TOWARDS A REALISTIC SCENARIO DESIGN ETHIC

[Franko]

I am starting this thread for the purpose of coming up with a set of standards/observations/guidelines, about how the AI behaves, for the purpose of designing realistic scenarios that will be challenging and enjoyable.

To that end, I am posting some observations, and would like you all to weigh in. I will edit this post occasionally to update this. Naturally, all of these points can be modified, if you have some better knowledge to share. Again, we are working to develope a set of guidelines based on how the AI behaves, so we scenario designers can avoid common mistakes which result in crappy scenarios.

1. Don't have the "computer" opponent be mounted on tanks or trucks. The reason apparently is that the computer doesnt know how to handle mounting, dismounting (has anyone ever seen the computer player have a unit "mount" a vehicle?). Is this true?

2. Don't design scenarios where the computer is allowed to setup, period. Is this a good rule? Does the AI ever do anything intelligently?

3. Don't make the computer player be the "Attacker", because it won't move well and lacks agression. This is a common assumption, but is it true? Is there something that can be done from a design standpoint to make the AI give you fits?

4. This space reserved for the definitive ruling on whether setting an exit zone on one side or the other "encourages" the AI to behave a certain way.

5. This space reserved for a rule on whether to count on the AI to use on-board mortars effectively.

6. This space reserved for a rule on whether the computer can use roads effectively, or avoid using them when it would be suicidal to do so.

7. This space reserved for some observations on how AI commanders (e.g., platoon leaders) behave vis-a-vis their subordinates.

8. This space reserved for observations on how the AI chooses target priorities.Treeburst states that the AI targets the most "experienced" unit it sees, and it knows what weapons that unit has (e.g., Panzerfausts). True?

9. This space reserved for observations on how the AI is effected by limited visibility.

10. Scott B -- raises an issue as to curious behavior when the AI is set to "hide" at the beginning of the scenario. Developing.

11. It appears that the human advantage can be mitigated somewhat by the use of harsher weather conditions and more static defenses on the AI side -- serving to level the playing field (Lindan).

I think we'll all benefit from this. Although this post is abstract, I think I communicated what I want. A set of principles whereby designers can make effective scenarios.

Frank

[ May 28, 2002, 02:09 AM: Message edited by: Franko ]

[Treeburst155]

Given any single unit type, the AI will ALWAYS target the most experienced unit it "sees". This is true even if the more experienced unit is further away and appears later in the turn.

The AI knows which squads have panzerfausts too. I'm not sure about rifle grenades. It's safe to say the AI knows everything about your units the second they are spotted, and will select targets accordingly.

Treeburst155 out.

[Gyrene]

The AI is a gamey bastich.

Gyrene

[Pzman]

Also in many cases the AI has trouble controlling forces over company in size. Whether it be a meeting engagement, attack or a defensive battle.

[Michael Dorosh]

I'm glad you started this thread, Franko, because you made some great points in the other thread about scenario creation.

IIRC, you stated that you were dissatisfied with the level of understanding that scenario designers have been able to achieve of how the AI and TacAI "does business"? Am I reading you correctly?

Gathering intel in a thread like this is a great idea.

I would like to see some pointers, however, on victory conditions. We've had many discussions of same over the last few months, and I think they are especially pertinent, since a scenario is a complete waste of time if both sides don't have a chance to "win" it.

I pretty much regard any scenario that I download, or attempt to create myself, as being for two-player play.

Are you speaking specifically only about scenarios especially designed for play against the computer? I think vs-AI and vs-human deserve to be put into two completely categories of discussion, for as you point out, there are a lot of limits on what the computer can do, beyond the obvious of, say, assault boats and continuing on through your observations of mounting, attacking, etc.

[Scott B]

I've seen strange behavior in computer forces that begin a scenario "hiding." Sometimes they move out, and sometimes I'll finish the game and they'll be in their starting position, still hidden. Could anyone comment on how the computer handles troops that begin a battle hidden when they are not supposed to be defending?

Scott

[Franko]

Originally posted by Michael Dorosh:

I'm glad you started this thread, Franko, because you made some great points in the other thread about scenario creation.

IIRC, you stated that you were dissatisfied with the level of understanding that scenario designers have been able to achieve of how the AI and TacAI "does business"? Am I reading you correctly?

