Proposal for a new Visual Model

[MBot]

One of the problems the Lock On AI has is in my opinion their oversized situation awarness. One possibility to reduce their SA to a more realistic level might be the implementation of a more sophisticated simulation of sight with all the limitations the player and real pilots have to face. Because right now the AI seems to be seeing things instantly and everywhere. I will now write down some of my ideas to make a better visual model. Please note that I don't know if my ideas are practical from a developers point of view. Those are just some ideas that float around in my head, please feel free to rip them apart with toughtfull posts :)

 

 

The initial idea came when I tought more about the new missle seeker modeling of the 1.11 patch. You remember that now the radar seekers of missles have a scan cone with a fixed FOV and a "search FOV" in wich the radar cone follows a search pattern. How would it be if the modeling of sight ( for the AI ) would follow simmilar routes as the modeling of radar ? Lets see how that could look like.

 

First we have the fixed FOV of our AI pilot ( or Strela-10 commander or whatever ). This is the area in wich pilot's eyes are actualy seeing. Lets say this FOV is 60° wide. I am aware that the true FOV of a human is something like 180°, but the FOV where you see things sharp is a lot smaller. Lets just say 60° for the moment. Lets call this the 'search cone'.

Now we have the FOV that the pilot can see by turning his head. The area is limited by the pilots ability to turn his head and the cockpit design of his aircraft. Let's call this FOV the 'scan area'. Every plane has it's individual 'scan area', definded by canopy rails etc. Basicaly this is how missle seeker heads are modeled right now.

In order to check the airspace the pilot moves his search cone trough the scan area following a special pattern. That pattern would most likely depend on the situation the pilot is at the moment. In normal flight, the prioritys in the scan area would be: 1st straight ahead, 2nd at your wingman, 3rd other parts of the scan area.

The pilot would plot the visual contacts around him with a simmilar logic that is used right now in TWS radar modeling. By looking at a contact it's vector and speed is noted ( with a certain % of error ) and plotted in the pilots mind. Now the pilot knows the targets presence even when not looking at it, but not very precisely. The pilot has to check the target within certain intervals to update the plot, because sligh errors in the initial vector will cause bigger errors after time or the target could maneuver. The need to check the target often to update the plot would increase with decreasing range and in a dogfight a target would occupie almost all of the pilot's attention. Otherwise the pilot could loose the target. The pilot can only keep track of a limited amount of airplanes and the closer he is to a single target the less time he to track other targets or search for new ones. Hey, we just modeled target fixation for AI ! I hope my idea of visual 'Track while Scan' is understandable, otherwise I will explain further.

 

Now another aspect comes into play. We usualy don't see everything we have in our FOV. There are certain factors that determine what we can see. While trying to model that, let's keep close to our radar analogy.

 

1st factor: Target range. Targets farther away are harder to spot. Very much like with radar.

 

2nd factor: Target size. Bigger targets are easyer to see. Very much like radar, where a big RCS means earlier dedection.

 

3rd factor: Target aspect. Planes from front/aft are harder to see than from top/bottom/left/right. Just like with radar, where RCS changes with aspect angle.

 

4rd factor: Target height. Airplanes are harder to see against the ground than against the sky. This mechanic is already modeled for radar systems.

 

5th factor: Lighting condition ( time of day ). When it is dark you see less. Perhaps we could link the eyesight to time of day ( less eyesight ~ weaker radar system ).

 

6th factor. Eyesight of the pilot. Could be coupeled with pilot skill ( along with ability to do "visual TWS"/keep SA ).

 

7th factor. Focus. The eye needs several moments to actualy see something, you wont scan the sky by just turn your head from left to right. There should be some sort of minimum time a contact has to be within the FOV until it is dedected. Perhaps this should also depend on visibility of the target ( ~ radar return streight ), you notice something clearly visible faster then something barley visible.

 

The decision wether our AI pilot sees a target or not would be a combination of all the above factors. You see that a good visual model is very simmilar to the radar model, perhaps there could be even shared code by the two models.

One thing that would be quite hard to model tough would be fog or clouds blocking visibility. I am not quite sure how to simulate this. Perhaps by comparing contrast between the the target and the pixels surounding it.

 

Now I already hear the question regarding FPS costs of such a complex view model. It is a good one and I am certain that such a model would cost FPS. BUT please remember that also the radar model uses up a considerable amount of performance and I never heard someone complain about it. Why ? Because radar is essential for a modern aircombat sim. But as is the visual model. Even today the eye is an important sensor in AA combat, I would say on the same level as radar. So I would consider the performance costs as justified.

 

 

Now our juicy new visual model has given us many cool things, like degraded SA for the AI, target fixation for AI, AI loosing visual contacts and the possibility to disengage combat thereafter.

And it opens the door for one of the basic operational doctrines for NATO airpower. Mutual support. For those that don't know. This means that planes are operating in pairs with each pilot keeping an eye on his partner ( and his blind spots ) and warning him about threats.

