AI Use Of Cover

[Werezak]

I hope no one minds me asking a few things about the AI performance so close to release...

Recently I was watching this video here where the AI's forces pretty much got crushed, losing almost half their strength in the first 10 minutes while inflicting no losses on the player and gaining no ground. To Roach's credit, he did put a good deal of thought and care in the initial placement of his forces, which definitely went a long way to his success in that mission.

That said... the AI's placement of forces was kind of strange. Units were placed out in the open, and instead of making use of the somewhat-covered river valley to approach the objectives, the bulk of the AI's forces were sent forward over an open plain and got wiped out the moment they left the the treeline.

I'm curious how the AI evaluates the balance of good LOS vs safety of cover for a position on the map. Also how it chooses lines of advance when conducting an attack.

In another video here the AI seems to perform much better. While the player there also makes a whole lot of mistakes, one big difference is that the battle is set on a Finland map that is much more heavily forested, so that it is less likely for the AI to carelessly throw its units into the open, and units that do move into open areas are much less exposed.

[Veitikka]

I admit that I haven't watched all these videos. I've started watching the first series you posted, and I hope to finish watching it later.

So, quickly looking at the terrain in the first video, it looks like the AI chose the western river valley. It had low ground and was the shortest path to the objectives. Probably a mechanized/tank force, not an infantry force (the game has mechanized/armor/infantry 'force types' that have different AI templates), so it favored open terrain more than closed terrain. If it had chosen the Eastern approach then it would have had to go through the town of Mackenzell and open fields and hills after that. The player was waiting in the urban areas with a good field of view over the plains, waiting for the enemy to show its head. I don't see many surprises here...

The AI does analyze terrain cover, elevation and such. If the scenario designer wants the AI to use closed terrain to approach objectives then he should select the 'infantry' force type for the AI. Open terrain is usually seen optimal for tanks. Just like in any game, you must know how the game works to make good, balanced scenarios. The game offers some AI analysis tools for scenario designers.

Also, there was that river which was set to be impassable, I believe. It restricted AI movement quite a bit.

[Veitikka]

The AI's main deployment zone (I think very shallow compared to the number of units, overcrowding is expected) in south was far away from the eastern river valley. It was not even possible to get there without crossing the open hills. Whatever was allowed to start there were scout units.

[22sec]

What about in the defense? How does force type influence the AI’s deployment?

[Werezak]

[/quote]

The AI does analyze terrain cover, elevation and such. If the scenario designer wants the AI to use closed terrain to approach objectives then he should select the 'infantry' force type for the AI. Open terrain is usually seen optimal for tanks. Just like in any game, you must know how the game works to make good, balanced scenarios. The game offers some AI analysis tools for scenario designers.

This is absolutely true. I don't think anyone has ever made a good AI for a game without also carefully thinking about how to tailor the game mechanics, scenario/level design, etc. in a way that allows the AI to perform optimally. So it's important for a scenario maker to not create a battle where the AI is basically set up to fail.

I admit though, I really didn't understand why that battle was not a good setup for the AI to work well. At least until your explanation, I guess... however it's pretty clear to me that a Mechanized force (it was a Mech vs. Mech battle) is also really not well suited to open terrain - APCs/IFVs are extremely fragile, and not very good at attacking targets at a distance unless they have ATGMs or cannon (which some do, but many also do not). The APCs really should have been held back, or taken a different route.

Wouldn't it make more sense for the AI to apply these rules based on the type of individual formations, instead of the type of the entire force? Or forgive me if it does already do that.

I don't mean to criticize the game by pointing out an instance where the AI performs poorly, but I would like to discuss it. It does seem like creating a good scenario requires knowing more about exactly how the AI thinks and makes decisions in this game, beyond simply knowing the standard tactics and doctrines of the cold-war era (I don't consider myself an expert, but I do think I have a decent grasp, I hope [8|]).

Really curious about those AI analysis tools you mentioned.

[Veitikka]

ORIGINAL: 22sec

What about in the defense? How does force type influence the AI’s deployment?

I don't remember if it has any dramatic effect on the defensive plan.

Here's a chapter from the manual:

"The force type determines how many points can be spent on different equipment categories, and it sets the basic AI opponent behavior template. If the force type is ‘infantry’ then the AI usually advances with its infantry units dismounted in front, and tanks follow behind, supporting the infantry. The infantry AI favors covered terrain instead of open ground. Purchase points can be used freely if the ‘dynamic’ force type is selected. If the player has the dynamic force type, the opponent purchase points are calculated from the amount of points the player spends. However, if the generated battle is saved for later use, then the opponent side will not be playable, because the original player side has no AI template."

[Veitikka]

ORIGINAL: dholedays

I admit though, I really didn't understand why that battle was not a good setup for the AI to work well. At least until your explanation, I guess... however it's pretty clear to me that a Mechanized force (it was a Mech vs. Mech battle) is also really not well suited to open terrain - APCs/IFVs are extremely fragile, and not very good at attacking targets at a distance unless they have ATGMs or cannon (which some do, but many also do not). The APCs really should have been held back, or taken a different route.

Like I said I haven't watched the video carefully, but to me it looked like a mechanized force advancing on open terrain towards a well defended urban area. The defenders probably could use the long range weapons to kill the advancing units, without revealing their positions.

Wouldn't it make more sense for the AI to apply these rules based on the type of individual formations, instead of the type of the entire force? Or forgive me if it does already do that.

Tanks would take one path, mechanized units another path and foot infantry a third path? I don't think it can work like that. Some commentators have stressed that the attack should be concentrated in one place. But of course it depends on the situation, and there's an infinite number of different situations for the AI to deal with.

I don't mean to criticize the game by pointing out an instance where the AI performs poorly, but I would like to discuss it. It does seem like creating a good scenario requires knowing more about exactly how the AI thinks and makes decisions in this game, beyond simply knowing the standard tactics and doctrines of the cold-war era (I don't consider myself an expert, but I do think I have a decent grasp, I hope [8|]).

That's true. In many other wargames you need to know the scripting tools and use hours or more to design a scenario. In AB you can generate a battle like this in seconds. The more scenarios you play the better you'll know the game. Some videos were made by people who had played 2-3 battles before recording a preview video.

Really curious about those AI analysis tools you mentioned.

See my AI dev diary on this forum.

[schmolywar]

In most of these battles covered on twitch the soviet side would have fared much better if it used proper smoke cover. I dont see it though, which worries me.

[Veitikka]

ORIGINAL: schmolywar

In most of these battles covered on twitch the soviet side would have fared much better if it used proper smoke cover. I dont see it though, which worries me.

The AI opponent does use artillery smoke and smoke generators.

[Werezak]

For the sake of a counterpoint to the OP I'd like to bring up this video here. In this battle the AI does seem to perform better (though we don't get to see the whole battle).

One thing to note is the scale of the map and density of units is WAY different from any of the preview videos or streams I've seen so far (and probably more realistic IMO, I wonder if this is closer to how the game is meant to be played?). There's a lot more room for the AI to move around in, and it seems like it does a better job at not throwing its units into a death trap. We don't get to see the whole battle so I don't know that doesn't happen later, but I'd say this is probably the best showcase of the game that I've seen so far.

[Veitikka]

ORIGINAL: dholedays

For the sake of a counterpoint to the OP I'd like to bring up this video here. In this battle the AI does seem to perform better (though we don't get to see the whole battle).

One thing to note is the scale of the map and density of units is WAY different from any of the preview videos or streams I've seen so far (and probably more realistic IMO, I wonder if this is closer to how the game is meant to be played?). There's a lot more room for the AI to move around in, and it seems like it does a better job at not throwing its units into a death trap. We don't get to see the whole battle so I don't know that doesn't happen later, but I'd say this is probably the best showcase of the game that I've seen so far.

Unlike many other previewers, he's commanding formations instead of single units. The AI opponent does the same.

[Werezak]

Thanks for answering questions. I have one more if you don't mind: does the AI make use of the EFF/MAX range settings in the SOP, and does it make use of the different pathfinding settings (e.g. covered pathfinding)?

[schmolywar]

ORIGINAL: Veitikka

ORIGINAL: schmolywar

In most of these battles covered on twitch the soviet side would have fared much better if it used proper smoke cover. I dont see it though, which worries me.

The AI opponent does use artillery smoke and smoke generators.

I got to ask though, have the AI ever started using artillery smoke in the initial stage of an assault?

[nikolas93TS]

ORIGINAL: schmolywar

I got to ask though, have the AI ever started using artillery smoke in the initial stage of an assault?

It did and does.

[Werezak]

Tanks would take one path, mechanized units another path and foot infantry a third path? I don't think it can work like that. Some commentators have stressed that the attack should be concentrated in one place. But of course it depends on the situation, and there's an infinite number of different situations for the AI to deal with.

It must be possible for infantry to advance a long a covered path, tanks advance along an open path, while both are supporting the same coordinated attack... that seems to be what most players do. Maybe this is more of a discussion for after release though.

[Werezak]

ORIGINAL: nikolas93TS

ORIGINAL: schmolywar

I got to ask though, have the AI ever started using artillery smoke in the initial stage of an assault?

It did and does.

You can see the AI using smoke to try to cover it's attack in this video (skip to ~23:30). Pretty cool.

...I think I must have watched pretty much every preview video of this game on YouTube. It's not too late to join the beta is it? [8|]

[Veitikka]

I believe it's too late for Beta now...

I've been working on improving the AI smoke screen capability. The AI can deploy deceptive smoke screens on places where it's not attacking.