Comparision of battle AI in campaigns and custom battles

[Duke John]

Since R:TW came out I have been mostly modding the game instead of playing it. Lately I have been paying more attention at the battlefield AI, particularly in Custom battles as they are easy to setup. I then noticed that the AI is using very bad tactics; charge everything (including missile troops). I do not recall that too much from the battles in a campaign. Others say that the AI uses better tactics when the generals have stars.

This gives me the thought that Custom battles may be flawed since the AI general never has stars and thus only uses bad tactics (charge).

So what are your experiences? Does the AI indeed put up a better fight when the general has lots of stars? Note that you should see past the bonuses, caused by the stars, on unitstats. So I'm not asking wether the AI fights longer, but does it skirmish, use archers, flank before it commits its combat troops?

[professorspatula]

Try testing your theory with the new battle editor. You can assign command stars to generals. Pick the flatest terrain you can find, give the two competing armies a few units and play the battle, then make a custom battle on the flat grassland map and use the same units and see what happens. I'd try it myself, but I'm too lazy!

[Duke John]

Didn't knew you could assign stars. Many thanks!

[Red Harvest]

I am not sure. It uses some awful tactics in custom, for certain even when it has a "no star" general unit added to the mix, so it could be star related. The default placement for archers is behind the main infantry line (WHY?). But even if it has just archers it charges them into javelin like ranges before halting and firing...much too close for their own good.

I've not been keeping careful track of the general stars in campaign, but I am finding when I face multiple AI generals (or captains, or generals & captains) that the AI general is still suicidal--all of them that is. The AI actually beat me once like this early in my Scythia campaign. I was sandwiched between two family member led armies and their 40+ man faction leader came charging down a steep slope into my center. I couldn't contain him with my 3 star general and single light cav. I had a unit or two of infantry, 2 HA's, and 2 archers, but I could only lend one infantry unit to help, and the general had rear support. It was the perfect opportunity to use that sort of rush...but a broken clock is right twice a day.

In other battles I've faced 3 family members and a captain led army. It led off the attacks on my Scythian skirmishers with its generals/captains, then followed with its family members. If it sees archers in front it seems to charge them with heavy cav, regardless of the consequences. That is similar to 1.1 in many cases. What I don't understand about the tactic is why not do it with all the heavy cav at once? Or hit with screening units first to chase the archers back, followed by main line with heavy cav hammering through weak points? Or with all heavy cav/family members *except* the commander.

[Duke John]

3 tests:

Non-scripted historical battle with 10 star generals
- AI moved 1 cavalry to his left flank since on his right flank was a steep ridge (thus guarded).
- Archers started to shoot as soon as my troops were within range (or so it seems). The same for Velites but of course with lesser range.
- AI General made a charge against my skirmishers and was crushed.
- My foresters decimated charging Roman medium cavarly. I did nothing the Romans just routed after a few seconds (yes with the 1.2 patch).
- The Roman line became a mess for reasons I do not see. My infantry haven't moved since the beginning, the only units moving were my cavalry who mirrored the AI cavalry.

Non-scripted historical battle with 0 star generals
Same result as It seems that skirmishers in front of your line causes the AI to change his battle line. Although it might also be possible that the AI starts thinking about unit match ups when within archer range. This is too late since it always ends up with a broken battle line which is disastrous for combat results.

My next test will be to see wether the AI performs better when all combat units are of the same class (heavy, light, etc).

Custom battle on a different map
I hope that the change in map doesn't interfere too much. The army composition was the same.
- This battle was chaotic, triarii were fed in pieces to my standing army.
- AI general made a glorious charge against skirmishers and the general was hold and routed by my Chosen Swordsmen.
- The velites kinda behaved as in the other 2 battles, but it was difficult to see since all units intermingled.
- The archers were constantly reforming, realigning and in essence doing nothing. At least they didn't charge my line. One thing that keeps coming back is the archers spending too much time getting in loose formation and busying themselves with reforming instead of shooting. Another thing is a kind of long column (or a line of archers turned 90 degrees) which starts shooting.

I'm not entirely sure since the custom battle had a different map, but at the moment it seems as if the custom AI is worse than the historical battle AI. But more tests are needed to make it 100%.

In all 3 cases the AI general made a suicide charge against my skirmishers and this definitely needs to looked at. I doubt that we will get a patch though, so it depends on modders.

Edit: some more results with the historical battle setup:
- The Roman line also goes from a nice checkerboard setup to all combat infantry lined up infront of missile units.
- In 1 test the AI only had Triarii as combat infantry (besides missile, cav) and it made one big line of infantry with cavalry infront. I guess this is because the lack of pilum infantry causes the AI to look for a different army formation (see Data\descr_formations_ai.txt)
- The infantry keeps doing its chaotic dance:
1. infantry moves forward
2. then moves back
3. it runs a bit then halts.
And use this dance randomly amongst all infantry. Result is a chaotic swirl of AI units that only enter combat by "accident". While 1 AI unit charges the others continue to dance.

I cannot see much difference when the stars are increased. Sometimes it appears as if more stars are better, but at other times the 10 star AI general behaves exactly the same as 0 star generals. If there is a difference then I think that the AI uses its archers better.