Stupid AI Tricks

[dnk 13]

I'm going to post my issues with the AI here. The first part is one the stock "regular" difficulty AI in urban combat. Later parts will deal with altered AI values and situations. I think it's important to start with regular AI, as that's what most missions end up as and what most unaware players/mission makers will have to deal with ingame and in the alpha/beta/demos.

Primary Problems

1. The Dance of Dumbness: if you ever really pay attention to the AI, you'll see this. Basically, the guy stands up and starts walking around slowly in a tight circle, then usually walking away slowly in the direct opposite direction of the guys shooting at him (or not, they do it pretty much in any situation). An obvious broken mechanic, and a common breakage at that.
   
2. The Dunce Decider: again, highly common. This is the guy that decides to aim at, watch, and try to engage a 200m+ distant target that's not firing at him currently, while totally ignoring the 3 enemies standing within 10 feet of him lighting him up. "Priorities", I guess...
   
3. The Wall Watcher: pretty much that. Sit/lie in a building (or outside), watch a blank wall. I guess there's something interesting behind it... but what's wrong with using a window? Or a door? Or maybe stop being a pansy and get out there and start fighting? Nah. I guess this can be seen as BIS modeling in "morale" or something - these guys are clearly traumatized or something. But then the AI does this on a street, too. No, don't like watch down the street or scan the area or something, keep staring intently into that fence post.
   
4. Standing Stupid: if you're being shot at in the open, you should just stand up straight and do nothing for 30 seconds. Deer in the headlights much? Never, ever, ever lie down or even think about crouching to reduce your silhouette. And never use anything other than the 3 basic positions (ignore all the extra ones players can use).
   
5. The Dance of Dumbness pt 2: come to open ground with enemies all around. Run through open ground under fire. Stop. Turn around. Run back. Stop. Turn around. Eventually die.
   
6. The Bumbling Building User: wait, there's a second floor? Wait, there are windows and doorways? Wait, we're NOT supposed to just camp inside one of the rooms and aim at a wall? I thought these things were ground-floor-only covered walkways!
   
7. The Lone Loser: lots of ramboesque "I'm going it alone" tactics in urban scenarios. Too often I see guys wandering (they're certainly not in a hurry) through the streets, detaching themselves from their squads and supporting bases of fire to penetrate into enemy lines. While this makes for slightly more interesting "surprise 1-man invasion" scenarios, it definitively is not realism-enhancing.
   

Discussion/Solutions?:

1. I don't really know what's up with this, but I figure the devs can find it and fix it, maybe just by shortcutting whatever's doing it with a "lie prone and wait" command instead.
   
2. Really, really need to work on AI priorities, but this may also be due to broken awareness. Obviously, the closest target is not always the most important, especially if a farther one is actively engaging you, but when anything's within like 20m, it should definitely take priority over the guy 150+m away. I would suggest something like this for prioritization: [(500 x (weapon modifier))/(distance to enemy)] + [(490 if currently/recently shot at by)/(distance to enemy)]. Here is a graph showing the result (numbers can be tweaked, of course, this is a rough version - the basic thing is that it should be exponentially in favor of closer targets, with the "being shot at by" taking over in importance at medium distance):
   
   
3. Basically, I think there are two approaches for when a target is totally out of LOS. First, if engagement is not a priority (other orders are higher priority or it would require too much repositioning to accomplish) then the target should be stored but cease to be acted upon (in place, active scanning of the "front" or whatever should take over, supporting rest of squad, etc). Second, if engagement is a high priority, then repositioning should occur.
   
4. AI should react realistically to incoming suppressive fire and run for cover or GET DOWN. This should become a #1 priority in all situations.
   
5. The AI should never attempt to close range with any enemy that's in close quarters (20m or less, I'd say). The AI should also generally highly prefer cover/prone to running through open ground when multiple enemy contacts are around or when under fire.
   
6. This I leave to the devs to figure out.
   
7. Basically, improve CQC squad tactics. They do seem to like to bunch up a lot when in cover, but then they break up into 1-man tactical units when they press forward.
   