Now we can actualy implement this, because our pilots are no longer able to instantly recognise any threat to their aircraft. They need the help of their partner the see their own blind spots. This would work the followong way. If a pilot sees a direct threat to a member of his flight ( threat and teammate in the FOV at the same time ) he will give a warning to him ( with a small delay to simulat voice comms ). For the programmer, this would be something like a "visual datalink" between members of the same flight. If the plane in the FOV is the player, the warning will be done by voice ( "Break left, SAM launch 3 o'clock" ). But as the AI wingman doesn't spend all of his time checking you, there still is a chance that he will miss your threat.

Now the mutual support concept is about partnership, so the player should also be able to warn his wingman. This can be done quite easily. If the wingman and the threat are withing the players FOV, let's just assume that the wingman will be warned about the threat. It isn't practical for the player to use the comm menu and send a warning manualy, that won't be fast enough. I think we can assume that the player would warn his wingman instantly if he could. It doesn't matter that much, as the basic concept of mutual support is there, that is keeping an eye on your partner. The player will spend quite some time looking at his wingman, because this will increase his surviveability ( just like in real life ). With the wingman beeing safer, this also means that he is there to warn the player about threats. So both pilots gain from the partnership. We just included mutual support the first time in sim history :)

 

 

Ok, this was quite much for the moment :) If your are still with me, please feel free to comment on the subject.

[Mizzy]

:)

 

This game has good graphics.

 

Regards

 

Mizzy.

[GGTharos]

Yep, I agree that such a sensor model (not just visual) would be very welcome, and it has already been discussed to some degree.

 

Short answer: Probably not going to happen any time soon.

 

TO me it looks like the underlying code simply doesn't support it very well. Once there can be more agents (ie. AIs) on the field and there is a good communication model for the AI, then things can change.

[Shaman]

:)

 

This game has good graphics.

 

Regards

 

Mizzy.

Hey Mizzy, have you actually read Mbot's post?

It's not 'bout graphics, but all-seeing AI behaviour, and how it could be changed.

I say you better take a look. There's pretty good feedback there.

[Mizzy]

Hey Mizzy, have you actually read Mbot's post?

It's not 'bout graphics, but all-seeing AI behaviour, and how it could be changed.

I say you better take a look. There's pretty good feedback there.

 

Well I wasn't really being serious simply because I don't understand anything much to do with the technical side of flying simulations. Sorry if my post appeared disrespectful, it just look like the poster put so much effort into his/ her post and nobody replied as yet :)

 

I meant no harm.

 

nite nite

 

Mizzy.

[Shaman]

I was just thinking about Mbot's feedback. This just asks for improvement in the area of AI (vocal) radio communications. We could hear how they are exchanging informations around themselves.

About possible FPS drop: I don't think it would eat so much computin power. Well look at Operation Flashpoint, or FarCry. As far as I know, each soldier has it's own complex "mind" and makes independent decissions how to react at the moment - and there are many of them and it works quite well.

 

I meant no harm.

nite nite

I know, and there are no hard feelings on my side. Goodnight! :)

[Maverick-90]

Because right now the AI seems to be seeing things instantly and everywhere.

It doesn't, I've been flyin' on some Fulcrum's six for over 5 mins and he didn't even recognize me (until he felt my 30mm Bullets in his a$$ ) :icon_weed

[Guest IguanaKing]

True...airborne AI isn't omniscient in a tactical sense, but it is in a strategic sense. Now, the IADS assets knowing exactly where you are, on the other hand, that's a little annoying.

[Weta43]

Maybe a simpler start would be that if actively engaged with a target (A2A BVR, WVR or A2G) AI won't visualy recognise another threat/target unless it is within some range considerably smaller than their normal SA or within 20 degrees of their original target

[MBot]

It doesn't, I've been flyin' on some Fulcrum's six for over 5 mins and he didn't even recognize me (until he felt my 30mm Bullets in his a$$ ) :icon_weed

 

True, I have expierienced this every once and then aswell. So there already is some simple form of blind spots, wich is good.

Still I have expierenced weird AI behavior many times, like spotting me while beeing directly below them. And it seems impossible to disengage once combat has started, because AI will never loose sight once he has spotted you. This usualy means that every combat in Lock On is fought until one side has been completly extinguished, wich is quite unrealistic.

 

 

Maybe a simpler start would be that if actively engaged with a target (A2A BVR, WVR or A2G) AI won't visualy recognise another threat/target unless it is within some range considerably smaller than their normal SA or within 20 degrees of their original target

 

I like that idea. It would introduce some simple form of target fixation.

[MBot]

I will give it another go before let it slip into the unknown depths of the forum history :)

[Yellonet]

Yep, something like this would be very nice.. but would unfortunately add to LockOn's heavy CPU dependance.

[Weta43]

So if they started work on it now, by the time they put it out processors will be up for it :-)