DyDo 2.2.2, AI for skirmish games and challenges

[kringled]

I emailed the one set of savegames to you. Dydo did also get to Inferno on that map but didn't deploy any actual Inferno units (orange did deploy thermite cyborgs, which is how I know they got to Inferno)

The 3-P challenge - I played that for a first time last night, and also just sent you a zip of the savegames. I really need to write the commands down for setting dydo general and level status; the two were set to standard and turtle AI respectively, both difficulty 3. They definitely gave me some trouble in the middle game there, and I took quite a few losses. It was also my first play of that map, so I didn't really understand the dynamics of the map. Some of my groups spent quite a bit of time scouting parts of the map out. I can give it a second run and see if I can do better now that I know where all the oil resources are, and how the enemy bases are set up. In this run, Dydo definitely got to and deployed Inferno units; I wonder if having one of the two being turtle helped make the other much tougher (since it profited from maximal research). They did get to Lancers, but just barely at the very end of the game; medium cannon a bit sooner. I was a bit starved for oil, as the vast majority of the game I only had the 8 oil resources, and gray actually beat me to the 3 resources closer to my base, so I had to fight for those.
As a side note: there are some odd graphical glitches to the map, blue splotches on the sides of buildings. I'll have to see if I can manipulate the screenshot I got.

Keith

[kringled]

Here are two more savegame sets on that 3 player map; in both I achieve victory just after the 30 minute mark with few or no losses. In one I use the same rocket/machinegun research path, the other I use a flamer/cannon path. You can tell that I get faster once I know the map well.
I want to start looking at the .slo and .vlo files for dydo in the next week or two to see if I can make more specific suggestions for improvement.

Keith

[Black Project]

Hello DD

I just got DyDo 2.1.3 working on Warzone 2120, but due to the dayworks that i have in the weekdays, i can only test in on Weekends (Saturday, Sunday)

BP

[Black Project]

I will post some of my experiencing with DyDo 2.1.3 playing with WZ2120 Mod.

This AI, as you said before, is much stronger than the previous version (2.1.2b).

In maps which DyDo get enough oil (Not massively to not say that i play in hugely oiled maps) the AI seems to build a lot of defenses and as the times passes, they still build more & more defenses and getting more advanced units (even more if they are allied with another DyDo AI), forcing me to find another ways to break through it's defenses, i saw too that they build FORTRESSES, making a real PITA (Pain In The Ass) to penetrate their base.

They don't attack so often as before, their units mostly guard their base (at least they wont give you a easy win in the first minutes) and depending on map, they build defenses in scattered positions of the map (close of it's base). When any enemy units gets close of their base, the AI send it's units to defend it's base.

Also, when starting in No Bases set (no matter what Tech Level i start, they build sooner some factories + a research center and cap some good amount of resources.

Then there it comes the part that let me really annoyed and sad.

They build power generator too late (they build one after more than 6-7 minutes of gameplay), at this time, they already started manufacturing units and researching stuff (in about 3-4 minutes), under the risk of running out of power too early and costly haveing no power to build a power generator (rarely, but this time happened).

Taking out of this annoying issue, this AI version is great, good job Dylan Dog, keep it up.

Regards BP

[zydonk]

Also, when starting in No Bases set (no matter what Tech Level i start, they build sooner some factories + a research center and cap some good amount of resources.

Then there it comes the part that let me really annoyed and sad.

They build power generator too late (they build one after more than 6-7 minutes of gameplay), at this time, they already started manufacturing units and researching stuff (in about 3-4 minutes), under the risk of running out of power too early and costly haveing no power to build a power generator (rarely, but this time happened).

Taking out of this annoying issue, this AI version is great, good job Dylan Dog, keep it up.

Regards BP

I've had the same prob with two 8 player maps, except in my experience the AI runs out of juice before a generator is built. Happens to all seven AIs. I'm surprised that building a generator is not near the top of the to-do list. There certainly should be a watch on power levels, so that this cannot happen.

Otherwise, DyDo is looking great.