8. (ADDITIONAL TO ALL): broken awareness of open terrain. A way to handle this is to AI-map the levels, basing areas' "cover rating" on the amount of cover objects (probably needs a creative hand to accomplish) with at least a 1m2 level of resolution. It could be divided into (1) high cover [buildings, closed in yards, etc] (2) medium cover [forests, rocky areas, small streets with good cover on both sides] (3) low cover [low density forests, cargo containers, open yards, only low cover stuff like a low fence, a street with good cover on one side] (4) no cover [large streets, streets with no cover on either side, that dried up drainage thing in Agia Marina, open ground]. The AI should then be programmed to NOT go over open/low cover terrain when already in cover unless (1) very well supported by a base of fire element that can actively suppress (and DOES suppress) while they move forward, simulating proper "leapfrog" tactics, and when they do so it should be at a FAST PACE and with a preset path and destination; (2) there is no high-cover terrain between them and the enemy and they are being aggressive (due to higher numbers or a flanking operation); (3) they have very low skill; (4) they are taking up a prone covering position behind some obstacle. If the AI can cross <75m of open terrain to find more "good cover" terrain on the other side, then it should be a fast sprint to the other side's cover, and no dilly-dallying in that no-man's-land, and definitely no RUNNING BACK AND FORTH through it.
   

1. (ADDITIONAL TO ALL): broken awareness seems to be an issue. The AI should be actively scanning their environment more and have a much easier time of finding nearby enemies.
   

---------- Post added at 01:11 PM ---------- Previous post was at 12:04 PM ----------

I've played around a bit more. It seems the situational awareness is due to the AI just not looking around much and having minimal peripheral awareness. You can place an enemy basically 5ft away at 90-degrees to the AI's orientation, and he'll never see the guy. The vision cone in this game's AI is ridiculously restricted, and I think at least a couple issues might just boil down to that.I'm going to post my issues with the AI here.

This visual problem extends into acoustic space, where nearby gunfire is impossible for AI to pinpoint or acknowledge for fairly long lengths of time, even when it's directed at them. This needs to be fixed with an almost instant reaction time to incoming, close-range gunfire, where the reaction time to income gunfire is based on a formula similar to the one above for targeting priorities, additionally modified for gun loudness.

Edited March 19, 2013 by DNK

[ShadowPryde 10]

What i've also seen in the first mission is that while clearing out the area, my squadmates got down on the floor but didn't change there stance after the fight was over. They where just sticking to the floor in the open or in the middle of the street.

They didn't even bother to get up to get into cover, scout the area or whatever.

That's something i've allready seen in Arma 2.

My guess is that although the AI has some improvments over Arma 2 Vanilla it still shares the old sickness in many parts.

I really hope that the AI is one of the top priority stuff this time around and get's fixed.

[maturin 12]

Wow, an AI complaining thread that might actually be useful. Nice labels.

[-Coulum- 35]

Some valid issues and some nice names to go with them! Can't really reply now but I agree with most of what you are saying.

You can place an enemy basically 5ft away at 90-degrees to the AI's orientation, and he'll never see the guy. The vision cone in this game's AI is ridiculously restricted, and I think at least a couple issues might just boil down to that.

I don't think it is so much a problem of the ai having restricted FOV, but of them simply not glancing left and right every so often, or occasionally checking their six. they just stare straight ahead. I thought this was improved upon in a OA beta but it doesn't seem to have carried over to arma 3.

[fabrizio_t 58]

Nice thread. ArmA3 AI is basically ArmA2 AI with some config buffs in my opinion.

Most of the issues are inherited by the previous BIS game.

[*]I don't really know what's up with this, but I figure the devs can find it and fix it, maybe just by shortcutting whatever's doing it with a "lie prone and wait" command instead.

As you stated, it's clearly broken. Waiting is not a viable solution in my opinion, since a man stopping does stop its entire group, when in combat mode.

I think a unit having identified a threat should either open fire -or- move into nearest cover - at full speed.

[*]Really, really need to work on AI priorities, but this may also be due to broken awareness. Obviously, the closest target is not always the most important, especially if a farther one is actively engaging you, but when anything's within like 20m, it should definitely take priority over the guy 150+m away. I would suggest something like this for prioritization: [(500 x (weapon modifier))/(distance to enemy)] + [(490 if currently/recently shot at by)/(distance to enemy)]. Here is a graph showing the result (numbers can be tweaked, of course, this is a rough version - the basic thing is that it should be exponentially in favor of closer targets, with the "being shot at by" taking over in importance at medium distance):

Looks like AI prioritize more newly discovered threats than previously detected threats coming again into line of sight.

At least debugging danger.fsm shows that.

