Is the TacAI still as bad as before 1.06? Seems so

[drtechno]

In the mission with saving the burning tank at night. I am able to drive my Strykers (not through the trees) but down the roads, passed the compound area which apparently contains dozens of enemy troops, and rendezvous with the M1 abrams. What irritates me, is that the troops in the compound area never make an attempt to move towards the objective (M1 abrams). So far, they just sit there. There is some effort on the part of the AI to make half hearted attacks against the disabled tank from other directions, but like half his troops are still just sitting around doing nothing.

[SgtMuhammed]

That is a function of design. What setting does the designer have those troops on? What are those guys supposed to be doing? If the designer intended them to hold those buildings then you wouldn't expect them to move about. Unlike the CMx1 TacAI which would immediatly launch a massive counter attack if you took an objective flag, the AI in CMSF relies on orders from the designer. While sometimes this can be annoying or give strange results it does allow the designer better control of the AI.

Not knowing the settings on the scenario it is difficult to say that anything is actually wrong. If the AI has no orders for a unit then it will sit and defend its location. Even without movement or attack orders, units can be set to different states. If they are set to hide then they will only attack if someone attacks them. Or they could be set for 50m ambush and will not fire unless troops come within that range. It is also possible that the troops moved in after your force went past.

There are some tweaks that many feel need to be made such as event triggers or general orders, attack, defend, etc., but most of the issues you see are design issues rather than programming problems. Perhaps the designer had a reason for the passive behavior you see or perhaps it is a genuine mistake. It would be nice if units had a little more independence but it is also nice that your carefully crafted defense doesn't turn into a bonzi charge when the first outpost falls.

[Guest]

Or... whatever issue is affecting the campaign is causing problems with the scenario. Troops way out of place? Maybe orders getting dropped?

Can you give the scenario another shot, drtechno? Established scens will likely have multiple AI plans. If you get the same results again I'd be more inclined to think it's an issue with the scenario and not with your playing some way the designer didn't intend.

[rune]

For that particular scenario, depends on which AI plan was selected. There are multiple things going on, and it varies by AI plan. And no, I won't give more information then that.

Rune

[Guest]

Horse's mouth, then, and all that. Nice to hear the scenarios aren't having problems. Seemed for a minute like the campaign bug might have some farther-reaching consequences...

[MikeyD]

Unfortunately, some of the earliest-design scenarios were the victim of a combination of green scenario designers using a Beta of CMSF v1.0! AI orders seem to sometimes be given less for tactical reasons and more out of game engine concerns. BFC included heavily reworked QB maps with the v1.05 patch in an effort to better match-up the AI orders with the improving game engine. I'm looking forward to the Marine module largely because the scenario builders will have had an extra year of practice!.

[Edited] D'oh! others posted while I was typing. I forgot Campaign's now got an issue with out-of-position units. My apologies for defaming any of the scenario designers out there.

[YankeeDog]

I think the OP is really talking about the StratAI, not the TacAI.

As I am familiar with the terms, "TacAI" refers to the AI routines that make tactical choices for individual units on the micro scale -- pathfinding to a specific location, whether to fire, who to fire upon, what kind of ammo to use, etc.

the "StratAI" is the part of the AI that makes up and executes a plan of attack, where to move units before contact is made, who to react in a general sense once contact is made, etc.

As already noted, the StratAI is not much more in the hands of the scenario designer. So if the StratAI doesn't seem to be doing it's job well, first place to look is probably the scenario design. However, I have heard some criticisms that the Scenario Builder lacks some flexibility in terms of what the designer can tell a computer-controlled to do, and how to do it.

Regards,

YD

[SgtMuhammed]

Right now the major problem with the AI editor is that it is all time driven. You can have the perfect plan worked out but if the player does something different then there is no way to change it on the fly.

[Roter Stern]

Originally posted by SgtMuhammed:

Right now the major problem with the AI editor is that it is all time driven. You can have the perfect plan worked out but if the player does something different then there is no way to change it on the fly.

That’s exactly it – it’s not that the game AI is stupid, but rather that the scenario designers lack the tools to make it react to the player’s actions. Basically the AI “pendulum” swung from a CMx1’s highly reactive AI with almost no concept of a global plan, to the opposite end of what we see right now in CMx2 where the AI is able to execute complex plans but is unable to show any initiative and deviate from the set plan at the right time.

Idealy the AI Plans should be event driven, as opposed to timing. In other words each order should have a condition-check for the AI group to either advance to the next order or remain on their current.

Alternatively, and possibly easier to execute, this could be solved with a tool which will allow the scenario designer to define a number of conditions upon which the AI would switch between different plans. So instead of randomly choosing between the 4-5 plans at the start of the battle, the AI could attempt to adapt to the player’s strategy.

[ February 07, 2008, 09:14 AM: Message edited by: The Louch ]

[Battlefront.com]

YankeeDog is correct that this is a StratAI issue, not a TacAI one. Just think of it this way... tactical decisions, that are limited to a single unit, are TacAI. Anything that involves more than a single unit is StratAI. There's a little bit of AI inbetween that is specific to coordination of elements within a larger context, but it's better to just think TacAI and StratAI.

