TacAI in the attack: what it can('t) do?

[Chelco]

I am having a hard time designing good AI plans for the red forces.

It's hard enough to make good AI plans without triggers. Without triggers, the AI plans look a bit like orders for a turn in a WEGO game, where the "turn" lasts the whole scenario. The absence of triggers that could provide the designer with some sort of control during the scenario is to be alleviated by the TacAI, because it can make decisions on the fly. That's the theory.

In reality, I can't get the red TacAI to avoid these behaviors:

-Sending more and more troops through paths which are obviously dangerous. More and more troops are sent right into paths where piles of dead red troops accumulate.

-Not pulling out until it is just too late. It looks like the TacAI waits until a team is down to two or three men to order them out of there. This maybe a result of the high lethality and pace of the modern battlefield, but it would be more reasonable to make the TacAI to re-consider the path if just one man was shot (if one man was shot, it is likely very bad things will happen to the whole squad).

Maybe I am expecting from the TacAI something it can't do. I am not asking for a super smart TacAI. I just want to make an scenario that is believable.

Maybe I'm missing something here. Your feedback is appreciated.

[Guest]

Might be a good idea to cross-post this in the main forums.

I've had the same difficulties, incidentally.

[George MC]

I'd agree that having triggesr would be a great help. That aside you can still create good AI plans. A top tip is to play the scenario hotseat and see what you would do, that helps to get a feel for both sides and see what happens.

It is easier to create an AI defence plan, but keep it simple with some manouver units, whilst you have a large group who are pretty static. In this case I give em commands sporadically that have them hide, thise means that they can break contact for a while then latter on pop back up. You'llsee this work in Hammertime and in Al Amarah.

An attack is a bit more challenging - best bet here is either go for multiple attack plans and vary where the AI comes from, again play this hotseat and see what you would do. What behaviours you give the AI also affect how it attacks. Play around with the settings and you'll see which ones work.

Whatvere you do keep the plan initially simple then complicate it once you get the main 'concept' and timing correct.

Hope this helps.

Cheers fur noo

george

[SgtMuhammed]

Changing the size of the objectives for each order can also have interesting effects. If you scatter points about you can make a unit seem to patrol an area for a while. Then they will move to the next order.

When you expect contact I find that the best is to use assault if you want your units to engage and move. "Max Assault" units seem to forget manuevering and if you dismount infantry their vehicles will remain where the dismounted. This isn't as much of a problem with an Assault stance.

Like George said, play around. The system isn't perfect but it can give you some pretty good results.

[Thomm]

When you want the AI to occupy buildings, by all means only highlight squares that are withing buildings! This makes such a big difference, that I have difficulties winning Al Huqf now after I "tuned" the AI this way.

Best regards,

Thomm

[Chelco]

Thanks everybody. Good stuff.

I have been playing around with both Steel Beasts ProPE and Armed Assault's scenario editors and usually I can pull out decent behaviors for the OPFOR. Truth is, both editors feature triggers. Yet, even if you don't use those triggers, some very nice "built in" routines are there. For example: SBProPE bots will pull out or find cover if there is trouble. Other example: Armed Assault bots use covered/concealed avenues of approach if they detect a threat along their original path. All of these behaviors are dynamic, on the fly and in response to what happens in the game. No scripting necessary.

I cannot make CMSF's TacAI to do anything remotely similar to the above. CMSF's bots remind me of those electric football figurines, dialed in one direction which they will follow no matter what.

Anyway, I will keep trying. Your feedback is appreciated.

[MarkEzra]

Hi Chelco: You've gotten some good answers from guys who know. One thing that helped me was putting Scen I enjoyed in the editor and looked them over. How did the designer use his group selections. What AI settings, pauses. speed and deployment size was used. I really hope you'll continue to work with the editor New scen designers are ALWAYS welcome.

[Chelco]

Hi Mark,

That's a very good point. There are plenty of scenarios out there which are very well done and I'm going to see into.

BTW, Paper Tiger posted this very interesting point in the duplicate of this thread:

However, I have found that the two most important things for a good AI attack plan are stance and the length of time they have to perform an order.

Stance is obvious, 'cautious', 'normal', 'active' and 'assault' all make a huge difference to how the AI group reacts while following your order. I rarely use 'max assault' as that usually means that after making contact with the enemy the group will never move again. Yeah, it will EVENTUALLY, but assault does it better.