[Tenoh]

Same problem here with 2.1.3. ai seems to build everything else fast but not power generators and runs out of power, i am not pleased.

[DylanDog]

I have been working on a new version of DyDo and I have not checked email, forums and website for a while.

In the new DyDo version (to be released for test in 7-10 days I hope) I fixed many issues with the script which generated few minors not crashing bugs (bugs which reduced DyDo strength though, yes I fixed the issue with the power generators too...).

This new DyDo version has a major change: I am introducing new research paths and increasing the number of research labs working at the same time.

The new research paths will be concentrated on few technologies only to allow to DyDo to quickly reach the next T level.

DyDo will research:

• PATH #1: All technologies
• PATH #2: Flamers and Rockets on T1, Lasers and TK/Scourge on T2-T3
• PATH #3: MG and cannons on T1, Lasers and Rail Guns on T2/T3

DyDo will mainly play with the path #2 or #3 and in such cases will deploy only templates with that weapons (with few exceptions on Cyborgs templates).

The first tests I am running are good as in 4 players game 2vs2 DyDo team which is using research paths #2 and #3 beats the team using the research path #0 as the first team develops its technologies quicker.

I still need to develop the part of the script which takes care to coordinate the decisions between teammates, as it make no sense that AIs in the same team use different paths. Actually I am running the tests by forcing the paths.

If you want to help me to test the new DyDo send me an email or a personal message and I will send you few DyDo packages with the various research paths already pre-configured.

Ideally I would like to have at least 3-4 testers for this version which could be seen like a "beta" one. The testers should play T1/T2/T3 games not only to search for bugs but also to suggest some improvements. You can test the new script by playing yourself or letting the AI play against themselves using the autogame feature (by typing "/ag on" on the console) and printing off all the logs messages in a log file (useful to search for bugs)...I can instruct on how to make this, it is very simple.

[Tenoh]

woot,good to hear that generator thing is getting fixed! hmm.. i doubt i would be great tester as i work most of the times.how ever i got few ideas that maybe be useful.I usually play with a partner ai of my choice vs all others.First of all many times ai just sits in its base and builds with out venturing outside to fight.If i give it a mark then it only sends 10 units and that's hardly a help.Maybe i could tell ai how many units to send... like all or big group or just a few?Another idea if ai gets low on energy,it should ask for help from its allies.Also being able to tell ai what and where to build would be lovely as it would give a strategic element to the game.

[Per]

@DylanDog: Sounds like an interesting approach. I've been thinking about how to do multiple research branches properly for semperfi as well, and there seems to be no perfect solutions, just better or worse hacks

One thing I have been considering is to run research in multiple "tracks" - one for general techs that all branches need, and one for branch-specific stuff. Typically that would mean one Reseach Lab is dedicated to each research track. Still unsure if this is a good approach, though.

Another issue is that sometimes the AI needs to focus on things it suddenly needs right now due to things the opponents does, best example probably being AA tech. This means it would have to reshuffle its priorities. I am not sure how to make it do that.

[Tucalipe]

> I've tried it out on Sk-Manhattan just for the sake of testing (I like that map, for some reason).

> As many have stated the AI's aren't building power gen. Out of the 7 AI's, only 4 had built power gens by 10 minutes into the game. The other three were stuck waiting for something to happen.
> Also, as pointed by Per, the AI's could be a little adaptive. For example, when I was battling WZ AI, and I started using VTOL's against their bunkers, they immediately sent a bunch of cyborg trucks to build a couple AA batteries. When I did that against a DyDo AI, it just rebuilt what he lost.
> At last, how exactly does the /ag command work? What was is supposed to do?

[DylanDog]

> I've tried it out on Sk-Manhattan just for the sake of testing (I like that map, for some reason).

> As many have stated the AI's aren't building power gen. Out of the 7 AI's, only 4 had built power gens by 10 minutes into the game. The other three were stuck waiting for something to happen.
> Also, as pointed by Per, the AI's could be a little adaptive. For example, when I was battling WZ AI, and I started using VTOL's against their bunkers, they immediately sent a bunch of cyborg trucks to build a couple AA batteries. When I did that against a DyDo AI, it just rebuilt what he lost.
> At last, how exactly does the /ag command work? What was is supposed to do?

