AI reluctance to fire

AI reluctance to fire (Full Version)

All Forums >> [New Releases from Matrix Games] >> Tigers on the Hunt >> Tech Support

Message

UP844 -> AI reluctance to fire (5/17/2017 1:56:44 PM)
In the lunchtime break I played a game of "Hedgehog of Piepsk". The Germans have 7 1st line squads with 4 leaders and 1xHMG, 1xMMG, 3xLMG and a Radio with an 80mm Mortar OBA. I set up the HMG, the MMG and the radio on the hill so as to fire at the Russians as far as possible, hoping to reduce their numbers before they come close. Two squads with a LMG each monitor the flanks and a single squad in the center keeps ready to move in case the Russians come too close. The Russians have 36 squads (2/3 1st Line, 1/3 Conscripts) with 3 leaders, 1xMMG and 2xLMG. They set up along the bottom border of the map and must take a dozen VP hexes in the village beyond the German line. Despite their numerical superiority, they have to advance in face of long-range MG fire with little cover (some woods). [image]local://upfiles/53779/EC73CD9FBF864FE09CD89CA15B0DC3FA.jpg[/image]

UP844 -> RE: AI reluctance to fire (5/17/2017 2:00:40 PM)
And this is the end-game screen: a few Russians managed to come close (4 hexes) to the westernmost VP hexes by infiltrating through the woods, but they were stopped and repulsed by the reserve squad. Most of the remaining Russians are broken at the bottom of the map, harassed by long-range MG fire. After turn 6, most fire was directed on broken units, to keep them at broken+ status (and to kill the occasional one [:D]). [image]local://upfiles/53779/90DB6C502A3D469DB2EBA50625586C7D.jpg[/image]

UP844 -> RE: AI reluctance to fire (5/17/2017 2:14:55 PM)
During the game, I took notes about the number of times both sides fired: as you can see in the Excel sheet below, the AI fired very few times (and, accordingly, caused very few losses to the Germans). (on a side note, the German OBA was lost on Turn 1 and the HMG spent several turns broken). The AI fired 8 times vs 138 times the German fired: with a 17.25:1 fire exchange ratio, the 28:1 loss ratio should hardly come as a surprise. If the AI is allowed to fire even when its firepower is reduced (long-range fire, pinned firer or both), it will not become a master infantry tactician, but at least it will provide a more worthy opponent. On turn 2, I took note of how many times the AI could have fired: in the Advancing Fire, it could have fired 4 times and in the Defensive Fire it could have fired 9 times. In both cases, it never fired a shot. [image]local://upfiles/53779/0A386FF423274FFCBF09DAD4BD797DE2.jpg[/image]

rico21 -> RE: AI reluctance to fire (5/17/2017 3:49:52 PM)
Hypothesis number one Below a percentage chance of touch, no shot Hypothesis number two AI selects advance shooter and non-shooter and does not shoot when needed Hypothesis number three AI (or Peter) like fishing[:D][&o] [image]local://upfiles/53841/F4996CD153F1448A98FC340C2CE662E3.jpg[/image]

UP844 -> RE: AI reluctance to fire (5/17/2017 4:03:14 PM)
According to my observations, AI-controlled units only fire if their firepower is not halved. Specifically, they never fire: After moving: in the table above, you can see the AI never once fired in the Advancing Fire segment. I only rarely saw AI units fire in this segment after they have moved: they only did so when they were adjacent to an enemy unit and unpinned. When they are pinned. I do not know for certain if they would fire at an adjacent unit (it never happened in this scenario). When marked with "First Fire": they never use Final Fire, unless the target is adjacent. In any scenario not taking place in an urban setting or with very limited visibility, this gives a huge (and fully undeserved) advantage to the human player, which already enjoys far greater intelligence than the machine.

UP844 -> RE: AI reluctance to fire (5/17/2017 4:05:52 PM)
P.S. In my opinion, this is the greatest issue with ToTH, ranking far before than new terrain, nations and vehicles. If the same scenario is played with Russians, you only have to use the Conscripts as MG bait and then advance at range 7 or 8: then you can overwhelm the Germans through your superior numbers (even if you fire at half FP).

Peter Fisla -> RE: AI reluctance to fire (5/17/2017 6:24:34 PM)
AI will never be perfect, I will try of course to make changes and improvements for the future. Never mind the fact working on AI is a lot of hard work. The AI has to consider many variables, situations already. You know, you can always add more units to the AI when making new scenarios and/or play on higher difficulty settings.

Jeffrey H. -> RE: AI reluctance to fire (5/17/2017 6:58:02 PM)
quote: ORIGINAL: UP844 During the game, I took notes about the number of times both sides fired: as you can see in the Excel sheet below, the AI fired very few times (and, accordingly, caused very few losses to the Germans). (on a side note, the German OBA was lost on Turn 1 and the HMG spent several turns broken). The AI fired 8 times vs 138 times the German fired: with a 17.25:1 fire exchange ratio, the 28:1 loss ratio should hardly come as a surprise. If the AI is allowed to fire even when its firepower is reduced (long-range fire, pinned firer or both), it will not become a master infantry tactician, but at least it will provide a more worthy opponent. On turn 2, I took note of how many times the AI could have fired: in the Advancing Fire, it could have fired 4 times and in the Defensive Fire it could have fired 9 times. In both cases, it never fired a shot. [image]local://upfiles/53779/0A386FF423274FFCBF09DAD4BD797DE2.jpg[/image] I like the quantitative approach you are taking. Don't you think it should be done with reversed roles to get a more clear picture ?

UP844 -> RE: AI reluctance to fire (5/17/2017 9:36:25 PM)
@Jeffrey H. : Great avatar! [:D] I spent the dinner break to replay the scenario using the Russians. I set up on the bottom edge, west of the hill, with two groups of a dozen 1st line squads each (one west, one east) and the conscripts (a.k.a. the suicide scouts) in the centre. The conscripts ran forward without taking fire, so I guessed the German AI does not share my preference for the high ground. In fact, on turn 1 there was no fire at all and no Germans were spotted. On turn 2, the first Germans appeared, inside one of the VP building hexes. In the meantime, my two-part pincer plan went into the dustbin of history thanks to C&C (hint: never try sophisticated moves with Russians [:D]). I then reverted to a classic Russian steamroller approach (which does not mean a blind charge into enemy HMGs). [image]local://upfiles/53779/26BFB54E749B49DB96FD6195DA4C3402.jpg[/image]

UP844 -> RE: AI reluctance to fire (5/17/2017 9:45:55 PM)
And here is the end-game screen: the Germans were wiped out to the last man, but only managed to kill 2 squad equivalents (1 squad, 2 HS). The game ended with a Russian major victory at the end of turn 7, and on Turn 8 the last remaining German units (a leader and a broken squad) were also eliminated. For some reason, the German OBA never fired, and this is one of the reasons for such a low number of Russian casualties (another is the sneaky tactics I used [:D]). Most of the German casualties were not caused by fire: one leader and 3 1/2 squads were eliminated for Failure to Rout, two other leaders and two squads were killed in CC (usually at overwhelming odds: these are Russians, after all). Russian fire, therefore, only managed to kill one leader and 1 1/2 squads. [image]local://upfiles/53779/4B601F8BDB7B4126A15BAD71546E5D65.jpg[/image]

Peter Fisla -> RE: AI reluctance to fire (5/17/2017 9:49:40 PM)
Ok UP844, I was thinking...I can add a special attack command for AI personnel when the AI side that when assigned to an AI personnel unit, this unit would focus on taking on VPO hexes or a specific hex no matter what the odds. Kind of like suicide squad/assault squad/human wave. I can add this functionality into the scenario editor for the next update. What do you think? What other scenario designers think ? Cheers!

UP844 -> RE: AI reluctance to fire (5/17/2017 10:02:08 PM)
Finally, here is the fire analysis table: the Germans AI fired much more often than the Russian AI did, but fighting within the village meant it almost always fired within the normal range of the German infantry (6 hexes). The AI fired 33 times vs 71 times the Russian fired: fire exchange ratio was 2.15:1 but, as noted before, the ranges involved were all within the normal range of the German units. Russian fire effectiveness has been abysmal (1.5 squads and 1 leader killed) but almost all fire was made by individual squads against targets in buildings and 50% of it took place in the Advancing Fire segment after moving (35 fires out of 71). This is hardly a recipe to obtain straight KIA, even though it pinned/broke several German units that were subsequently killed through FTR or CC. [image]local://upfiles/53779/D7323FB9F7C0494E851D00EB87AE39E0.jpg[/image]

UP844 -> RE: AI reluctance to fire (5/17/2017 10:09:33 PM)
Excuse me, Peter: I am more or less a computer illiterate and so I am unable to estimate the technical issues involved, but it will be so difficult to have the IA fire even when its FP are halved? This change alone will make the scenarios much more challenging with no need to add hordes of AI-only units. The Germans are somewhat tough customers anyway because of the large number of MGs they usually have, but Americans, with all their firepower, can be easily slaughtered (try playing "Canicatti Ridge" with the Germans and take note of how many times the American squads fire).

