AI... Some consistency please?

So I know that people are often complaining about how the AI is bad. This thread is not like those threads. I don't care how it works, I just want it to always work the same. Or at least to know when it will do random things like what I'm about to show you.

Here is what brought up my question. I did an attack in war, and what my pekkas did stunned me. Unfortunately, I made this thread to late, and can no longer access the replay. But I made a bad quality representation of what happened.
Attachment 108416
There you go. It's quality is the future of HD. SO the dark pink rectangles are walls, pink circles are elixir storages, and the black box is dark elixir storage. A and B are my two pekkas, and the H.T is a hidden tesla. Now that you can understand this, let me explain what happened. Pekka A broke through that wall to attack the DE. The H.T popped up, and began attacking it. Pekka B begins attacking the wall. Pekka A destroys the dark storage. Then Pekka B stays at the wall. Pekka B goes through the attacking animation, does no damage, and begins walking towards the tesla. The walls I drew at the bottom might be too small, but you get the point.
My Question:
Why did the Pekka change targets? Why did it wait a second to change? Why did it go through the animation before deciding to go to the hidden tesla? And why, if troops will continue attacking the same wall after its target is destroyed, does this pekka decide that its target, probably the elixir storage, no longer mattered?

If anyone can answer ANY of these questions, I will be thankful. I'm not complaining about what it did, I just want to know why it did it. And if no one can explain this, then can someone in Supercell change this?


BTW
sorry if this is in the wrong subforum, I did not know where to post this, so I chose this one, because I put in the idea of a change. If this is the wrong place, please do not hunt me down and behead me, or make a comment about the placement of this thread. Thank you.

Completely understand what you mean and how the A.I. on some troops is ridiculously shotty(especially on Pekka). Though Ive always assumed pekka are so stupid is because they were meant to be that way. If you had Pekka with really good AI they'd be really OP, same with dragons.

Doesn't excuse the insanely stupid moments as you mentioned and demonstrated, but it's something we'll currently have to deal with, as I have no reasonable explanation and no decent solution, though some others on the forum might.

when I was a little th3, watching a replay of my defence,

i watched about 8 barbs run right past an elixer collector headed for an archer tower....they got about 2 squares past the collector then turned around and went back to it
of course at the time I was happy. the over shoot and return gave my defence more time to work, but if my troops did that I'd be seriously ♥♥♥♥ed off

I'd honestly have to see the replay, but I've noticed oddities as well, and been able to explain most of them to just bad luck with AI edge cases.

I don't think the 'stupid movements of <PEKKA, Dragons, King, Queen>' are specifically programed in. If anything, I see them as being glitches with what I call 'core AI behavior', as in the code used for almost all troops for common and basic things. Can think of it like thier 'common sense'.

Why would basic troops seem more or less fine, while big units seem to have moments of stupidity? Because the Core AI was made for barbarians, archers, and small things that are usually ran in large numbers. They spread out, they have a set attack radius, and so on. When you scale it up to say, a dragon, with far different attack ranges, a modified spread factor to make them 'act larger' and only able to run very few of them when compared to the number of barbarians, now you have problems. In the case of the dragons, the randomization factor to make them spread out now makes dragons act almost like a Plinko Game, and the village buildings are the pegs.

Being mission critical units, players keep an eye on them with more scrutiny. Looking at a mob of barbarians doing just fine, but with a few acting stupid isn't really noticed in the chaos, but the problem gets magnified when it happens to the few powerful units you run.

To a large extent, I've mentaly reverse engineered how most of the AI basicly works, and been able to see the problem, and yet again :facepalm: when I do, and kind of chuckle as a coping mechanism to the annoyance.

For the OP's PEKKAs, I'm still having issues mentally picturing it.

But it kind of sounds like the Teslas cause a re-target trigger on PEKKAs, which honestly, is the first time I've heard of troops reacting to things hurting them. Then again, when a H.T pops up, it does cause an area check to see if a re-target is needed. That is done since it would be silly to attack a building, and a closer H.T. comes up, and the troops don't react. Mostly that is for defence attacking troops.

The pause seems to be a common thing when there is a re-target, and can be seen everywhere when looking for them.

The game has a very simple events system, so what you see isn't what you always get. The attack animation is independent of the actual damage being delt. Damage is instantaneous, in one fixed sum depending on the source. It is not uncommon for something to be within a split second of dealing damage, before something interrupts it. This interruption can be a re-target trigger or out right unit death. Ask any Balloon user, you'll get an ear full of complaints of how those balloons fail to do damage when part way into the animation, with half of it consistenting of the actual bomb in free fall.

The obsessive wall attacking is symptom to a coding problem Supercell has yet to efficiently figure out. Like I said, there are re-target triggers. And when they are done, there is a pause, and then the troops act. Troops, even Wall Breakers, target buildings first. Then if needed they target whatever is in the way. If a wall is in the way, it then does a wall targeting function to find the best wall unit to attack. Closest wall is preferable, but one that can open up two or more closed areas, even better. After that, the location of the building they want to attack is set aside, and now targets that wall. They attack until the wall is destroyed, they are killed, or a re-target trigger is given. Thankfully, a re-target trigger is given if a near by wall is breached, and is useful, but if not, they will keep going at that wall, even if thier original target was destroyed in the meantime. This is a problem since where else do you put an effective re-target trigger? You can't spam it, or nothing will be done. And if done too much, you may have neurotic looking troops.

Anyway, sorry for the long post, but I hope it helps make sense of the frustrating troop behavior.