in Rise of Flight, I can't understand why people think this is preferable to what you described above.

One enters a cloud at 90 degrees to the AI plane at, oh, say, 200 meters distance. Let's say it's a smallish cloud, too.

Now the human would tend to go high and look to the opposite side of the cloud for the enemy to emerge, anticipating the path of the opponent. We'd circle and be very keen to find out where an enemy went after entering a cloud.

AI, however, have perfect situational awareness. They never ever forget where you are or lose your position, no matter what the visibility.

So what we see is "clouds don't exist for me stupid human" and the AI continues to follow you into a thick cloud and shoots you down as if the cloud didn't exist at all.


I totally agree with this, it is something that would make engaging the AI more 'human like'

I don't think anyone "prefers" the current situation; it simply is what it is.

Let's not attempt to polarize this into a wedge issue of those "for" the AI seeing through clouds and those "against."

It's more like "huh, well I wish they'd fix it, but whatever," rather than "hell yeah, that's what we want, can we get more of this!"

Likewise, wanting it to change doesn't make one a simulation hating zealot.

Because, after all, if the sim wasn't damned good at its core nobody would have any sort of opinion at all.....since none of us would be playing it.

It's more like "huh, well I wish they'd fix it, but whatever," rather than "hell yeah, that's what we want, can we get more of this!"

Likewise, wanting it to change doesn't make one a simulation hating zealot.

We're only chewing the fat. The SimHQ forum rules are in operation, anything not forbidden is compulsory. Ok yes that does apply to physics but it's a start

a) Nothing is between ---> target is visible
b) Something opaque is between ---> target is invisible
c) One or more transparent textures are between.

Find the texture for each one, look up the transparency and "add the transparencies up". Have some threshold, say 50% and if it is more transparent than that, the pilot sees the object, if it is more opaque than that, he does not see the object.

Very interesting Buddye1, let's hope the clouds are static or face CPU meltdown

Confirmed on IL-2 objects not being able to see through solid objects like buildings

Ta Dart, Buddye introduces a wrinkle: levels of transparency. That's very ambitious

The workings of RoF with a combination of volumetric clouds with sprites around hints that they were coding in the direction of AI blindness. I wonder how the AI should be programmed when blind.

Yes how would an AI pilot deal with finding himself inside a cloud. At 300mph. We're going to need bigger CPUs

But it does make me wonder, if LOS occlusion (you're an Il-2 AI lorry driver and you cannot see the Panzer up ahead waiting to spring the trap, hiding behind a building) is already in Il-2 then: it may be because objects on the ground have only slow speeds (so the CPU can keep up while checking for LOS occlusions), Oleg has incorporated effective slow-speed LOS occlusion citing Dart's report. So it's not beyond the realms of fantasy that for SoW and our faster CPUs, Oleg's SoW pilots will lose sight of us behind clouds. The not-too-transparent is-it-or-isn't-it ones

Would be great to sneak in and out of the gaps between clouds waiting to pounce is what I mean, and in RoF we already have an effective overcast with variable thickness <dreaming one day>

Ming

As usual, Winston Churchill provides the solution in a succinct way:

"When going through hell, keep going."

The AI simply holds "control position," maintaining the same track as when he entered the cloud.

That was rhetorical

I meant he'd be within many layers of textures so lots of CPU needed

Of course he'd keep going, they're more intelligent than us when we get into a mass panic and start throwing the stick around when the pencil rolls off the dash

Ming

I usually blip the engine off, throw my hands in the air, and squeal like a little girl.

Oddly enough, I have yet to wind up inverted or otherwise compromised.

AI sucks, it always will because besides direct instructions, you have to include tricks and cheats to make it appear like it's smart. It's not, it won't be for a very long time because it can't learn in it's current form with today's processing power. You have a human giving it 'and', 'or', 'else' and 'if' statements, so that means that every maneuver a human could perform - the programmer has to be aware of it and how it will interact with the hard set boundaries surrounding it (ground, buildings, other aircraft) and what it should do in different atmospheric settings (high alt/thin air, low alt/thick air, windy conditions, etc).

AI is far easier to code for shooters because the environment is far more simple, but there are so many variables for air combat simulations that it would take a long time to code the AI plus it needs so much more computing power than we have today.

AI is a practice tool in today's networked world, it doesn't come close to how a human would act in the same circumstances and it will be a very long time before that happens. It'd be nice to redirect resources, of any company that produces an air combat game, towards an online war like WWIIOnline... but better and eliminating the ground forces being human controlled.

The problem with realistic and Human like AI is that it is so very costly.

The developing of realistic Flight Sim AI can “not” really be coded to a technical spec (like graphics, landscape, textures, hardware interfaces, multi player, canned missions, etc.) where the cost can be accurately estimated and bounded. Developing AI is a process (develop, test, refine, test, refine, etc). You are trying to simulate a human whose options and thinking are almost endless.

