[JVM]

Hmmm, for what I understand, NeoqB people are not done yet: the AI is still able to see across clouds....

This is a very difficult problem to solve, because the software needs to be able to compute if an A/C coming into the AI detection bubble (or something to that effect: it is not possible to calculate a detection probability for every A/C in the environment; I suppose you have to set a limit arbitrarily at a certain distance, say 10 km) can be seen or not...
For this to work with a cloud, the cloud should logically exists as a bona-fide 3D object (and possibly an associated "detection box") , not a sprite or pseudo 3D object.
To simulate a human pilot reaction, one may possibly compute a detection "event" taking in account, alongside the straight line between the pilot eye and the A/C to detect:

- Pilot quality (rookie - ace) -> a first coefficient between 0 and 1
- Pilot fatique -> again a coefficient between 0 and 1
- cockpit geometry (taking in account structural members) -> 0 or 1
- A/C geometry (parts like wing or cowling, dead angles) -> 0 or 1
- atmospheric transparency (there is a difference between CAVOK in Arctica and summer haze in Tuscany) -> another coefficient between 0 and 1, also depending of distance between the two A/C and atmosphere model (inversion layer level is crucial here); if you or your opponent are in the cloud, it becomes 0.
- difference in height (depending of respective atmosphere layers, and also camouflage quality) -> another coefficient between 0 and 1 (maybe a discret set of values here)
- cloud part presence along the line of sight
- respective flying direction (a fleeing fighter as seen from the observation point is more difficult to spot than a parallelly flying bomber)

All this will give you after computation a value either of 0 or someting between 0 and 1 at a time t0 (entrance of the A/C in the detection bubble); logically this computation will be done X times/s should be combined to the time interval between computations (as long as the A/C is in the detection bubble):

The addition of the "time interval X the computed detection coefficient" over the time since the A/C has penetrated the detection bubble or since the coefficient have changed from a 0 value for a "certain" time ("I believe I have seen someting, but I cannot see it anymore...") represents the probability of detection; when this number goes over a certain value, and never goes back to 0 for longer than two or three intervals "detection is acquired"...

I believe this would emulate quite well the way a real pilot would act...and I suspect it may be difficult to fully implement in a sim due to calculation power constraints (but with multicore CPU, who knows?)...So this can be simplified by just integrating the pilot/cockpit characteristics as a detection probability factor, the existence or not of a part of cloud in the line of sight direction, the dead angles and the real distance between the A/C (not 300 m like in IL2, even in the dead angles)

I just hope I did not lose anybody here...

Amically,

JVM

PS: the RoF clouds as seen from above or from aside are beautiful, very much real-life, but the cloud bases are still not very realistic (they should mostly be "fuzzily flat")...