No skirmish AI?

[Alexey K]

I'very played several QBs. Initially I thought that QB is analogue of "skirmish mode" of classical RST. However, behavior of AI units tells me that no AI of any sort is playing against me. Enemy units just execute prescripted plan.

Am I right in my conclusion?

[domfluff]

Sort of. There are three AI levels.

The "TacAI" is what tells your troops what to do on an individual level - taking cover in the 8m action spot, vaulting over walls, cowering from fire, using the right shells against tanks, running away.

The top level "strategic" AI is a scripted plan per-scenario, but it's more like "go here, at X speed and Y amount of caution". 

In between these, there's a level of AI to control units that have been grouped together in the editor - deciding how to move from A to B (fire and movement, etc.), using cover and adhering to the overall plan (e.g., if the plan was for this group to sprint in this direction, without regard for the enemy, then they'll try to do that.


There can be multiple AI plans, which can be weighted - this can cover everything from setup onwards, so you can set up a scenario that may change drastically on a second play through.

 

You can also set reinforcements on a random timer, so you may be unsure as to what opposition you're facing, even if you've played the scenario before.
 

 

The challenge with Quick Battles is that the map designer has no way of knowing what forces you're going to pick (or even which sides), so the plans have to be generic by design. With regular scenarios, they can be much more sophisticated.


The way this used to work (back in CMx1) was pretty simple - there were objectives on the map. If those objectives were friendly controlled, the AI would defend, if they were enemy controlled they would attack. There is still some merit to this approach (it reacts dynamically to your actions), but it's very predictable and not terribly sophisticated - the AI might lead an attack with mortar units, for example, which is entirely avoidable now.

Quick Battles are great, but they're probably best done with a human opponent, PBEM. They can be fun to knock around, but the AI is always going to be weaker than in a proper scenario or campaign.

[Alexey K]

Hmm... weird quiestion, but can I design and put my own AI in place?

Any kind of API for that kind of thigns?

[domfluff]

The scenario editor will let you design scenarios and write AI plans - it's the same tool used to make them in the first place.

It's pretty powerful (it's mostly *this* AI group goes *here*), but writing a good one is more difficult.

Probably the easiest way to get started would be to open a quick battle map in the editor, and then save it as a new scenario. You can then delete all of the existing AI plans and write your own.

Here's a decent guide, including triggers - it's for CMRT, but this should be identical to CMBS in it's current state.

http://www.youtube.com/watch?v=f02X2cygzFY&safe=active

[agusto]

The problem with QBs is that the AI plans must fit all sorts of different force selections and tactical situations. That makes them different from scenarios, where the AI plan is specifically tailored to one specific force and one specificn tactical situation. However if you want a challengeing QB, play a Meeting Engagement as Russians vs. Americans and manually pick the US force for the AI. Give the AI lots of M1A2s with APS, FOs and artillery, and it is going to be an interesting battle. The general superiority of the US equipment and the extremely fast american artillery strikes will compensate for the lack of a good AI plan.

Edited February 26, 2015 by agusto

[Alexey K]

I haven't yet seen AI plans mechanisc but I suspect that is not what I'am talking about.
I was talking about adaptive AI that is not tied to scenario designer prescriptions but in fact implements OODA loop.

Personally, I'd love to try and design one myself if CM engine permits plugging in 3rd party functionality.

Edited February 26, 2015 by Alexey K

[Stagler]

Alexey, you are not wrong at all.

 

The AI follows a prescripted plan. This needs to be upgraded. Just dont even ****ing get me started on how far this has to go to be considered modern. Thats all im saying.

 

Steve & all who argued against me in the Graphics thread - take note of another person concerned with the same issue.

Edited February 26, 2015 by Stagler

[agusto]

I haven't yet seen AI plans mechanisc but I suspect that is not what I'am talking about.

I was talking about adaptive AI that is not tied to scenario designer prescriptions but in fact implements OODA loop.

Personally, I'd love to try and design one myself if CM engine permits plugging in 3rd party functionality.

 

3rd party content is limited to textures, sounds, scenarios (within the limits of the built-in scenario design tool) and campaigns. CM is currently not capable of having a dynamic AI and you wont be able to implement one unless you get your hands on the source code.

Edited February 26, 2015 by agusto

[Alexey K]

3rd party content is limited to textures, sounds, scenarios (within the limits of the built-in scenario design tool) and campaigns. CM is currently not capable of having a dynamic AI and you wont be able to implement one unless you get your hands on the source code.

 

And I won't get source code because CM is not opensource project

 

Ok, I got it.

[A Canadian Cat]

Sort of. There are three AI levels.

The "TacAI"

<snip>

 

An excellent overview.  The only thing he did not touch on there is triggers.  That lets the scripted plan react to things on the battlefield.  Really this is not unlike what you and I do: We create a plan and start executing it and then make changes along the way to react to the enemy.  Clearly the current AI plans are not nearly as flexible as we are but they do reasonably well

 

 

Hmm... weird quiestion, but can I design and put my own AI in place?

Any kind of API for that kind of thigns?

 

I just checked there is a design guide for creating scenarios that was done for the Market Garden release.  The pdf ships with CMBS.  Look in the install directory for A Scen Design AAR PDF Book.pdf.

[A Canadian Cat]

And I won't get source code because CM is not opensource project

 

Ok, I got it.

 

Yeah, as a friend of mine pointed out Combat Mission is not really mod-able it is skin-able.  We cannot change the behaviour or add units or otherwise create new stuff what we can do is create new skins for the existing stuff.

