ME + Large map +AI = bad idea?

[Sgt AA]

Hi!

Just tried to play a battle with small armed forces on an 6x3 km map (CMAK)- time was 60 turns. Both forces started at their edges of the map and in the middle we had 2 big flags. After 40 minutes I started to wonder where the AI was, so I had to go after him, found him close to the starting-point. He seems to hardly have moved at all and driven all his tank off the road in to some bad terrain in one big lump. When I showed up he seams to start to function, fight back and moved his things in a more logical way.

Does the AI need some sort of constant stimuli? Is it impossible for the AI to move his forces 10-15 minuts before he gets combat contact?

BTW - could it had make any differents if I had set the battle as an "AI probe/attack / assault" instead?

[JasonC]

The AI orients on spotted enemy and on flags. If there are none visible close to where its own units are, it has no reasonable way to pick one path rather than another - all seem equally unpromising.

String flags or readily detectible enemies along its path like breadcrumbs, and it will find its way better.

Note, however, that you will get pile ups along single routes if you only provide one line of these, and it has a long way to go. Which may be what you want for vehicle columns, but is seldom sensible with infantry.

To get the AI to orient infantry in a reasonable manner, it needs to start with a reasonable spread and to follow multiple promising routes to objectives. If you put only one flag near it, or only one covered approach to the nearest flag, it will pile up hopelessly.

Basically the AI is a path finding routine. And it doesn't weight "stay away from friendlies" nearly highly enough. n units all solving the same problem from basically similar starting points will all arrive at very similar answers.

[Sgt AA]

Thanks for your answer, it seems to be coherent with my experience. A little surprising that the AI "don't know" where the flags is, unless he didn't spot them! Maybee they have to made it that way unless the AI would have had knowleadge of my troops as well.

Perhaps a "big map" ME can work if it's triggered, e.g with a recon sharpshooter (without ammo) in sight of the flags.

It's a pitty that it wasn't programmed to, at least, follow and stick to the road until something appears.

Good, but sad, to have been aware of another weeknes of the AI.

[Slappy]

In my experience, the problem stems from having the AI weigh things happening close to it much more heavily than things happening far away. Generally this makes sense. Given that it can only choose one, you want an AI halftrack to take cover from the tank 200m away rather than the one 200m away. Also, you want the AI to seek out and try to take the flag 200m away rather than leave it alone to take the one 2000m away, at least generally. I think there is a sort of "inverse square law of AI response" that helps this happen and prevent the whole AI force from reacting to things on one corner of the map (a problem with Borg spotting). This generally works fine for maps with the size and flag and unit density that you genrally see in CM, but in novel situations the AI will definitely fail.

As an aside, I'm sure you've been attacking the AI and taken one flag only to see at least some of the defenders leave perfectly good prepared positions to wildly charge the taken flag. This is exactly what the distance response curve must have been implemented to overcome. The problem is, I doubt that the drop in response for distance is proportional to the size of map. It is likely a static number.

As another aside, you can see what sorts of issues this might present when trying to play CMC matchups against the AI (along with a host of other potential AI issues). In a world of 2km x 2km maps with exactly four flags, how will the AI react? Your guess is as good as mine.