IMO the length of time the group has to perform it's order is the more important of the two though. Tell group 1 to move from point A to point B with only 5 minutes to 'Exit before' and it will perform the action very differently than if you give them 20 minutes. The units will use any terrain they can find between the two points and won't necessarily approach it in a straight line. The longer you give them to perform the order, the better, and more unpredictable, the results will be.

It makes sense. I'm looking forward to try this too.

[Pandur]

The units will use any terrain they can find between the two points and won't necessarily approach it in a straight line.

i would really like to know if this is still the case when pathfinding is totaly fixed one time

i guess they are partly, just zig-zag´ing(as you see vehicles move along the grid) through the map, thats why its so unpredictable.

dont know for sure, but for me it looks like this.

[Paper Tiger]

I first noticed this when I designed a simple AI plan for a Red force moving along a road into an ambush. There was nothing fancy at all about the plan, move from point to point asap and of course, it pretty much did that and got whacked!

So, to make the situation a bit more challenging, I decided to have the AI act a bit more cautiously. After all, it's moving at night into enemy held territory so it's not going to behave the way it did in the first example. So I gave the AI more time to perform it's order and changed it's stance. WOW! I got stuffed, almost my entire force wiped out because I was expecting the AI to do the same thing, only more slowly. Some units worked their way around under cover and came at me on my flanks. Only the stance and the timing changed, the waypoints remained the same and my units in the same set up. Two COMPLETELY different results.

You might be able to avoid too much of the bunching up behaviour by making their objective box even bigger. I'm having some very good results with this in another scenario under development. I had given the AI companiies very long, thin objective zones thinking that they'd space out and line up but that wasn't happening. This morning, I made the boxes even longer and much wider and they didn't bunch up so much.

In my campaign, the AI is doing most of the attacking (but not all!) so I have to make sure that there's at least 1 GOOD attack plan in there. I feel like I'm learning a lot while I'm doing this. And, it's fun to watch too.

[Secondbrooks]

Will AI hunt down opponents in their objective zones? Currently i'm wondering that how i can make AI's reserve forces come and help defenders with flashy counterattack. Basically i can aproach AI positions from any direction i like and engagment and spotting distances are just 40 meters so AI's reserves literally have to hunt me down, it's not enough for them to be somewhere nearby.

Problem is that defenders have 2 platoons in seperate defencepositions and i basically can take them both out one by one, when i want and in what order i want. AI should be able to act dynamically acording to situation, to make mission even somewhat tough.

[ January 30, 2008, 09:51 AM: Message edited by: Secondbrooks ]

[Paper Tiger]

Secondbrooks:

-Problem is that defenders have 2 platoons in seperate defence positions and i basically can take them both out one by one, when i want and in what order i want. AI should be able to act dynamically acording to situation, to make mission even somewhat tough-

Second problem first. That problem is down to the scenario designer. If he sets up two AI platoons in positions where they can't provide mutual support for each other then of course you can do that. So it's the designer's intention to put them in that situation. That might be a dodgy proposition in real life too.

Of course, you can try to anticipate what the player will do and script a counterattack. Once you get a feel for the map, it's possible to do that but it's still difficult to time it right. Sometimes it will work and people will think that you're a genius for making the AI do that and sometimes it won't. You can also script a withdrawal too.

I haven't done much in the way of AI defensive behaviour scripting but if I were designing a situation like that, I would have to consider the placement of AI forces at the start very carefully. You must try to maximise the resources you give the AI side, place them in positions to close down avenues of approach, set up hidden positions with ambush orders and force the attacker to approach the objective from the direction YOU want them to.

When I'm designing a scenario, the AI side is MY side and I want it to win. The kind of AI scripting you're looking for doesn't exist in the game yet and probably won't for a long time.

[Secondbrooks]

Yeah. I've tested i too, no luck this far with dynamic-behaviour. Well i have to skip that mission idea. System doesn't fit to forests at all, yet.

[SgtMuhammed]

Try using objective points at opposite ends of the area you want to attack. I set up some widely scattered waypoints for my guys and they sent units to every point and then everyone moved to the farthest point (from the last order). All with a single order.