[*]AI should react realistically to incoming suppressive fire and run for cover or GET DOWN. This should become a #1 priority in all situations.

True. We are somewhat able to script the units to lie down, but we are not able to consistently make them run into cover. We lack reliable commands (findCover? ForceRun?).

[*]The AI should never attempt to close range with any enemy that's in close quarters (20m or less, I'd say). The AI should also generally highly prefer cover/prone to running through open ground when multiple enemy contacts are around or when under fire.

It should not lie down in CQB, while crouched stance may be the preferred one imho.

[*]Basically, improve CQC squad tactics. They do seem to like to bunch up a lot when in cover, but then they break up into 1-man tactical units when they press forward.

This problem happens when the AI leader orders single units to engage targets, which is pretty common.

Infact unit enableattack false does mitigate the issue.

Some more points:

1) You mentioned the fact that AI units keep RUNNING BACK AND FORTH on open terrain.

That's true and it also happens in urbna environment, it's just less noticeable.

Whenever a single AI units moves, all the following AI units in the group feel they should adjust their own position, even if they're actually in good cover.

This leads to a lot of "running back and forth", cover swapping and other ugly and useless movements.

The root issue is the AI group is always trying to keep formation too tightly, while it should favour keeping actual cover/advantage positions instead.

2) Another big problem is a stuck AI unit will stop all subsequent AI units in the group.

These will wait indefinetely for it to move on.

3) Another ridiculous problem exists with fleeing / withdrawal: an AI group is just plain unable to withdraw as a whole from the battlefield.

Sometimes a single AI unit may start fleeing: when this happens all the group instantly (and magically) does the same.

The interesting thing is that these fleeing groups usually do move slowly away trying to keep into formation, but without caring to keep heads down.

4) AI machinegunners are still insanely inaccurate.

Edited March 19, 2013 by fabrizio_T

[NordKindchen 12]

AI needs to be high priority. Its the main reason Arma got critizised in the past:

[Samfisherchaos 10]

AI needs to be high priority. Its the main reason Arma got critizised in the past:

This ^ and more importantly more of this ^.

[krycek 349]

Some more points:

2) Another big problem is a stuck AI unit will stop all subsequent AI units in the group.

These will wait indefinetely for it to move on.

3) Another ridiculous problem exists with fleeing / withdrawal: an AI group is just plain unable to withdraw as a whole from the battlefield.

Sometimes a single AI unit may start fleeing: when this happens all the group instantly (and magically) does the same.

The interesting thing is that these fleeing groups usually do move slowly away trying to keep into formation, but without caring to keep heads down.

4) AI machinegunners are still insanely inaccurate.

The irony in this is they keep group cohesion when one is stuck but they can't do the same when the AI leader sends them to attack you or your squad.

Also I wouldn't say about mg that are that inaccurate,I've been one shot(my AI pilot too) by them quite a few times when I was buzzing around and my pilot couldn't hit a group with the miniguns.That happened with friendly side on full unit skill while the opfor on zero skill.:o

[bigshot 62]

I have to agree that the AI needs to be the number 1 priority. It's always been the oxymoron of the arma franchise, meaning that it's always been one of the game's strong points AND it's weakest link. It seems lately that whenever BIS put out a patch for AI, however, their aiming accuracy and spotting keep improving (now to the point where it needs to be pulled back alot as default) but they don't seem to be much if any smarter than from years ago. That may not be technically correct, but I can't help but feel that is the way it seems from the "naked eye" of an average gamer.

They still don't follow orders while in Danger Mode, they still have major issues following/keeping up with you, it still takes twice as long to let them drive you around as compared to just driving yourself...these are a few of the very basics you need the AI to do very well in a game like this, but after many years still do not.

There are 3 things that BIS branded themselves upon: Huge landscapes, Dynamic AI, Plethora of drive-able vehicles. It is the combination of these 3 things that to this very day have set BIS and the Arma franchise apart from every other game...and it is also why they have attracted so many talented modders which are responsible for the game still being alive, well and even more popular 12 years after its initial inception.

While BIS has obviously been working hard on visuals and sound (which is VERY important of course) I think it's long overdue that they invest alot back into their AI. I see their AI as part of who and what they are about, it is part of their branding and needs serious consideration if it is to remain that way.