Yes, we do need to expand the pallet of StratAI choices in the Editor. Event (aka "trigger") driven features are what we'll be concentrating on. We'll be getting to that sooner rather than later.

Steve

[Vergeltungswaffe]

Well then, we're already good to go for Russians 41-43.

"I'm hit"

"Stay on target"

"But I..."

"Stay on target"

"Aaaaaaaaaahhhhhhhhhhh!"

Future upgrades will take care of everyone else.

[Roter Stern]

Yes, we do need to expand the pallet of StratAI choices in the Editor. Event (aka "trigger") driven features are what we'll be concentrating on. We'll be getting to that sooner rather than later.

Great to hear that it’s on the to-do list! I wonder if I should hold my breath for a LUA-type AI scripting with branching conditions and endless nested If-Then statements?

One suggestion I haven’t seen mentioned yet, for when triggers are implemented into AI plans, is for each AI group to have a “Break off engagement upon…” condition. Mostly useful for an offensive AI plan, since as is stands right now it’s rather hard to force an AI to preserve it’s troops – once it commits to an assault it doesn’t seem to stop if its troops are being decimated.

Actual conditions could range from simple casualty/status/ammo levels to perhaps something as complex as evaluating the defending enemy’s strength and composition. It would also be nice if we could set an area to for unit to retreat into, just to help the AI figure out where the safe-zone is.

[drtechno]

I agree with you guys that there may be multiple things going on with the StratAI, but if the sole objective (as judged by the points distribution at the end of the scenario) is to take out the tank, then I am not exactly sure what the AI gains by hanging onto a sector that is uncontested and gives it no victory points.

Its the Baker one scenario that comes with the game, so anyone can try it out.

[Battlefront.com]

Conditional triggers are also on he list. I would also like to see the Plans be linkable, so if Plan A fails the AI can switch to using Plan B, but if successful go onto Plan C. This would reduce the variety of overall strategies, however it would increase the flexibility of the one being used.

Drtechno, the answer is that the scenario designer made it that way This could be accidental (i.e. made a mistake in the Plan somewhere) or on purpose just in case the player tries something like a flank attack. Whatever the case my be, the StratAI has *no* ability to act on its own initiative. So if a group of units has no orders to move out, they will not move out.

Steve

[rune]

And you aren't supposed to know why. It is the reason there are 3 AI plans in that scenario, and I just confirmed, none of them stop in that area.

Rune

[Omenowl]

Conditional triggers would be fantastic. Also if at any point you could get Digital elevation models to be imported so we don't have to do our own topography this would be be fantastic.

[Paper Tiger]

I really like the idea that the AI could switch plans mid-game too. Since the game keeps casualty figures as percentages, it would be nice if it were tied to a plan as a condition, ie, if casualties exceed 50% stop attacking. However, I can foresee how this might cause problems occassionally when the defender is on the verge of collapse and the AI attacker hits the magic number and switches to defensive mode instead. But that happens in real life too.

[Topo]

Conditional triggers (and eventually WWII) at this point can lead CMSF from a very, very good game/sim to a fantastic, incredible game/sim IMHO.

Saluti

[Hev]

Import digital elevation models, is that even possable?

If it is how cool would that be, you could have a basic map in seconds!

[Redwolf]

That's not the TacAI, BTW. It's the programmed opponent, or strategic AI or whatever you call it.

[Darkmath]

Basically the AI “pendulum” swung from a CMx1’s highly reactive AI with almost no concept of a global plan, to the opposite end of what we see right now in CMx2 where the AI is able to execute complex plans but is unable to show any initiative and deviate from the set plan at the right time.

Bump for this thread as the TACAI works properly now (for me at least ); it's time to deal with StratAI. IMHO, TheLouch highlighted an important point.

The time based AI in CMSF makes some people complaining about a false AI (everything should be specified by the designer) ; just do what the AI hasn't been instructed to and the AI would do nothing, killing the gameplay and game replayability.

However, there is still a factor of unpredictability ;the AI could randomly choose between 5 plans at most.

Programming 5 plans is of course a pain for the designer, but is it worth the effort?

-Under which circumstances is randomly chosen complex AI plans significantly better than previous CMX1's AI?

Defense seems to be suited for this kind of design but does it still remain ways for the player to get the AI inactive?

[Darkmath]

I should have added that I observed an annoying problem with the AI in WEGO in original scenarios :

within a 60 seconds turn, units move until 30 seconds are elapsed then stop.

30 seconds staying at the same point.

Seems that the AI work according to a 30 seconds turn.

[hoolaman]

I saw this too Darkmath in the 1.02 demo, but haven't noticed it since then. Still happening?

[Darkmath]

I didn't put a lot of effort in checking this.

But I think it happens only for the original maps ; it may depend on stance used by an AI group.