[Paper Tiger]

They definitely don't hunt down enemy units in the objective zones. I was watching them this morning while I was playtesting a situation and the enemy overran the objective zone, which also happened to be a victory location. However, a couple of friendlies got isolated in a building and, as long as they kept their heads down, nobody bothered them. Being IN their objective zone doesn't seem to alter the way the AI units interact with the enemy. If they can't spot them, they won't do anything.

I think SgtMuhammed's method would do a good job of flushing the opposition out of buildings but as long as the human player's troops are in good order and hiding, they could still be missed.

[Secondbrooks]

SgtMuhammed: Sorry, i don't get this: 'Try using objective points at opposite ends of the area you want to attack'. You mean that i put objectives in corners of area i wish to attack?

I think i will have to make map more tunnel like, restrict player's movement to avenue of approach.

[Paper Tiger]

-I think i will have to make map more tunnel like, restrict player's movement to avenue of approach.-

That's not really necessary. You can get the same effect by leaving areas of the board without cover. And it will look better too.

[Secondbrooks]

Originally posted by Paper Tiger:

That's not really necessary. You can get the same effect by leaving areas of the board without cover. And it will look better too.

It would look like sacrilege... I just have to start battle from point when unit is already starting it's assault to objective. Most manuvering has already been done. Well, i quess i was dumb when trying to add earlier phases into it.

[Pandur]

when i designate a Tank Company as one AI group, and tell em to "Advance" somewhere. do they advance platoon whise or in chaos order!?

[SgtMuhammed]

If you give the entire company a single order it will be race day. Also be sure to give the objective enough space for the entire company or they will end up in all kinds of strange positions.

Secondbrooks: What I mean is, say you have a hill as one of your order points. If you put one objective point then all your units will go there. If you put two points then they will tend to send some to each point although most will initially head for the closest point. After a little bit though they will all tend to move towards the farthest point.

I found this out when I tried to set a recon screen. I set three points for the AI and expected them to move to those points and sit. Instead they moved to the points and then moved to the point that was farthest away from the previous move order. In the end they all ended up bunched around one point.

[Pandur]

If you give the entire company a single order it will be race day. Also be sure to give the objective enough space for the entire company or they will end up in all kinds of strange positions.

oh well, as far as i understood there are 8 AI groups possible.

i have 3 Tank companys and 2 mech inf Battalions(not all at full strenght). thats five already. than one for the ATGM group and one for the HQ´s and i have 7. i cant give them more accurate orders with 8 groups.

also, it would be a pain, so its race day than

[SgtMuhammed]

You can be kind of creative depending on what you are doing. I usually group all my HQ units and vehicles (Ash and trash as we used to call them) and units that aren't going to move in the plan into one group. Defending gives you a lot more options because you can set a unit to defend and then just put it out of the way if you don't want it involved. On attack you don't always have that option.

And yes, most designers have complained about the limited number of groups.

[Chelco]

Patch 1.06 has alleviated some of the problems I pointed out before. In the "test scenarios" I had previously won with just two clicks, I'm getting my rear handed to me by the AI.

[Chelco]

I’m still confused regarding the TacAI sending troops on covered/concealed paths as mentioned by PaperTiger.

Let’s suppose the TacAI is moving from A to B as in the image below. Let’s suppose a player’s unit is in area B with a good field of fire all the way through A, except in the green patches, which are trees. Yellow areas in the image are like the ones we see in the editor. The red arrows are the paths the TacAI uses to move the units from A to B.

In case 1, the TacAI sends all troops across the open into B.

In case 2, some units are sent through a covered/concealed route.

In case 3, some units are sent through a covered/concealed route (as in case 2) but just because the B area is wider.

I’ve seen the TacAI using paths like in case 1 and case 3, but never like in case 2.

Can the AI do something like case 2?

[Pandur]

two simple questions;

1. when reinforcements arive at turn/minute 5, that does mean when i want to move em by AI plan they will move no matter if i gave em "exit befor" 0030 or 0500 in their setup orders!? means, does time count from scenario start no matter if reinforcements or not!?

2. if i have a unit moving to "orders 2" and its at minute 10. order 2 has "exit befor" 0700 minutes. does that mean they will continue to "orders 3" as soon as the next turn after they arived at "orders 2" becouse they are too late already...!?

thanks