Again, I have fixed the issue with the power generator in the new DyDo but I am still running test and the new AI is not 100% stable. DyDo already builds up AA defences around the attacked structure, maybe it has not built any because AA tech was not available or because of already too many AA in the base.

/ag is the AutoGame command, if you give this command then DyDo takes control of your player.

[DylanDog]

One thing I have been considering is to run research in multiple "tracks" - one for general techs that all branches need, and one for branch-specific stuff. Typically that would mean one Reseach Lab is dedicated to each research track. Still unsure if this is a good approach, though.

Another issue is that sometimes the AI needs to focus on things it suddenly needs right now due to things the opponents does, best example probably being AA tech. This means it would have to reshuffle its priorities. I am not sure how to make it do that.

I think is a complex theme. The issue as you said is not just to identify a path and pursue it but you also need some basics techs and you need to adopt the units production to the researched techs.

I think the AI should identify 3 main families of researches:

• essential/basic researches
• tech oriented researches (Sub branch should be VTOL researches)
• survivor researches (really do not know how to call them)

Basic Researches
Basic structures, main T1 Bodies, main T1/T2 Defences
These researches are always researched by the AI

Tech Oriented Researches
Ideally these researches should be grouped by technology (MG group, Flamers group, Cannon group...) and by weapon type (AR,AP,AT,AA). The AI chooses the researches path to pursue keeping the balancing on the weapon type (AP+AR,AP+AT,AT+AR). All the member of the same team should pursue the same paths, actually DyDo 2.2 chooses a team leader which advises the team-mates on the path to be taken (Human player can force AI in his team to pursue some research paths). The AI should build only templates which have weapons whose techs are pursued.

The VTOL researches are part of this branch but managed separately, the AI decides the % of research to dedicate to VTOL stuffs and should modify it when necessary. DyDo assigns around 20% to VTOL stuff at the beginning and this % is increased after some time if the VTOL main techs (factory and rearming pads) have not been reached or if any enemy is using already VTOLS.

Survivor Researches
These are researched to force when certain events happens like the research of AA stuffs if attacked by VTOLs, mortar researches (with CB tower) when attacked by mortars and power researches if that particular AI "needs" power.

Further comments...
The AI should really operate as team and should be able to calibrate the power spent for researches. Let me explain a bit...if he game is 1(Human)vs3(AI) then AI could allow to not only pursue the Basic and the Tech oriented paths but could also search survivor stuffs for example or increase the % given to the VTOL stuffs; or another option could be to keep one AI with only one ResLab working and using almost all money on units production. The team could also decide to spend power mainly on researches for a certain period of time to gain a strong advantage on the techs and only then run attacks (but the game would be boring I think...).

This is just the 22:00 O`clock "brain storming"...I would like to get to a more accurate and good planned approach.

About 80% of what I explained above is already in the DyDo 2.2 script. I am not sure this is the best approach but I am sure it is a better one then the "random research" or "research a bit of all" approaches. Also I just applied all this for General 1, the main AI personality.

[guciomir]

Hello,

What is the progress on Dydo? I haven't played the game for quite a while and i can see brilliant plans in your latest posts. I think a lot of players would love to play new version.

[DylanDog]

It is proceeding well also becasue Keith (I forgot his nickname) is helping a lot with the tests and is providing usefuls advices.
I think I will release DyDo 2.2.x soon. Apart from the bug fixing, DyDo 2.2.x is visibily stronger then the DyDo 2.1.x., it uses more power on researches and concentrate on few techs. I will release it soon but I want to be sure no bugs are present, would you mind to test it? if so send me your email and I will send it to you.

[NoQ]

I have just seen a weird scorpion bug. I'm not sure it's realted to the DyDo, but i've never seen anything like this without it.

A standard DyDo 2.1 ally was building some machine gun scorpion half-tracks, which itself sounds like a bug. But still when i click on it, i see "Heavy Machinegun" instead. Here is a screenshot: