[matt7785]

Something strange I just noticed about the AI, even though it always does this, how does it know to not try to send ground based attacks if there's not a path to your base, either from the bridges being out or just starting on seperate islands ? Is it something hardcoded that tells it to ignore triggers that use land units or is it just something in the rules file ? I don't really think it's based on the trigger weighting probabilities, because it knows not to use them before ever trying them and having them fail.

[EmpirioN]

Yep, done the same thinking myself a couple of times. And I haven't found anything from the ini files, that could have something to do with this.

It seems, that it is hardcoded. When the AI detects, that there is no ground passage to the enemy base, it does not spend money on forces it can't use. So it uses planes, etc...it even uses hover units, which is quite good for the RA2 AI.

[matt7785]

I looked around in the rules file and it looks like it might use the MovementZone and Locomotor tags to decide if the units are usable before accepting the trigger. What's not really clear is if all of the units in a taskforce need to be able to reach the target, either by using a transport in the taskforce or on their own, or if only some member of the taskforce needs to be able to reach the target. Also in the case of the taskforce with a transport I don't know if it checks if all the other units can fit before using the trigger, or just making sure a transport is present and taking what units it can.

It is actually a pretty good thing that the AI checks if it can reach its target, and I wouldn't try to change the MovementZone or Locomotor tags to get around it, but I think it would be possible to create better taskforces if the path availability logic was better understood. One idea to make good use of this logic would be to have a good number of teams with transports or aircraft, but set their trigger weighting probabilitys pretty low so that normally they aren't used that often, but when the other triggers are blocked they become much more active.

[Master_Fenring]

Actually, I think there's a coefficient that AI uses when choosing the triggers to use. If a team fails to reach its destination cell, the AI will begin choosing teams that WILL reach it or will sucessfully complete their missions. IIRC, if a team fails X amount of times, use something else.

[matt7785]

If your talking about the AITriggerFailureWeightDelta and AITriggerSuccessWeightDelta coefficients I don't think thats what it uses, or if it does it uses some other stuff too, because I thought those were only applied to triggers that had actually been used and had their teams built and failed in completing their script somewhere ? I'm interested in the times it knows it's going to fail before any of the teams are ever built, and what it uses to figure that out. After figuring that out it might add the AITriggerFailureWeightDelta to that trigger without the team ever being built, but if it does I think the AITriggerFailureWeightDelta is just a result of its path availability check, not the cause of the initial trigger not firing.

I think a way to test if the AITriggerFailureWeightDelta is applied to triggers where the teams are never built would be to block the AI's path by destroying the bridges to your base, then later fixing the bridges; If the AI then starts building alot of teams that were previously blocked then it probably wasn't applying the AITriggerFailureWeightDelta, but if it continues to use the teams that have been most successful when the path was blocked and only rarely sends a team that previously would've been blocked then the AITriggerFailureWeightDelta probably was being added to those triggers.

[Karan Sjet]

Deltas work, but you need Track like 20, and Deltas like -500,+5000,
then set trigger minimums low and maximums high, and it will work.

[matt7785]

That's not really what I was talking about, but it does still look useful. I thought the Deltas and the trigger probabilities just had to match up in relative size to each other, are you saying it's best to use large scale numbers instead of a small scale ? Also, why would you only set the AITriggerTrackRecordCoefficient to 20 ? Wouldn't it be best to set it close to the total number of triggers you have so that it keeps track of all of their current probabillities, instead of just using the default probabilities of the ones not in the cache ?

[Karan Sjet]

Large numbers- if AI sends something, you can be sure it wont send it again if failed, and that it will repeat effective tactics. Track- you can increase it to 999 if you want, but Im not sure if it will have effect (It might have limit of 100 or something.