Gathering intel in a thread like this is a great idea.chance to "win" it.

Are you speaking specifically only about scenarios especially designed for play against the computer? I think vs-AI and vs-human deserve to be put into two completely categories of discussion, for as you point out, there are a lot of limits on what the computer can do, beyond the obvious of, say, assault boats and continuing on through your observations of mounting, attacking, etc.

You are reading me correctly. Hit the nail right on the head, as it were.

F

[Franko]

Originally posted by Gyrene:

The AI is a gamey bastich.

Gyrene

Funny. But not helpful.

F

[Michael Dorosh]

Well, one subject I would like to see addressed is the problem of reinforcements arriving. I like the use of random chance in scenarios and wish the editor allowed us to do more of that sort of thing - but we are not here to put up another wish list, but to discuss how to work with what we've got.

I tend towards using multiple reinforcement groups, with at least a little bit of randomness thrown in, for variety's sake and also because I think the AI handles small reinforcement groups better than large ones. It is rather jarring - and I just noticed this in a game with Captain Wacky - when a company of reinforcements arrives out of thin air in the midst of a firefight. Especially if they are poorly deployed. They also don't know how to logically arrange themselves mounted so that the smallest groups sit in the kubels, say, and the squad sized units go in the trucks. In my experience, some enter walking instead of shifting around to all mount up properly.

I suppose if you want them dismounted, you should enter them as seperate reinforcement groups arriving in the same place?

I also tend to try and locate their arrival away from the action. Sorry if this seems obvious and elementary.

But I also try and break them up a bit and spread their entry around, where possible. It helps replayability, at any rate.

But at what percentage chance are you likely never to see a reinforcement group?

Simple math tells me that a 50% chance of reinforcements arriving on turn 5 pretty much guarantees their arrival by turn 7. Is it really that simple? Has anyone noticed anything out of whack with the way the computer handles these numbers?

[ May 28, 2002, 01:50 AM: Message edited by: Michael Dorosh ]

[Lindan]

Hi Franko,

this surely is a valauble thread for scenario designers. In fact, you stated nearly all the reasons why the AI is what it is: it helps you pratice the game as a sparring partner but can't substitute a human playing you. (because it doesn't use roads effectivly, embark/disembark, attack behaviour, needlessly shuffling troops around on defense etc., etc.)

What improves the enjoyment of playing a scenario vs. the AI is, IMO, making weather conditions and time setting far from ideal. a foggy night battle can be great fun vs. the AI. the less intel the human player has, the more he has to plan ahead and the less is the impact of his superior skills at evaluating the situation. giving the AI assets that it can't misuse, like bunkers, minefields and such also helps a lot.

just my 2 €-cent.

[Ozzy]

The AI is a gamey bastard regarding extreme flag rush. I want to say that he is VERY flag oriented.

(a flag fetishist...? :confused: )

In a test battle I once setup a realistic WWII British Armoured Battalion with support of 1 battery 40mm Bofors to test the success of a fighter-bomber squadron (6-10 FBs) against the amd formation. I placed three large flags randomly on the map to spread out his forces. Within three or four turns he rushed 25 tanks to each flag. Of course, he lost 24 vehicles (including Gun Damage and Track Hit) to the FBs...

In another occasion during a scneario testing phase I wanted to determine how many and where victory flags to place. I played defender in a small town behind a narrow river and 3 bridges crossing it, AI is attacker with dense wood on his right flank and open ground to his center and left. Initially I placed two large flags, one in the town and one on the rightmost (AI view) bridge (considered primary objective). AI has 1 rifle bn., with arty, AC, halftracks and 8 AFVs. I have two rifle coys, arty, some HMGs and handheld AT, and four Close support tanks.

What was his tactic? He sent virtually ALL his vehicles full speed over the open ground, and ALL his infantry through the forest, which ended at the mentioned bridge (excellent artillery target).

The test battle ended with the AI surrendering and me having lost 31 men (12 KIA)...

Was pretty much a kind of turkey shooting....

Conclusion: if the scenario is designed to play against AI, place 5 or 6 small flags rather than 1-2 large ones. If not you will see REALLY foolish and unrealistic moves from the AI.

[ May 28, 2002, 04:14 AM: Message edited by: Ozzy ]

[Brian Rock]

Originally posted by Franko:

3. Don't make the computer player be the "Attacker", because it won't move well and lacks agression. This is a common assumption, but is it true? Is there something that can be done from a design standpoint to make the AI give you fits?

I've had limited success by micromanaging set-up areas. If there are five possible routes to advance down and two will get the attacker killed, set up the AI attacker so it won't take the non-lethal routes.

Not always possible of course, but a useful principle to keep an eye out for.

Also setting units up in an "assault formation" can help, putting contact elements well in front of the follow-on forces. In theory, that is - this one seems to break down once the AI hits wooded areas, or if there are tricky pathing problems.

[Hoopenfaust2]

Franko you said that the AI doesn't attack very well. Sometimes he does, sometimes he doesn't. I just finished "Hello Second Armour" and I found the AI (germans) attacking quite well (infantry). But the AI is too predictable.

AI does not use TRPeez that well. I've created a scenario where the allies are visible in the assembly area, with TRP everywhere. He does not use them ever.

[WWB]

It is about time for one of these discussions. We have had a good 2 years to make battles for CMBO, so we should have that mystifying AI thing figured out by now.

Originally posted by Franko:

[QB] Snip . . .

1. Don't have the "computer" opponent be mounted on tanks or trucks. The reason apparently is that the computer doesnt know how to handle mounting, dismounting (has anyone ever seen the computer player have a unit "mount" a vehicle?). Is this true?

I have never seen the PO mount a vehicle, but I have seen it dismount. It only dismounts when under fire, near as I can tell. It will not do something like "drive to the back of the woods and dismount infantry platoon." I do think starting the AI mounted is valid, when you want it to drive into an ambush. Also, it does go a long way towards keeping tanks moving with infantry, instead of having them blindly rush into AT ambushes.

2. Don't design scenarios where the computer is allowed to setup, period. Is this a good rule? Does the AI ever do anything intelligently?

I pretty much agree with you here, especially for the defensive. For the offensive, it can do a half decent job, because it sets itself up in a manner which it wants to attack in. The major weakness here is that it pays no mind to formation, leading to lots of broken, out of command infantry.

3. Don't make the computer player be the "Attacker", because it won't move well and lacks agression. This is a common assumption, but is it true? Is there something that can be done from a design standpoint to make the AI give you fits?

The AI can attack, in the right situation. I would not ask it to assault a dug-in foe in a dense urban environment. Making an effective AI attack requires the right kind of map, the right kind of forces, and a proper flag deployment. Moreover, it is quite unpredictable how the AI will react to a given flag deployment and map setup until one actually tests it. It tends to work towards big flags, going over little ones on the way, but aside from that nothing is certain.

4. This space reserved for the definitive ruling on whether setting an exit zone on one side or the other "encourages" the AI to behave a certain way.[./quote]

Good question, anyone got time to run some tests?

No. The AI does best on the offensive without heavy weapons at all. It tries to run them forward like regular infantry. This, for some godawful reason includes FOOs. On the defensive, it can make effective use of on map mortars, but will not fire them indirectly.

The AI definitely follows the path of least resistance, usually roads. Great way to guarantee a nasty ambush is setup the AI with AFVs and a road into a defended village. They will come naked and alone. Which is why mounting infantry on them makes sense, it will give the AI a fighting chance.

The AI really does not have a concept of subordinates. In setup, it deploys platoons all over the place, and then it just rushes forward as a mass. One of the better AI improvement ideas I have seen floated about is from Mr. Johnson, who suggests giving the AI lots of high-bonus Coy or Bn. HQs to stiffen it on the attack.

BTS--if you are listening---One thing that might help it, but would be a cheat , would be to make all the AI platoon leaders effectively company commanders. It does a good job of distributing leadership, it just does not keep squads together. So, if any HQ could command any squad it would be helped immensely.

I suspect the AI targets using the Tac AI, just like a human player would should he not use the target command.

Lindan's suggestion ties into this. One thing the AI does not do well at all is mass fire, so limited visibility helps it inasmuch as massing fire is difficult with limited LOS. Limited visibilty also helps the AI out with the fact that it really does not understand LOS and can move and mass unobserved.

OTOH, I find that the AI does a lot better when it is in contact with the enemy. Usually the pre-contact turns are spent aimlessly wandering until someone gets shot at. Then it tends to get its stuff together.

