Skynet AI

[z3n]

Oh well, that was originally a bit of my AI but now I think he was trying some on his own now (Which is fantastic).

In fact, I think that descr_campaign_ai_db is solely his own now. I'm not really sure about the AI he wrote since I never tried it myself.


But I will say that I previously tried making large amounts of decisions like he did (to the point of my AI file reaching 30k lines or so at one point) and found it wasn't great for the AI. In my experience the most effective AI has a minimalist approach to the db, you want to limit its behavior to prevent it from making too many plans (since making it simple for the AI is best).

[Point Blank]

That makes sense, thanks. Might use some of his hidden AI traits, perhaps at a lesser level though. Look forward to trying Skynet this week

[MagusCaligula]

Thanks for making this free resource, AI was the most problematic thing for the mod I'm making, I'll just add all this and call it a day...

[MagusCaligula]

I have a question: how big is the influence of movement points on the AI? I'm asking because due to the design of my mod the ones you recommend are too much.

[z3n]

No problem, and influence of movement points isn't gigantic but it is there. Movement ability factors into the AI's plans in terms of how quickly it can move to get to another settlement (which is why in later games the base movement is often enough to hop from one town to another).

I don't think it will necessarily cripple the AI though.

[beermugcarl]

hey there, I've been playing EBII and have absolutely been loving the AI switcher, do you have a template for that?

I was trying to modify and use it myself but I couldn't find all the correct triggers, or at least I couldn't get it to work and so I assume I was doing something wrong.

but honestly, having used the AI switcher, I don't think I ever want to go back and so i want to try to adapt it to a few other mods.

[z3n]

The gist of it, is that the script swaps AI labels depending on what the situation of the campaign is.

What do you mean by template exactly?

[bitterhowl]

After some time I continued experiments with formations and seems that AI don't break initial formation if there's no additional BAI scripts added. So there's may be a possibility to make different formations for AI by some methods but without scripted AI behavior on battlefield.

[z3n]

I read this last night but was not able to write a longer reply until now.

Very interested in this. What methods do you refer to? Entirely avoiding scripts and focusing on descr_ai_formations?

[bitterhowl]

Sorry for late answer - I got no time for proper testing. Well yes, I can say I added new formation and added it into config_ai_battle. AI attacked me several times without changing formation or partially regrouping. My tests are not 100% valid because I changed too many things at once in config_ai_battle, so sometimes AI just stand still even when I attack, and all AI units act in battle separately. This behavior depended on friendly/enemy strenght ratio. At 1:1 AI was passive.

So first I got many questions about your new lines in config_ai_battle from EB 2.35 which I added into my file.

[z3n]

The only new lines that are relevant, are ones that are in the config_ai_battle files by default.

Those will be parsed by the exe. I experimented with the behavior a bit in that sense but there's no guarantees to getting the exact behavior you want. Honestly M2TW engine gimps the battle AI. There are some really stupid design decisions that I can only attribute to lack of time by the developers.

[bitterhowl]

I can't see files before monday (they're on my PC at work), I can restore some parameters which I wanted to ask about by memory. The section about attack plan (from front/left/right/back). I haven't seen them previously in other AI mods. Where did you find them

[z3n]

Those are actually based on a guess I made about them being in the exe based on some ollydbg debugging someone else posted about a few years ago.

I never knew for sure whether they worked as intended but generally even if they don't the worst they could do is not be parsed by the exe.


You are able to add any xml lines that exist in the config by default in vanilla and see an effect or two. Beware however adding them in specific places can cause crashes for example during sieges if the exe doesn't expect or can't find something it knows should be there.

[bitterhowl]

Ok, thanks! And what about that great feature with order executions delay for units on battlefield in EB 2.35? I liked it very much in G5 system too, but you did it with .xml!

[z3n]

Ok, thanks! And what about that great feature with order executions delay for units on battlefield in EB 2.35? I liked it very much in G5 system too, but you did it with .xml!

Well that was actually thru the pathfinding files from what I remember. Pretty sure It's the budget that changes that. It's been a few years but to modify pathfinding you need to delete the db for it to regenerate.

