EU4 - Development Diary - 16th of October 2018

[mikesc]

Welcome all to today’s dev diary, my name is Michael, and I've been on the EU4 team as a programmer for a bit more than a year now.

Many of you have noted that the AI has been showing some undesirable behavior after recent updates, especially when it comes to defending their homeland and when choosing which forts to siege. So for the upcoming "Spain" update we spent some time on improving those aspects of the army AI. In this dev diary I will be explaining some of the changes we made. In addition to the improved hard-coded AI behavior, with the upcoming update modders will be able to influence some parts of the army AI with scripts. I'll end this dev diary with a short explanation/tutorial of this.


First I'll go over the changes and improvements made so far. Since we are still working on this update all of this is of course subject to change.

- The selection of armies to go on invasions (overseas) does now consider potential threats to the homeland and will be smaller if necessary. There still is the option that the AI will decide to invade with the full force if it thinks it cannot win otherwise. The calculation of the required strength has been overhauled and should be better in determining how many troops are actually needed.

- Region assignment of armies used to be done about once per year. There was a random factor involved so in some cases it could take even longer. In addition to the random reassignments it is now recalculated whenever the threat to the homeland changes considerably, i.e. there are sieges, large amounts of hostile troops moving towards homeland, etc..

- If there is a threat to the homeland these regions will have a much higher priority when assigning AI agents to regions. Even if not threatened, the homeland has higher priority in region assignment, if enough armies are available. On the province level (Within the assigned region) provinces close to the own country are now preferred when looking at what to siege/where to defend and what war goals to go for first.

- Armies running away from a superior force will still try to find a safe place, but provinces that require using military access or long walks will have lower priority, and if reasonably safe, the homeland receives much higher priority. AI armies are more likely to stand their ground, especially if they are clearly inferior and there is no real chance of winning the ongoing war anymore. Previously the player was often forced to chase them around the world thanks to easily available military access.


Scriptable army AI
A central part of the EU4 army AI is the evaluation of provinces, which determines where and when armies move to. There are already many defines you can change to tweak the outcome of this evaluation, but starting with the coming update you will also be able to change or even overwrite the evaluation score by using any trigger you like.
You will find new script files that have some basic functionality in commmon/ai_army, but you can add as many of your own as you like.

Let's look at a simple example:

province = {
    war = {
        active = {
            is_at_war = yes
        }
        eval_multiply = {
            modifier = {
                factor = 0.1
                is_in_capital_area = yes
            }
        }
    }
}

For now the top-level scope is always "province". We might add different options later. The second level is just a name you can choose freely and it serves to separate different modifications you want to apply. Just like event probabilities the starting value for these is 1. The interesting scopes are "active", "eval_multiply" and (not shown) "eval_add" and "eval_overwrite". With those you can control the modifications that will be applied to the evaluated province score. In this case it will be applied whenever the country is at war.
It is important to note that the evaluation of provinces happens very frequently for all armies in the game, so this "active" trigger will not be checked every time a province is evaluated (Which might happen hundreds of times per daily tick).
It will be evaluated when any of the following happens:
- Whenever you gain or lose ownership or control of a province
- Whenever a war starts or ends
There are three ways to modify the score, you can multiply it, add to it or overwrite it. The factor can be negative and < 1.0 as well so the inverse operations are also possible.
Note that armies prefer the province with the lowest score.

While you can influence armies a lot with this, there are still some hard-coded things that will always take precedence and obviously armies will still always adhere to the rules of the game (e.g. shattered retreat). It is a nice option to have however and I'm looking forward to seeing if and how it will be used in mods. Of course it is also possible to break AI behavior with this, so I'd recommend being careful, especially with "eval_overwrite".

To debug your script you can use the AI mapmode. In this mapmode you can see the evaluation scores for every province if you select an army. You can enable this mapmode by running this command in the console:

mapmode aieval

That’s all for today! Next week we will be looking at new features of the upcoming immersion pack. In the meantime, I’ll be watching this thread, so feel free to ask any questions you might have.

 

[Groogy]

Hmm, so how will the moddability of the AI affect rebels and such? There should be differences on how the different types of rebels "act". I'd expect separatist and pretender rebels to be the most organised and dangerous compared to say peasant tax dodgers.

Always been scriptable in the rebels own definition

 

[mikesc]

Scriptable AI sounds really cool! Is there any functionality on how AI will divvy up/organize its forces, or is it mainly a siege/movement/positioning thing?

For now it is just for movement/positioning, we might add more things in the future where it makes sense.

Shouldn't the AI be more courageous, especially against rebels or when having strong military bonuses?

The AI does know about the defensive bonus, but it will often err on the side of caution, though the way you describe it certainly makes no sense from an AI perspective. If you have a save where it's reproducible please do submit it. (It's difficult and very time-consuming to investigate these kinds of things without being able to reproduce it)

Would this make scripting something like "rebel" armies going out into neighboring countries from their home province possible? I am thinking of something like robber barons.

Rebels will still stay in the area defined in the rebel type, setting the area attribute to some invalid token and then scripting specifically rebel AI could work, have not tried it though.

- How do these changes (and others being made) affect the overall A.I's decisiveness? I may be wrong but it feels like these last patches at least the friendly A.I. has been, in analogous scenarios, noticeably less willing to commit to an action than before (e.g. whether to siege province A or chase army B, or be in region A or B), resulting in much walking around but not much getting accomplished (e.g. ottomans at war with mamluks and doing multiple laps of the arabian peninsula). Im curious about the results of the changes to region assignment of armies and how it relates to this.
- How does the focus on homeland defense affect the willingness of friendly A.I. to invade relatively distant foreign lands in a big war (e.g. france going after bohemia or poland in a protestant league war)? Will it be far less common now?

- If there is one evaluated province score that is so good that no other comes close there is less of a chance that armies will change their mind frequently, so these changes might affect this behavior.
- If the AI is in the war and considers it an important war goal it will still go long ways, but not at the expense of leaving the homeland undefended. It's not all or nothing though, just a balance change towards the homeland.

Hmm, so how will the moddability of the AI affect rebels and such? There should be differences on how the different types of rebels "act". I'd expect separatist and pretender rebels to be the most organised and dangerous compared to say peasant tax dodgers.

Things defined in the rebel types still apply and hard-coded evaluation for armies (including rebels) is still being done just as before. The scripting just allows you to work with the result of that and modify it.

@mikesc What is the one thing that you would like to improve in the AI in the future? (doesn't matter if its possible at the moment)

Many things obviously, sadly there is not enough time to do it all. One thing I would like to tackle is the indecisiveness of armies which leads to all kinds problems. Basically armies will look at the state of the world every day and then decide what to do. Changing conditions will lead to some action being preferable to what the army was trying to do previously and the army canceling whatever it is doing at the moment. There is a lot of code to handle specific cases of this. I would like a more general system that will make it possible for armies to stick to a task, consider already invested time/resources and only change it if it really makes sense.

 

[Caligula Caesar]

Sounds cool, looks like something fun to play with for modders! One thing I have noticed with the AI as it is currently is that when I have allies in a war, the allies will often suicide themselves attacking the enemy's strong stacks, even when my troops are stationed too far away to help them. It is almost as if the AI thinks I am also an AI and it can control my armies to coordinate attacks, only it can't and doesn't realise my armies aren't moving into position.