Also, the AI really likes to re-arrange your carefully planned defense when it cannot see the enemy just yet. Making the AI want to stay at home a bit more on defense would help it out a bit.

Scott's observation is good. See McAufflie's Belgian CM homepage's scenario design guide, he mentions this as well.

So long as the static defenses are padlocked . . .

One more thing (for now): there is a grave pitfall in scenario design of attempting to be all things to all people. When I design a battle I design for one aspect (2 player, single side). If it works for any of the others, great! If not, I just advise the players in the briefings. Not that people read briefings, but at least I can say they were warned.

[Gpig]

originally posted by wwb_99

The AI really does not have a concept of subordinates. In setup, it deploys platoons all over the place, and then it just rushes forward as a mass. One of the better AI improvement ideas I have seen floated about is from Mr. Johnson, who suggests giving the AI lots of high-bonus Coy or Bn. HQs to stiffen it on the attack.

BTS--if you are listening---One thing that might help it, but would be a cheat , would be to make all the AI platoon leaders effectively company commanders. It does a good job of distributing leadership, it just does not keep squads together. So, if any HQ could command any squad it would be helped immensely.

Hey! That's a great idea!

I'm always amazed at the AI's lack of C&C. Giving all the A.I.'s platoon leaders a +2 command bonus/radius, that would at least help some.

Also, (even though this is cheating as well) could you indeed have an enemy force controlled by the A.I. with just Company commanders? No platoon leaders at all? Or does deleting the platoon leader in the unit editor delete the platoon as well? (I've forgotten.)

I suppose you (as the scenario designer) could institute the ugly cheat of adding EXTRA company commanders across the battlefield. These company commanders would have great bonuses and +2 command bonus/radius. This way many more A.I. controlled squads would remain under C&C and give a better account of themselves.

Oh. I just reread the bit about Mr. Johnson saying this exact same thing. But now that I've typed this all up, consider it a supporting position.

Gpig

P.S. I think I'll try this for me-self.

[WWB]

Gpig: No, deleting platoon leader deletes the squad. You can add extra BN and coy commanders though, by buying companies and deleting everything.

WWB

[Scott B]

Interesting. On the subject of workarounds, there are a few things I've experimented with to put a little more "fight" into the AI.

Antitank guns - you can "set the timer" on these by angling them away from an area where the enemy is expected to appear; if you don't want the AI to shoot at the first enemy vehicle that comes into view, instead waiting for a number of the player's vehicles to expose themselves, you can rotate the gun so that it faces away, and will take longer to be able to engage. Those guns with exceptionally slow rates of traverse can take up to two turns or more to spin 180 degrees, which can be plenty of time to let the player move vehicles into the "kill zone."

Artillery - a human player often has artillery missions plotted on suspected enemy positions such that if an enemy does appear there, the wait for the fire mission will not be so long. While the AI will not deliberately target a TRP, it still benefits from its use; if you want the AI to have faster artillery, it might be worth buying it a few dozen more TRPs and marking all the likely terrain, as well as boosting the AI spotters' quality.

Reinforcements - a bit heavy handed, but timed reinforcements can be a better way of coordinating an AI's movements than expecting it to just move through the "covered line of small flags." Since the AI commonly zooms its tanks forward in a vulnerable fashion, don't give it all of its tanks until some point midway through the battle, when it will likely have located reasonable targets.

A note - it may just be me, but I've never seen the AI be able to place antitank guns that begin a battle or arrive still mounted on their transports.

My apologies if this is too much of a digression from the initial topic of the thread. This is a very interesting discussion. Also, thanks, Wyatt, for pointing me toward McAuliffe's Belgian CM site with regard to my earlier post.

Scott

[Ace Pilot]

Great idea, collecting all these scenario design lessons!

Originally posted by Franko:

2. Don't design scenarios where the computer is allowed to setup, period. Is this a good rule? Does the AI ever do anything intelligently?

Frank[/QB]

I've had some success allowing the computer to set up its forces. At the very least, this needs to be an option for a scenario to be replayable - otherwise you always know where the enemy units are at the start.