[bitterhowl]

What do you think about this stuff? http://www.twcenter.net/forums/showt...r-AI-(new-CAI)

I added it into my mod and I got awesome results. And things that guy says in his OP and answers looks like a revolution. Seems that thread doesn't get attention in community as it deserves.

[z3n]

What do you think about this stuff? http://www.twcenter.net/forums/showt...r-AI-(new-CAI)

I added it into my mod and I got awesome results. And things that guy says in his OP and answers looks like a revolution. Seems that thread doesn't get attention in community as it deserves.

I don't really agree with some of what he says but I liked his minimalist approach to the descr_campaign_ai_db, I adopted something similar in late 2017 / early 2018.

- AI trains the troops of good quality, generally better than in the majority of mods I’ve seen and I think even better than in the original EB2 (no offence to the original EB CAI coder) since the troops quality depends mainly on the tactics settings in file descr_campaign_ai_db – I don’t know whether the twc modders know that.

Troop quality does not mainly on tactics settings in the descr_campaign_ai_db. Instead what it really depends on is a) AI personality in descr_strat, b) recruitment_priority in the EDB, and c) upkeep & recruitment (financial costs). Invasion decisions like invade_buildup (recruit own troops) or invade_immediate (use mercenaries), don't really matter as much.

[bitterhowl]

Well, truly to say at this moment I use your pathfinding and config_ai_battle settings and his descr_campaign_ai_db. I included all good settings into descr_campaign_ai_db as sections SkynetAI, Beermugcarl, PiterAI, and some custom settings such as improved horde behavior from twc thread - and switch between them with link_ai_label. Now I got the best performance on strat level. I understand that custom AI may have their own settings in descr_diplomacy, but at this moment I'm satisfied.

[bitterhowl]

Now I got the .xml and want to clear some new lines. I searched .exe with Notepad and understood now how you find this new parameters. I'd be very pleased if you explain how do you suggest they working. I have only one goal, as Beermugcarl wrote in his thread - "I had a dream. a dream of competent AI! "

<resourcing>
                <!-- backup priority -->
                <melee-infantry>10</melee-infantry>
                <cavalry>0</cavalry>
                    <unit-suitability>
                        <!-- all missile units -->
                        <missile>0</missile>
                        
                        <missile-infantry>0</missile-infantry>
                        
                        <!-- all artillery units -->
                        <artillery>0</artillery>
                        
                        <melee-infantry>10</melee-infantry>
                        
                        <!-- all cavalry units (elephants, horses etc) -->
                        <cavalry>0</cavalry>
                        
                        <elephant>10</elephant>
                    </unit-suitability>
                    <!-- amount of threat a single unit can counter (controls how many units will get assigned) -->
                    <threat-per-unit>25</threat-per-unit>
            </resourcing>
            <assignment>
                <melee-infantry-lockout-distance>6400</melee-infantry-lockout-distance>
                <missile-infantry-lockout-distance>6400</missile-infantry-lockout-distance>
                <cavalry-lockout-distance>6400</cavalry-lockout-distance>
                <unit-suitability>
                    <!-- all missile units -->
                    <missile>0</missile>
                    
                    <missile-infantry>0</missile-infantry>
                    
                    <!-- all artillery units -->
                    <artillery>0</artillery>
                    
                    <melee-infantry>10</melee-infantry>
                    
                    <!-- all cavalry units (elephants, horses etc) -->
                    <cavalry>0</cavalry>
                    
                    <elephant>10</elephant>
                </unit-suitability>
            </assignment>

[z3n]

I apologize if anything I say after this is filled with disclaimers but it really has been years since I looked any of this.

Much of what I said before is relevant however -- look for whether the tag was in vanilla and if it was, then it's relevant. Much of what I was trying to do there from what I recall is prevent cavalry from being completely stupid as they normally are in vanilla M2TW. The point was to make them flank more, and fortunately I think I partly succeeded somewhere along the way.

In my opinion however, the most successful is scripting the AI to react to certain events. You may even have success with scripting.

Also read this thread, it answers one of your previous questions more in depth. I just rediscovered it
http://www.twcenter.net/forums/showt..._gta_plan_sets