Battle ai

**Lusted** · May 25, 2007, 05:51 AM

Im making a couple of threads about the ai for people like DrJambo and myself to have discussions about what can be done to improve the ai. This one will be about the battle ai. I'll post more later.

**DrJambo** · June 04, 2007, 05:29 AM

Never noticed this thread!

Anyway, for reference I've copied and pasted across the content of the post I made on the battle analyser:

I have something regarding the new 1.2 battle-analyser section of the config_ai_battle file that you might want to investigate for your mod.

Have you ever noticed that if you bring superior missiles to a fight against the AI, the AI will quite often sit out the duel using its inferior missiles, essentially allowing you to decimate the AI army? Well, in the battle-analyser section there's the following code:

Code:

        <battle-analyser>
			<!-- ratio of friendly to enemy strength to be considered more powerful -->
			<friendly-to-enemy-strength-ratio>0.8</friendly-to-enemy-strength-ratio>
			<!-- ratio of friendly to enemy melee strength to be considered overwhelmingly powerful -->
			<enemy-melee-strength-multiplier>3.0</enemy-melee-strength-multiplier>
			<!-- ratio of enemy to friendly ranged strength to be considered overwhelmingly powerful -->
			<friendly-ranged-strength-multiplier>3.0</friendly-ranged-strength-multiplier>
			<!-- ratio of friendly strength to enemy ranged strength to force attack -->
			<friendly-ranged-strength-divisor>5.0</friendly-ranged-strength-divisor>
			<!-- distance behind the defensive line the enemy must reach to be considered to have broken through -->
			<enemy-position-buffer-distance>25</enemy-position-buffer-distance>
        </battle-analyser>

This code was introduced in 1.2 and probably has a big impact on the passive AI. Now, as I see it and from the tests I've done there's seems to be an oddity here. I don't see how a friendly to enemy melee ratio and an enemy to friendly ranged ratio can both be greater than 0 and make any sense to the AI, when the end product is the same, i.e. for us (I think) to be perceived as being overwhelmingly powerful.

Anyway, in the tests I did where my missile heavy (4 missile units, 2 infantry units, 1 general) Byz army faced a Sicilian army of 2 Norman knights and 1 pavise crossbow (army ratio was 6:5 in my favour), the AI just sat and traded missiles until it was dead or routed. No attempt was made to engage the infantry and almost all Normans were dead. If the AI isn't going to retreat immediately this is clearly a daft tactic. So, I changed the

Code:

<friendly-ranged-strength-multiplier>3.0</friendly-ranged-strength-multiplier>

to

Code:

<friendly-ranged-strength-multiplier>0.3</friendly-ranged-strength-multiplier>

and this had a great impact. The AI appeared to trade a couple of rounds of missiles and then engaged its infantry. It still lost of course, but it played sensibly given the situation and took more casualties from me as a result.

Try it and see what you think. It's possible the last value could also be out of sorts but I haven't tested this one fully since changing the aforementioned one produced good results.

**DrJambo** · June 04, 2007, 05:55 AM

The other critical part I had suggested investigating in some thread over at the ORG was infantry skirmish units that have a small range, i.e. javelin, hand gunners and naffatun units. There's something not quite right about the way they're set up in the default battle_ai file and when the AI uses them, they more often than not get stuck in a seesawing motion between skirmishing and trying to get within range for shooting. I also found that a well placed unit of cav, i.e. just in front of your front line missile troops, can effectively render these units in the hands of the AI completely useless!

The questionable value seemed to be this entry:

Code:

<!-- don't skirmish until attackers are within this fraction of the missile range -->
	<range-factor>
	<moving>1.1</moving>

The fact this value is higher than 1 seems to be the root cause. As far as I can tell this means that if the unit is moving and attackers come within 1.1 x their range, they'll skirmish. If this is the case, then no wonder they never seem to get settled to fire. Now, I have made other changes like increasing their standard range, and increasing their "stop at this distance..." value, but the main change which seemed to help prevent this AI seesawing motion was setting this value to lower than 1 like so:

Code:

<!-- don't skirmish until attackers are within this fraction of the missile range -->
	<range-factor>
	<moving>0.8</moving>

The complete file is attached if you're interested. Like I said above, I also increased the default range of all javelin, hand gunners and naffatun. This is now 70.

**Lusted** · June 04, 2007, 06:18 AM

Those changes all look good to me. Have you had a look at any of the siege ai stuff?

**DrJambo** · June 04, 2007, 06:50 AM

A little yes.

I made it so the AI is less likely to retreat from defending around the breaches. Turning and running isn't always a very good plan and the AI always seemed a little over eager to abandon this stage. Plus in moving from breach defence to defending the streets the AI would face their units the wrong way, i.e. towards the plaza! This is especially bad for pikemen as you can imagine.

I also play with a slight modification of Klink's cool garrison script and so I increased the strength ratio at which the AI would sally. Sallying AI usually equals dead AI and getting the AI to sally is a relatively easy way to capture settlements.

In term of attacking siege AI I made some modifications to the formations file. Mainly trying to ensure the AI general unit is kept as far back from the main force as possible. This is most important for the "Wall Assault" formation which the AI is quite fond of using, so I added a special Generals block and placed it -50 behind the main bulk of units.

If you have a file compare utility you can have a look at these changes I've made and any others I might have made to the formations and config_ai_battle files, which I've attached.

Any questions, just ask.

**Klink** · June 04, 2007, 06:45 PM

I'm tickled now, what did you change about my garrison script?

This line is interesting:

Code:

<!-- controls the spread of the units along the wall as a percentage of unit width -->
<unit-spread>1.0</unit-spread>

This is a problem because the AI puts several units on walls with one-rank thickness right behind each other, which makes the missile units unable to shoot. Increasing this to 20 helped, but caused new problems when apparently the AI couldn't deploy on the smaller wall levels.

Also, this line:

Code:

<!-- reform to this amount times the tower range -->
<tower-range-scale>1.5</tower-range-scale>

That has to be 2.0 at least, buffer or no buffer.

I added a special Generals block and placed it -50 behind the main bulk of units.

Ha, I put at -120. Won't do to have it within range of anything.

Request, could you make a general's block for all the bridge crossing formations? I just easily killed three many-starred Mongol generals at a bridge because the terrain somehow put the general unit in front of all the others.

Another problem is when the AI is pushed sufficiently out of the breach defense formation it tries to reform, in spite of the hundreds of enemy soldiers hacking at it all the while. Maybe there's a formation tolerance there to mess with?

**DrJambo** · June 05, 2007, 06:19 AM