[Alexey K]

Yeah, as a friend of mine pointed out Combat Mission is not really mod-able it is skin-able.  We cannot change the behaviour or add units or otherwise create new stuff what we can do is create new skins for the existing stuff.

Is it a technical limitation or Battlefront doesn't want anybody else to produce content for the game?

[womble]

Is it a technical limitation or Battlefront doesn't want anybody else to produce content for the game?

There may well be technical hurdles, but BFC wants to keep things historical, and in-house, so they aren't going to address those obstacles, even if they could be overcome.

[Stagler]

There may well be technical hurdles, but BFC wants to keep things historical, and in-house, so they aren't going to address those obstacles, even if they could be overcome.

 

Shame!

[sburke]

Is it a technical limitation or Battlefront doesn't want anybody else to produce content for the game?

You can produce content (scenarios, campaigns, maps etc). What you can't do is access something that doesn't exist (SOPs) or the base structural elements of the game. i.e you can't create some new type of buildings, vehicles etc. as noted above you can only skin them.

What the implications are for BF and whether it is a technical or business based decision is something only BF can answer.

[Michael Emrys]

What the implications are for BF and whether it is a technical or business based decision is something only BF can answer.

 

Which they have, at least partially. The reason so far is to maintain compatibility between different copies of the game. So that if you and I decided to play a game, I can be sure that you won't show up with a fleet of tanks armed with masers and tactical nukes. That sort of thing. There may be other factors involved in the decision, but that was one that got mentioned. (Well, sort of. The masers and nukes were my idea, because I figure you're that kind of guy.)

 

 

Michael

Edited February 27, 2015 by Michael Emrys

[Alexey K]

Actually a lot of games with h2h play are moddable.

They just compare hashes of game data to ensure that both copies have same mods.

[Stagler]

Actually a lot of games with h2h play are moddable.

They just compare hashes of game data to ensure that both copies have same mods.

 

This is true.

Mods aside, the ai logic should categorise units into armour, infantry, recon, and support. And basic algorithms to use each of those.

• Move forward with recon at the start - find the enemy
• Use its support assets on what it discovers - fix the enemy
• Attack with a combination of armour and infantry category units using cover to close with the enemy it discovers and occupy the objective areas. (we have seen that the tac AI will utilise trees and forest type terrain to move through as a preference).

[womble]

Unfortunately the AI as it stands falls at the first hurdle and doesn't have the foggiest idea what its troops can do. This starts at the force picker: who'd rationally pick immobile units (light AA guns) for a Meeting Engagement? Thats what one of the suggestions in a CMBN game offered recently.

[MikeyD]

I've used this analogy before. Learning the functions of the editor is like learning to drive a car. Don't expect to be competing in the Indianapolis 500 within the first five minutes. And if your first attempt at parallel parking didn't go well don't blame it on car's 'fatally flawed' design. Its perfectly simple to do once you've properly learned how to operate it. If you're lobbying to get access to the source code when you haven't even touched the game editor you're kind'a putting the cart before the horse.

[Alexey K]

I've used this analogy before. Learning the functions of the editor is like learning to drive a car. Don't expect to be competing in the Indianapolis 500 within the first five minutes. And if your first attempt at parallel parking didn't go well don't blame it on car's 'fatally flawed' design. Its perfectly simple to do once you've properly learned how to operate it. If you're lobbying to get access to the source code when you haven't even touched the game editor you're kind'a putting the cart before the horse.

I wasn't lobbying anything, just exploring what is possible.

I like game in it's present state so far

[Stagler]

Unfortunately the AI as it stands falls at the first hurdle and doesn't have the foggiest idea what its troops can do. This starts at the force picker: who'd rationally pick immobile units (light AA guns) for a Meeting Engagement? Thats what one of the suggestions in a CMBN game offered recently.

 

To do this, the units coded in the game would have to have some sort of category system I have outlined.

 

Maneouver units, reconnaissance units, support units - stationary units?

 

The algorithm would pick a ratio of which units to use whether it was attacker or defender. Say in a meeting engagement the algorithm divide spending its points on 20% on recce, 60% on maneouver units, 20% on support units.

 

The coding for that wouldn't be hard. It would simply be an addition of the following or something in each units code block. That's how I would code it in the VR3 engine anyway.

maneouver_unit = true;

support_unit = false;

recon_unit = true;

 

That unit would then be bought as either of those units.

If I had access to the code I could see how its done more clearly.

Edited March 1, 2015 by Stagler

[womble]

If you're lobbying to get access to the source code when you haven't even touched the game editor you're kind'a putting the cart before the horse.

Thought this was about the QB AI. There's no need for any automatic role-assignment in scenarios because the designer can do that, putting units in order groups that have orders that make sense for their type. The QB AI, though, needs a bunch of help in making sure the "right" units get assigned to the "right" order groups; static units being assigned to a "hold out of sight then advance on the objective" order being the most obvious case-in-point.

[Alexey K]

Thought this was about the QB AI. There's no need for any automatic role-assignment in scenarios because the designer can do that, putting units in order groups that have orders that make sense for their type. The QB AI, though, needs a bunch of help in making sure the "right" units get assigned to the "right" order groups; static units being assigned to a "hold out of sight then advance on the objective" order being the most obvious case-in-point.

 

Yes, I was talking about dynamic QB AI. Such AI need additional metadata on units to use them appropriately.

[Stagler]

Yes, I was talking about dynamic QB AI. Such AI need additional metadata on units to use them appropriately.

 

yes, the metadata I had in mind is in the post above.