[Touch Off 10]

when you lot finish the game for BI, who gets the money?:rolleyes:

[dnk 13]

I don't think it is so much a problem of the ai having restricted FOV, but of them simply not glancing left and right every so often, or occasionally checking their six. they just stare straight ahead. I thought this was improved upon in a OA beta but it doesn't seem to have carried over to arma 3.

I absolutely agree. I thought I mentioned this, but I clearly forgot. It's both issues, but probably the one mentioned here is more important. "Head on a swivel, wait, scratch that, head on a pike."

Looks like AI prioritize more newly discovered threats than previously detected threats coming again into line of sight.

At least debugging danger.fsm shows that.

Any info on debugging that? I would like to get more into the nuts and bolts of things here.

True. We are somewhat able to script the units to lie down, but we are not able to consistently make them run into cover. We lack reliable commands (findCover? ForceRun?).

This could be solved by the AI map mentioned above, giving different cover values for direction of cover in addition to quality. Then it's a simple matter of locating the nearest directional cover tile and pathing to it, then letting the typical cover usage programming take over.

For this form of the AI map, each tile would have 4 or 6 or however many directional values you like. You can then aggregate these directional values for the overall tile value (when considering movement when not being actively suppressed).

It should not lie down in CQB, while crouched stance may be the preferred one imho.

Well, if they aren't going to find proper cover and they're otherwise in the open in an urban environment, it's the next-best thing to keep them alive for more than 5 seconds.

Anyway, a final note regarding accuracy. I'm currently finding a value of 0.1 for aimingaccuracy and 0.3 for aimingshake to be fairly appropriate. They still get kills up to 250m in under a clip, but they're not all super marksmen between 75-150m anymore (under 75 is near-instant death with any reasonable setting). 0.15/0.4 for can make the mid-far range less "ammo spew"y. I wouldn't push them much higher, though, unless you like short firefights. After considerable testing, aimingspeed seems to have no noticeable impact at all right now. Also for spotting distance, they seem to be hard capped at about 200m, though they can engage from much farther if a squadmate sees someone.

Edited March 19, 2013 by DNK

[fabrizio_t 58]

I have to agree that the AI needs to be the number 1 priority. It's always been the oxymoron of the arma franchise, meaning that it's always been one of the game's strong points AND it's weakest link. It seems lately that whenever BIS put out a patch for AI, however, their aiming accuracy and spotting keep improving (now to the point where it needs to be pulled back alot as default) but they don't seem to be much if any smarter than from years ago. That may not be technically correct, but I can't help but feel that is the way it seems from the "naked eye" of an average gamer.

They still don't follow orders while in Danger Mode, they still have major issues following/keeping up with you, it still takes twice as long to let them drive you around as compared to just driving yourself...these are a few of the very basics you need the AI to do very well in a game like this, but after many years still do not.

There are 3 things that BIS branded themselves upon: Huge landscapes, Dynamic AI, Plethora of drive-able vehicles. It is the combination of these 3 things that to this very day have set BIS and the Arma franchise apart from every other game...and it is also why they have attracted so many talented modders which are responsible for the game still being alive, well and even more popular 12 years after its initial inception.

While BIS has obviously been working hard on visuals and sound (which is VERY important of course) I think it's long overdue that they invest alot back into their AI. I see their AI as part of who and what they are about, it is part of their branding and needs serious consideration if it is to remain that way.

Look like oldtimers share the exact same attitude towards this game ...

---------- Post added at 16:17 ---------- Previous post was at 16:07 ----------

Any info on debugging that? I would like to get more into the nuts and bolts of things here.

Depbo characters .pbo and open "scripts" folder.

You'll find danger.fsm, which you can open with a FSM editor (http://community.bistudio.com/wiki/FSM_Editor).

Also you may wish to take a look here:

http://community.bistudio.com/wiki/Arma_2:_Operation_Arrowhead:_AI_FSM

http://forums.bistudio.com/showthread.php?118989-Research-on-AI-FSM

https://dev-heaven.net/search?q=danger.fsm

Anyway, a final note regarding accuracy. I'm currently finding a value of 0.1 for aimingaccuracy and 0.3 for aimingshake to be fairly appropriate. They still get kills up to 250m in under a clip, but they're not all super marksmen between 75-150m anymore (under 75 is near-instant death with any reasonable setting). 0.15/0.4 for can make the mid-far range less "ammo spew"y. I wouldn't push them much higher, though, unless you like short firefights. After considerable testing, aimingspeed seems to have no noticeable impact at all right now. Also for spotting distance, they seem to be hard capped at about 200m, though they can engage from much farther if a squadmate sees someone.

