AttackOrders may not prioritize ranged attacks due to rotation conflicts

[IceRaptor]

Another sucide charge from the AI in round one
Hellound charged at the King Crab with spawn protection still on
Attachment file type: archive
RogueLogs.zip
18.91 MB
Harkonnen — 05/28/2021
@FrostRaptor CG decided that charge is the best move here, with ToHit:0.275. maybe there is some logic lapse?
Interactive Rubber Dolphin — 05/28/2021
maybe evasion remove is weighted that much?
FrostRaptor — Today at 9:46 AM
I think it's a combination of things
Looks like all the ranged weapons were marked as wilLFireAtTarget=false in the evaluation
1:30:16.961 Checking range 172.2269 and LOF from attacker: ((384.0, 199.2, -540.4)) to target: ((300.0, 161.3, -685.9))
21:30:16.961 -- weapon => 'M. i. Heavy Laser (C)'
21:30:16.961 -- can be used in melee, adding to melee sets.
21:30:16.961 -- willFireAtTarget: False withinRange: True
21:30:16.961 -- weapon => 'Repeating TSEMP'
21:30:16.961 -- can be used in melee, adding to melee sets.
21:30:16.961 -- willFireAtTarget: False withinRange: True
21:30:16.961 -- weapon => 'SRM6 Streak MKII'
21:30:16.961 -- can be used in melee, adding to melee sets.
21:30:16.961 -- willFireAtTarget: False withinRange: True
21:30:16.961 -- weapon => 'AMS (C)'
21:30:16.961 -- can be used in melee, adding to melee sets.
21:30:16.961 -- willFireAtTarget: False withinRange: False
21:30:16.961 -- weapon => 'Grenade Launcher'
21:30:16.961 -- can be used in melee, adding to melee sets.
21:30:16.962 -- willFireAtTarget: False withinRange: True
Which results in CG thinking the unit is a melee bot
21:30:16.974 == Creating melee weaponSets for attacker: Hellhound HND-RX_Lieutenant_A3C9A6F1 versus target: King Crab KGC-666_Poet_2B0CDE33
21:30:16.974 Unit isMeleeMech due to rawMelee: 153 >= rawRanged: 0 x 2
I evaluates its choices and figures out melee is the only thing that yields damage
21:30:16.980 AEHelper found 2 different attack solutions after evaluating attacks
21:30:16.980 evaluated attack of type Melee with 1 weapons, damage EV of 47.025, heat -81
21:30:16.980 evaluated attack of type Shooting with 0 weapons, damage EV of 0, heat -81
21:30:16.980 best shooting: 0 melee: 47.025 dfa: 0
So it goes for it
Harkonnen — Today at 9:49 AM
so its basically back to the question why " willFireAtTarget: False"
FrostRaptor — Today at 9:50 AM
Yup
I can think of two things off the top of my head
First is that being the first turn, sensors are offline - so pilot might be outside visual range
Harkonnen — Today at 9:51 AM
it was able to charge, caanot be
FrostRaptor — Today at 9:51 AM
Yup
Plus LV logs have this:
21:28:01.349 MH:PCM Parsing current map.
21:28:01.360 Calculating vision range as Math.Ceil(baseVision:40 * 30.0 * visionMulti:0.8) = visRange:961.
21:28:01.360 MapHelper: Vision range for map will be ==> 960m (normalized from 961m)
So shouldn't be a signature/vision issue
Harkonnen — Today at 9:52 AM
lloking at your other explanation.... could it be that yore are checking this from the unit intial position and there is no LoS?
FrostRaptor — Today at 9:52 AM
Wrong element, there's two missions in this log
Yup
That's the second idea. That the current rotation may be what's throwing it off.
Basically if HellHound had KGC in rear arc... most weapons would not figure from current rotation
This is probably a bug in HBS code that I've carried over.
Yeah, that's seeming more and more likely
HBS code:
bool flag = weapon.WillFireAtTargetFromPosition(target, unit.CurrentPosition, unit.CurrentRotation);
bool flag2 = unit.Combat.LOFCache.UnitHasLOFToTarget(unit, target, weapon);
bool flag3 = (target.CurrentPosition - unit.CurrentPosition).magnitude <= weapon.MaxRange;
It's not setup to consider that you can freely rotate from a stationary attack.
In most cases it works because the influenceMap will direct you to a different position before this logic kicks in
KGC sprints in phase 11:
2021-05-27T23:13:07 CombatLog.ActorActivation [LOG] Actor King Crab KGC-666 (ec9d2280-ca9a-4d90-8a20-963d8a4c0a39.1) - Activated by Team Player 1
2021-05-27T23:13:07 CombatLog.ActorActivation [LOG] Actor King Crab KGC-666 (ec9d2280-ca9a-4d90-8a20-963d8a4c0a39.1) - sprinting

