AI Discussion (dev branch)
Hi, there
When it's possible/useful, we will try to post brief descriptions of recent AI changes done on the development branch. We hope this will help demonstrate the tweaks that we're making and generate useful feedback to help us validate and polish our work.
For now, the first batch of notes to kick things off:
It wasn't supposed to fix rotation speed of AI soldiers, but is letting them use "aiming deadzone" as specified in the unit config (minGunTurnAI, maxGunTurnAI) which is typically 60 degrees cone.
Therefore currently AI should be able to follow targets better without moving whole body.
RPG and NLAW are supposed to be unguided, but AI was still able to use them as guided weapons. This was corrected.
Quite big change, now AI is able to calculate and predict accurate ballistic of bullet, therefore is able to aim and engage targets on all ranges, both static and moving. Also, this improves AI ability to hit moving targets by more precise time of leading calculation. Good way to check this change is try to fight AI sniper on 1km+ range. Of course it still depends on his skill.
AI was considering animals as friendly, therefore denied to shot in theirs direction. Example: put yours character, enemy soldier and snake animal between you. After this fix AI won't bother with friendly fire against animals.
That would mean something like setSkillFinal. I'm afraid this is something that would require many exceptions and would be tricky to handle from another user's PoV (why some AI are affected by difficulty and some aren't...)
AimingAccuracy skill or skillFinal? First one should always be the same (it's what you set via slider or script). The second one is after AI Level interpolation - the one actually used in simulation.
Revive / incapacitation is currently a scripted/modular system with a specific MP use. Full sandbox support is definitely desired, but atm we've got no ETA. Not anytime soon.
The best example is probably the aiDispersionCoef from fire mode config of a weapon. Let's have a weapon with a firemode with aiDispersionCoef 10 (ignoring the Y/X atm). But the actual coefficient that's used to increase bullet spread depends on the skill. The following table (still using that aiDispersionCoef 10) could better show the difference between then old and new state. AI's aimingAccuracy 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 resulting coef. (old) 23.5 11 6.8 4.75 3.5 2.67 2.07 1.63 1.28 1 resulting coef. (new) 9.1 8.2 7.3 6.4 5.5 4.6 3.7 2.8 1.9 1
You can see the "most interesting" differences were happening ~ between 0.2-0.4 AI aimingAccuracy. Now we are talking about "final" skill - after taking the difficulty/AILevel into account.
To put it in a better perspective - on "Recruit", with AI Level Precision @ 0.2 - an AI with raw aimingAccuracy 0.38 actually translates to 0.194 final aimingAccuracy. And with raw aimingAccuracy 0.2 (still possible via editor) = 0.01 final aimingAccuracy. You can already imagine the resulting dispersion at that value using the old interpolation ;) Broken AI. AI tends to throw a grenade at a target that the AI doesn't directly see, but it's able to estimate its position and knows it's an enemy.
As for UGL - the AI should be more prone to fire them on condensed groups of soldiers or vehicles - if the probable outcome has a higher value than firing a bullet. But there might be some tweaking necessary.
Revive / incapacitation is currently a scripted/modular system with a specific MP use. Full sandbox support is definitely desired, but atm we've got no ETA. Not anytime soon.
The best example is probably the aiDispersionCoef from fire mode config of a weapon. Let's have a weapon with a firemode with aiDispersionCoef 10 (ignoring the Y/X atm). But the actual coefficient that's used to increase bullet spread depends on the skill. The following table (still using that aiDispersionCoef 10) could better show the difference between then old and new state. AI's aimingAccuracy 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 resulting coef. (old) 23.5 11 6.8 4.75 3.5 2.67 2.07 1.63 1.28 1 resulting coef. (new) 9.1 8.2 7.3 6.4 5.5 4.6 3.7 2.8 1.9 1
You can see the "most interesting" differences were happening ~ between 0.2-0.4 AI aimingAccuracy. Now we are talking about "final" skill - after taking the difficulty/AILevel into account.
To put it in a better perspective - on "Recruit", with AI Level Precision @ 0.2 - an AI with raw aimingAccuracy 0.38 actually translates to 0.194 final aimingAccuracy. And with raw aimingAccuracy 0.2 (still possible via editor) = 0.01 final aimingAccuracy. You can already imagine the resulting dispersion at that value using the old interpolation ;) Broken AI. AI tends to throw a grenade at a target that the AI doesn't directly see, but it's able to estimate its position and knows it's an enemy.
As for UGL - the AI should be more prone to fire them on condensed groups of soldiers or vehicles - if the probable outcome has a higher value than firing a bullet. But there might be some tweaking necessary.