Peter Fisla -> RE: AI reluctance to fire (5/17/2017 10:15:05 PM)
quote: ORIGINAL: UP844 Excuse me, Peter: I am more or less a computer illiterate and so I am unable to estimate the technical issues involved, but it will be so difficult to have the IA fire even when its FP are halved? This change alone will make the scenarios much more challenging with no need to add hordes of AI-only units. The Germans are somewhat tough customers anyway because of the large number of MGs they usually have, but Americans, with all their firepower, can be easily slaughtered (try playing "Canicatti Ridge" with the Germans and take note of how many times the American squads fire). This is a lot more complicated because firing long range has other implications like weapon breaking etc...while the weapon could have been used instead when an enemy unit is in full fire range. I will look into to see if what I can do for the next update....but I'm not promising anything.

rico21 -> RE: AI reluctance to fire (5/17/2017 10:21:45 PM)
I do not know enough about how the AI works. In general, an AI obtains a priority but can change when it encounters certain events. Personally I prefer to be able to script AI with trigger events. I am aware that it is not simple and any improvement will be better than nothing. We will do with what you give us! Thanks.[sm=00000436.gif]

UP844 -> RE: AI reluctance to fire (5/17/2017 11:13:07 PM)
quote: This is a lot more complicated because firing long range has other implications like weapon breaking etc...while the weapon could have been used instead when an enemy unit is in full fire range. I will look into to see if what I can do for the next update....but I'm not promising anything. I am afraid I have not been able to clearly explain my worries. The issue I reported has nothing to do at all with AI aggressiveness when moving (): the AI Russians charged forward to the last man (literally: a single Russian leader surged forward when all the remaining units were broken several hexes behind him()). () There was an issue with shy tanks, but you solved that a long time ago. (*) This indicates another issue I left out: all too often leaders abandon broken squads and move towards the enemy all by themselves: 99% of the times this means the leader gets killed in CC and the broken units only have a minimal chance to rally through self-rally. quote:* This is a lot more complicated because firing long range has other implications like weapon breaking etc...while the weapon could have been used instead when an enemy unit is in full fire range. I will look into to see if what I can do for the next update....but I'm not promising anything. Playing the devil's advocate against myself [:D], I can see how a smart player can expose a unit at long range so that the AI wastes its fire opportunities against it, and then move other units within the full fire range knowing they will be safe. This is a problem that only concerns the defensive fire taking place in the Movement Segment, however. It could be countered by having the AI firing only after it passes a check. By way of example, an AI-controlled firer with a 10% chance of damaging a target, will have a 10% probability to fire. If this is too complex, fire at long range can have a fixed "X%" probability of taking place. What's important is that the human player will no more be able to be safe when it faces the AI at long range. For all other types of fire, it looks to me as if the AI will only have to consider more targets before selecting the one it wants to fire to. Currently, if the only available targets are at long range, the AI will not fire at all. This can be entertaining one or two times, but when I see Russians pouring huge amounts of firepower on longer-ranged Germans who only reply with scattered MG fire I think there is something wrong... Finally, I know I am an outstanding pain in the *** and I beg your pardon for bothering you so much, but I am raising this (and other) issues this because I really, really love this game.

UP844 -> RE: AI reluctance to fire (5/20/2017 2:41:04 PM)
Some updates after playing Rico's last scenario: AFVs, unlike infantry, fire in the Advancing Fire Segment (mostly with MGs, but sometimes with their MA too) at ranges greater than 1 hex (so, their MG fire should be halved). I noticed several unbroken leaders routed with some broken units, but then advanced out of the hex in their Advance Segment. Then, on the subsequent turns, they charge forward all alone. Wouldn't be possible to inhibit advancing for any leader that is in the same hex with broken units at the start of the Advance segment?

Page: [1]

Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI

1.484375