In cases where I've had success, it's generally when the AI was on the defense, and the set up areas were well tailored. I've also noticed that changing a scenario to switch the AI from defense to a meeting engagement results in a very different set up. In a meeting engagement, it seems the AI spreads out its forces in the set up area, whereas in a defense engagement, it does a better job of establishing a defense around flags. A better understanding of this ought to be useful in some way.

Speaking of flags, does anyone have any methods or ground rules for helping determine what kind of flags to use and how many? I was playing around with a dynamic flag scenario and put together a small Excel spreadhseet to see how changing the value of the flag would change the victory outcomes under different possibilities (50%/50% attacker/defender casualties, 25%/25%, 75%/75%, etc.). Any thumb rules on what the "range of outcomes" should be, i.e., taking the flag ensures at least a minor victory, regardless of casualties, and minimal casualties earns you a total victory, perhaps?

Ace

[CombinedArms]

Originally posted by Franko:

7. This space reserved for some observations on how AI commanders (e.g., platoon leaders) behave vis-a-vis their subordinates.

Frank

I once did a test of the effects of various types of Allied arty on an Axis infantry battalion in foxholes--Axis was defender, troops were all regular, and there was no fog of war, so I could check arty effects. I had about 8-10 FOs blasting away and discovered to my great surprise that the squads would rush forward to attack the flags in the center of the board, even under heavy arty fire, whereas platoon leaders just sat in the foxholes and didn't move. Eventually, it looked like a series of well spaced cribbage pegs (the platoon leaders in their holes) surrounded by the empty foxholes of their leaderless squads. This was consistent through half a dozen different tests. Very strange, and it suggests that the HQs are less prone to aggressive action than their squads. I don't know if this is a general phenomenon, but it really surprised me and might explain the disconnection between leaders and troops.

[russellmz]

these are the most relevant links i have in the faq. other scenario tips are there as well taht might be useful. the part about quick battles has lots of things that make the ai a little tougher to beat. plus an equation or two to help with scoring as requested above...

How is scoring done?

http://www.battlefront.com/cgi-bin/bbs/ultimatebb.cgi?ubb=get_topic;f=16;t=019346

Also, additional explanation of exit points and killing units.

How to make a good Quick Battle (has many excellent ideas to improve AI behavior)

Scenario Editing Guidelines:

Suggestions with comments, located in the Scenario Editing Forum.

Tips and Tricks of CM Scenario Design

How can I put a gun in a house?

Same as above, place unit where you want, then put a terrain tile over it.

How can I have a pillbox with a turrent?

How do I make the AI attack better/more aggressively?

There are several ways that may or may not work.

TIPS & TRICKS

Swap the map edges. So if Allied friendly edge is West, make it East in the scenario and vice versa for the Axis. Making an exit zone in the direction you want the enemy to advance also may help. You can also leave a path of flags for the AI to follow.

Also there is this tip:

Originally posted by James:

Here are some tips on senario design which give better AI performance from the infantry and more unpredicatble better fought meeting engagements. Here are the three steps.

First: In map preview split all squads. This may seem strange but what happens is the split squads behave with more caution, a more sensitive TacAI - they react more sensibily without needless sacrifice. Also the global moral of both sides drops to 55-70% making the overall AI more cautious and less prone to giving you a turkey shoot. The infantryman in CM is abstrated into squads, by splitting the squads that abstraction is also half as less.

Second: Remove all victory flags. What!? I hear you say! The trouble with VLs is that you know where the enemy will head for, even with dynamic ones, and all you have to do is get there first into cover to let the AI force his men relentlessly into your incoming until their all casulties or broken.

Third: Place exit zones for each side opposite each other. ie Allies start in the west advance to exit zone in the east and the opposite for Axis. What it actually does is give the Strat AI a fine balence between attack and defence. It will utilse both strategies with flexibility according to the situation and use any part of the map. It even withdraws in a hot spot to a better position. Actual exiting by the AI hardly ever happens, and there is no need for you to exit if you are concentrating on defeating the enemy. The exit zones are there not there as an objective but as a modification to the behavior of the AI.