(related: HTF is a KGC going on phase 11?!? That's round 19, WTF!)
That's wrong, that is previous mission
Harkonnen — Today at 10:03 AM
actualy, LoS from current explains all weirdness in Ai reported.... stupid melee attacks, mechs running from player....
FrostRaptor — Today at 10:03 AM
@Interactive Rubber Dolphin to your question - the attack is weighted highly because of evasion break:
21:30:16.974 Evaluating damage for state: CHARGE
21:30:16.974 -- Checking ranged weapons
21:30:16.974 - Calculating utility based upon total projected instab of: 200
21:30:16.974 Adding 60 virtual damage to EV from evasivePips: 6 x bonusDamagePerPip: 10
21:30:16.974 Reducing virtual damage by 30 due to potential loss of 6 pips.
21:30:16.974 Reducing virtual damage by 50 due to attacker damage on attack.
21:30:16.974 Virtual damage calculated as -20 = evasionBreakUtility: 60 + knockdownUtility: 0 - selfDamage: 50 - selfEvasionDamage: 30
21:30:16.974 State CHARGE exceeds previous state damages, adding it as highest damage state
Interactive Rubber Dolphin — Today at 10:04 AM
wait

-20
wouldn't that be instead low?
FrostRaptor — Today at 10:04 AM
Yes, that's applied to the EV of the charge attack
The point being without the +60 the charge would probably go negative in this case, and end up being a 0 damage attack - AI would have braced instead
Interactive Rubber Dolphin — Today at 10:05 AM
Ah
I see
FrostRaptor — Today at 10:07 AM
Actually, at a different attack position it's even higher:
21:30:15.193 === Calculating expectedDamage for Hellhound HND-RX_Lieutenant_A3C9A6F1 melee attack from position: (384.0, 199.3, -540.4) against target: King Crab KGC-666_Poet_2B0CDE33 at position: (300.0, 161.3, -685.9)
21:30:15.193 useRevengeBonus: True
...
21:30:15.194 Virtual damage calculated as 180 = totalDamage: 200 + evasionBreakUtility: 60 + knockdownUtility: 0 - selfDamage: 50 - selfEvasionDamage: 30
21:30:15.194 Setting weapon: Melee to virtual damage: 180 for EV calculation
Another factor that could be complicating this
There are multiple enemies:
21:30:16.961 Evaluating all actors for hostility to Hellhound HND-RX_Lieutenant_A3C9A6F1
21:30:16.961 -- actor: King Crab KGC-666_Poet_2B0CDE33 is an enemy.
21:30:16.961 -- actor: Corsair COR-RA_Raven_F9FAE558 is an enemy.
21:30:16.961 -- actor: Mauler MAL-RX_Ronin_9D156E66 is an enemy.
21:30:16.961 -- actor: Munin LAM MN-1B_Tinker_E325BAAE is an enemy.
The AI logic looks for 'opportunity fire' shots against targets, and so long as it finds one that beats the threshold it takes it.
If the influenceMap logic results in a node position & rotation that is facing say the Munin, but the KGC evaluates first in the list and beats the threshold - you'd see behavior like this.
It picks the KGC as the first opportunity fire candidate instead of evaluating all of them, as a performance boost.
This one will be a bitch to fix

[IceRaptor]

Issue fundamentally is that ranged attacks are evaluated from current position + rotation, because AI logic assumes movement then eval of attacks. In the case that we have a stationary attack node, ranged attacks should allow rotations of the unit to turn to face their target. I think this already happens but need to include more logic to try to make this more intelligent.