Lowering accuracy definetely helps in making AI more human, but AimingAccuracy at 0.1 will originate friendly fire accidents, caused by machinegunners especially.

AimingShake is tricky: lower it too much and AI units will be unable to aim nor to open fire. 0.3 may look fine, but fatigued or wounded units are potentially going to have problems.

SpotDistance and SpotTime did not make a whole lot of difference in ArmA2, i dont' know in ArmA3.

AimingSpeed was just useless ...

I think the most negative thing about AI accuracy is that is not affected by morale nor suppressive fire.

That may be scripted, but should have been hardcoded.

Edited March 19, 2013 by fabrizio_T

[dnk 13]

Well, you can always give MG units higher stats. I do need to do longer testing (mostly just <1min test setups so far) with these values still. Thanks for the info, I'll be looking into all that soon.

[Levinin 1]

Something I'm realising is that a bunch of tactics are not truly possible with the AI as is. I have an editor mission where there is a central OPFOR on a hill. I have a blue rifle squad engage front and a second waypointed with "hold fire" to go right and flank on the hill the other side of some rocks at fast pace and then cut in to engage.

That doesn't happen at all! The AI just all engage fairly centrally. It looks as though the AI won't move under fire to take up a flank even when told to hold fire and move. They either move forward engaging as they go or try to do that and get shot to pieces.

[bad benson 1733]

well i think it's good that they react to a threat. this sounds like something that needs some "advanced" mission scripting.

play around with this.

http://community.bistudio.com/wiki/disableAI

and this.

http://community.bistudio.com/wiki/enableAI

this basically allows you to make the AI ignorant to threats. you can use disableAI to make them perform your maneuver without hesitation and then use enableAI to switch them back into their normal awareness. since arma is a simulation you will need to resort to tricks like this to get those cinematic results. they have their own opinion on things which is good in my opinion ;)

[Levinin 1]

well i think it's good that they react to a threat. this sounds like something that needs some "advanced" mission scripting.

play around with this.

http://community.bistudio.com/wiki/disableAI

and this.

http://community.bistudio.com/wiki/enableAI

this basically allows you to make the AI ignorant to threats. you can use disableAI to make them perform your maneuver without hesitation and then use enableAI to switch them back into their normal awareness. since arma is a simulation you will need to resort to tricks like this to get those cinematic results. they have their own opinion on things which is good in my opinion ;)

Thanks I'll have a look at those.

Actually not a good thing if the AI does too much thinking for itself, it's behaving like a bunch of badly disciplined troops who can't take direction. They'd fair much better if they did the flanking instead of attacking head on :-)

[Maslofski 10]

dont forget AI teammates that just turn their complete body in comat and just look at you. or the enemy wall watchers that turn their head to you but dont turn the weapon

[NacroxNicke 11]

One of the things I think it's broken is the "scan the horizon" command, because I think it's a common thing that the AI does per default, and that's makes them to TOTALLY ignore enemies near them. So if you have tried CQB you should have experimented with AI looking aimlessly to the horizon while you are at the side firing at their ground, and they never try to make a turn to kill you, i mean, never ever infiniteness. I remember that in OFP early days the AI didn't looked for targets while doing the same schedule as the "search the horizon" command, and they actually responded in CQB if you were that bad to don't kill them in the first shot.

Check for yourself, put your squad in scan the horizon mode, see how they start spinning the same as in when they don't respond in CQB situations (I mean, the turn the head and not the gun situations).

Edited March 19, 2013 by NacroxNicke

[wipman 1]

Hi, one question about the AI on the ArmA3 Alpha... do the AI squad mates enter in formation facing the direction that theye were created?, instead form up in the told formation facing the direction that the Squad/Team Leader is facing?.

On the ArmA2, that mission where you had to try to leave the island in time... if you's ordered your AI mates to enter into Line formation... as they were created looking West, for get that formation shape looking South... you had to order 'em to enter into Column formation. Still the same on the ArmA3 Alpha?. Let's C ya

[NordKindchen 12]

refreshing this thread;) Its an important one.

[mrcash2009 0]

I think in most dev blogs the only AI changes they mentioned was the looking for wall position for cover was lessened, but all the way though Ive never really seen anything BI's end mentioning AI specific changes in these area.