These 3 steps may seem a bit strange but try designing a senario useing them all. Even an experienced CM player can actully be given a real challenge and be beaten by the AI without it having superior weight of numbers! I use forces of approximatly equal point strength all the time and the AI really gives as good as it gets. I have had some excellent battles and lost a number of times, drawn most times, won sometime. Not only does it feel more realistic it is a better game for useing these 3 steps.

not in the faq, but personally i try and do as many things as possible. i switch friendly edges, make exit zones, make ten big flags in one spot to make the ai rush there, split squads to reduce global morale(reduced global morale for the ai apparently makes it a little more active, according to deanco and a few others in a post i read a long time ago in the archives), make use of all the reinforcement slots, and sometimes for fun put a reinforcing ai unit behind the player's units to simulate a flanking action.

[ May 28, 2002, 11:37 PM: Message edited by: russellmz ]

[WWB]

One thing I really disagree with doing, even if it helps the AI, is reversing the map sides. If you do so, you end up allowing people to withdraw forwards, making the command either useless or a cheat.

Regarding the points issue, it is a matter of taste. If you want to setup a battle so that the terrain held is the key factor, then use lots of flags. If you want losses to be determinant than use fewer and smaller. Large flags are worth 300 points, while small flags are 100. Units are worth essentially their purchase price in KO points (yes, I know not exactly, redwolf, but the factors effect both sides equally so they take care of themselves). Also remember that flags held can cancel each other out. One interesting way to do an ME is give both sides a rear area 'HQ' flag, which should be held, as well as some flags in the middle. That tends to encourage players to not rush everything forward.

WWB

[Olle Petersson]

Originally posted by Franko:

1. ... the computer doesnt know how to handle mounting, dismounting ... Is this true?

I agree with what the others have posted;

- The AI never embark.

- The AI disembark only if shot at (in enough quantity).

2. Don't design scenarios where the computer is allowed to setup, period.

With well defined setup zones the AI might be allowed some slack, as a means to create surprises for subsequent plays. The AI most definately have trouble with passive fortifications though, frequently massing all minefields in one corner or so...

3. Don't make the computer player be the "Attacker", because it ... lacks agression.

There are workarounds. One has to be careful with what units the AI control, the terrain and setup.

I agree that the AI have a tendancy to cancel hostilities too soon though.

4. ... whether setting an exit zone on one side or the other "encourages" the AI to behave a certain way.

It most certainly does. I tried using an exit zone for the AI in one scenario, and it raced all units at full speed up along one edge of the map.

In this particular scenario that is a very effective way to achieve the exit objective, but it won't wipe out the defenders, as is the primary objective.

5. ... AI to use on-board mortars effectively.

As noted; direct fire only. Will move mortars so that they do get a LOS.

Sometimes this is very effective, sometimes it's a waste of resources.

6. ... roads ...

7. ... AI commanders ...

8. ... target priorities ...

No comment.

9. ... AI is effected by limited visibility.

One general behaviour I've noted is that the AI never target suspected enemy locations, only known enemies. I further suspect that off map artillery is only directed at targets within LOS from the arty spotter. The result is that with poor LOS, as a result of terrain and/or weather, the AI will almost never use off map artillery.

10. ... the AI is set to "hide" at the beginning of the scenario.

Very unpredictable?

11. It appears that the human advantage can be mitigated somewhat by the use of harsher weather conditions and more static defenses on the AI side ...

Yes, provided one take all other aspects into consideration. (Like the arty comment from point 9 above.)

Cheers

Olle

[Ligur]

As stated a few times, IF you have a case where the AI is on the attack, place the tanks far away to the AI map edge or make them reinforcements. Drop a few scout vehicles (the AI will react when it encounters resistance, if not, it just mills around) on the front, a few moving guns to the back, and take most of the armor in as reinforcements.

Otherwise the AI rushes them madly to their horrible death. This way, it will actually support the infantry.

[SuperTed]

Frank,

Please email me tedquincey@yahoo.com.

[Lawyer]

The scenario designer must be permitted to plot the first moves of a scenario to get the AI going. Or perhaps certain units can be plotted to move as set by the designer, no matter what. This is necessary to get convoys moving and columns going where that is essential to the battle.

The AI can react as usual when the action starts. Mounted units will hit the ground and start fighting as normal, but the AI does not seem capable of doing the the moves that lead to action. All this business of trying to "coax" the AI to go down a road or attack is nonsense. Let the the designer do it.

This is particulary true when dealing with scenarios of actual battles and the initial actions of reinforcements arriving after the game has started.

Everything else can be set by the designer. Why not unit moves? A movement command should be available for each unit when designing scenarios.