AI Robots can be built today if their functions are limited (repeatable task like the auto robots) but the building of human robots would require the refinement and testing process with endless delays and the cost gets too high or out of control. It is almost impossible to "bound" human thinking (say a good pilot, as individuals solve problems and invent solutions in an almost infinite manner).

Combat Flight Sim AI has the same problem. The development cost to make the AI more human and thinking is just too high. The game developers are force to get the AI working and call a stop to the refinement process. They can not continue the process too long because of cost.

In BOBII, we work for free. We get endless ideas from our players and testers which we can use for the BOBII AI refinement process.. The implementation and testing is still very difficult requiring much refinement and some really good testers who can offer refinements. We have been refining the AI in BOBII using this refinement process for 5 years. We could go on forever with the only limit being “new” ideas and experienced manpower.

The answer IMHO is simple, the Flight Sim game industry, the developer, can not afford the money for an effective and complete human like AI system (a commercial game developer really can not afford a BOBII like refinement process). Only the military or NASA can afford to develop smart, effective, thinking, and human like AI and even they have had limited success, as yet..

Could an AI modelling engine become a viable licensed property, along the lines of speedtree or even a graphics/physics engine?

If I were developing a simulation it'd probably be cheaper to buy and modify an AI engine than to build one from scratch.

I need to become mega-wealthy so I can play venture capitalist for such ideas.

" ... I need to become mega-wealthy so I can play venture capitalist for such ideas. ..."

You just gave some kind of goal and meaning to my life.

Greetings,
Catfish

Yes, IMHO, it is possible if an interface spec could be written between the new game engine and the new AI engine. A big if , I bet, as the AI has many tasks for the game (takeoff, landing, danage, FM, engage, disengage, formations,etc).

in some ways it's surprising there isn't really a front running AI engine at this point, especially considering what incorporation of physics engines like Havok and ODE have done to the way players enjoy their games. Suddenly developers could throw realistic collisions and other interactions *almost* as an afterthought and consequently such things started showing up in games where it was a genuine surprise to see them.

I think in many other ways, the nature of AI makes it much harder for this type of thing to come about. While your WWI biplane, fancy sports cars, and miscellaneous thrown objects all behave according to the same laws of physics it's harder to see how an AI engine designed to fly planes, drive cars, and throw grenades at players would have any sort of redundancy at all (meaning all these situations would have to be looked at, in detail, almost as if the developer was starting from scratch each time). I think there's money in it so I think dev's must have tried it by now and failed. Still, there's always room for clever innovation!

Havok and ODE

And PhysX running under a GPU

I've seen reports of an AI API, for game developers to licence (like SpeedTree for example). Not sure that that will ever take off

Ming

To put this into perspective, the reason why I think AI "cheating" (quotation marks used since a totally cheat-free AI is probably not possible at the moment) matters is that it actually affects gameplay a lot. If you read WWI or WWII memoirs, after action reports, or tactical instructions the role of the enviroment plays a huge part. The use of sun and clouds to ambush or escape an opponent is particularly important.

In sims, however, these things take a back seat to "gadget" factors like how you use flaps or radiators. I do not find this particularly realistic. I think it's partly a reflection of the mentality of a lot of simmers and aviation buffs, who often have a "gadget" mindset, and partly a reflection of how the sims work. If the AI doesn's recognize sun or clouds it does not make sense for the player to incorporate them into his tactical thinking. It is analogous to the discussion about micromanagent vs tactics in the wargaming community.

The other big missing piece right now is formation tactics, which is also partly an AI issue. The one-on-ones so beloved by simmers are inappropriate even for 1917, let alone 1941.

Sad, but true. Many do think think that flight is all about buttons. Also flight instructors say that people coming to them from sims always stare at instruments instead of flying by VFR like they should. Comes from the fact that sims rarely provide full visual clues about speed, climb rate, altitude etc. RoF's pretty good in that respect, BTW.

I guess I buck the trend, but I hardly ever look at the instrument panel in flight sims!



Actually, in WWI it pretty much always devolved to one-on-one, which was as much a factor of training as it was machine. Remember Dicta admonishment that as many planes need to be engaged as possible, and that no two planes should go after the same target.

Think forty planes fighting in a combat box that could be as little as 800 meters to the side and you'll understand why the wingman concept didn't really take off in WWI.

Lol, I usually forget to check the instrument panel. I'm in the process of uploading a vid (nasty compression on it tho) of a 1v1 RR in a Dolphin - for pretty much the entire fight you can't even see the airspeed indicator lol.

Of course, this goes some way to explaining why I stall out and have insufficient height to recover the spin fully, lol.

As for combat tactics - I think you've both got it right, to be honest - in that solo flights had more or less ended by the end of 1917, and you would have squadron-sized formations fighting together - even if it was 1 v 1 on an individual level, you still had those 40-odd aircraft in the combat zone whereas in earlier years you'd find much smaller groups up together. Or at least, that's my understanding.