I dont have Alpha but seeing as confirmation is coming from Fabrizio_T (AI Fiddler 101 :)) then its a shame things still seem to be Arma2 vanilla requiring the usual suspect mods to be ported over to get it better.

I have a question, when you get your team to form up while walking, do you still have it that the AI will lag behind or start going too far forward, so you have to stop still for them to realise, then they stop and back step into formation?

[-Coulum- 35]

Fixed: AI no longer fires on targets it does not see (but which are reported by other group members)

This should help lessen some of the complaing about xray vision ai. I am glad to see it and happy to see BIS still tweaking the ai. Next step is to stop their super memory and precise/instant info sharing along with their "sixth sense" they gain when they spot you.

[whisper 0]

I have a question, when you get your team to form up while walking, do you still have it that the AI will lag behind or start going too far forward, so you have to stop still for them to realise, then they stop and back step into formation?

I think it's still the case.

But tbh, do exactly the same with player-filled team, and see the result. The AI will look incredibly organized, compared ;)

[instagoat 133]

At bottom I can confirm all of these problems, and I would guess it´s virtually the same for everybody with this as it was in Arma 2.

Currently writing up a list of things to prove/disprove about the AI for my own testing thread. Pretty much all that´s mentioned here´s already included, and more. I hope I will find the time to work through all of those and provide some data.

I´m also learning to use troopmon (downloaded it today) for figuring out what´s going on with the AI. The more light we can shed on this, the better we´ll be able to define what and how we want it changed.

Or, if it turns out that the inherited problems are too big, we can maybe lobby for a complete bottom to top new build for the AI, but I doubt that´ll happen. :I

A lot of old problems were config problems. However, the system may turn out to have native problems unrelated to any config errors, and that is when we´ll need to see what we can deal with, and what BI can provide to fix whatever the community finds.

[Jex 1]

Some valid issues and some nice names to go with them! Can't really reply now but I agree with most of what you are saying.

I don't think it is so much a problem of the ai having restricted FOV, but of them simply not glancing left and right every so often, or occasionally checking their six. they just stare straight ahead. I thought this was improved upon in a OA beta but it doesn't seem to have carried over to arma 3.

When they changed the Zombie's in DAYZ to be looking where their heads were turned rather than a fixed FOV coming out of their chests (is the best way I can describe it) it made the game impossible not to be seen. Obviously the zombies looked around lots more like mental cases but this is what the devs would need to look out for as well. AI wouldn't turn their heads that much or would they. I do in RL but i'm not a computer that sees everything around me. If I look around whilst talking to someone I might not be concentrating on what I'm looking at, maybe just glancing around bored and would miss things.

I always thought the AI should be modeled along these lines.

AI has x% chance to spot the player - this is a base level to reflect their alertness and skill and environment (drunk militia vs trained soldiers). This could be 50% for this example. (You could also break it further down and have 2 or 3 factors rather than just one and make quite a complex algorithm for how the AI look around and see things.

Then other factors are assigned to this base value as either a + or - (remember these are only examples)

- Distance (-5% per 100m away from the player, say)

- light (- x% depending how dark it is, modifiers for flashlights or NVG's)

- Weather (thick fog will make it harder

- Objects (how much does an object cover a player and also can you see through it, like a bush perhaps)

- Movement (if the player is moving then the bonus applied would go with how fast they are moving)

- Distraction (the FOV could narrow to soldiers distracted or caught in a fight. Normal would be 160 perhaps but in a fight it narrows to only 20 degrees until the threat has gone say, or maybe better trained soldiers won't get as much of a reduction.

and the list can go on for what other relevant factors there are but you get the idea. You could even add in camo values, bonuses if the AI use bino's, etc

What we don't want is robot eyed AI that merely have to glance in your direction to see you, or dummies that won't see you because their skill lever only let's them notice you at 100m and you're at 101m. I would like a set of values that reflect the different factors involved in how we see things in real life and also add in a random factor as well because nothing is set in stone.

---------- Post added at 13:58 ---------- Previous post was at 13:51 ----------

This should help lessen some of the complaing about xray vision ai. I am glad to see it and happy to see BIS still tweaking the ai. Next step is to stop their super memory and precise/instant info sharing along with their "sixth sense" they gain when they spot you.

It would be nice if the AI made an animation of using their radio to call in the spot but if they get killed before they can do that, the call never goes out and the other enemy don't get that information.

Edited April 4, 2013 by Jex