[3.1+] NullBot: an adaptive skirmish AI

[aubergine]

NullBot lays siege to an enemy base...

Click for large view

[NoQ]

Oh, btw, the bot was on medium

This means that it didn't research rockets early enough, researching flamers simultaneously, but probably this didn't affect the outcome too much. Not sure though.

[effigy]

NullBot lays siege to an enemy base...

Click for large view

:O quite the variety of units

[NoQ]

Many outdated designs btw. Probably it actually needs to recycle things after all (:

[aubergine]

So, what I didn't tell you is that I built a huge wall around MR nullbot base to prevent it's units getting to the enemies, waited for the turtle nullbots to build themselves up, then demolished the walls to let the MR nullbot go play. There was much carnage

[NoQ]

Note: Turtle AI behaves much better on master builds on maps with well-put gateways. It is even able to defeat the NullBot classic/hard on this map.

Iluvalar: I'd really like to listen to your opinion on some more important issues. For example: How should the AI decide how many labs and how many factories or vtol factories it should use? How should the AI decide how much it focuses on fundamental research vs. weapon branch research? Any ideas?

[Arreon]

Perhaps you can add a text command that the player can use at the start of a game, like "ignore player 1". That way I can watch two of the AIs fight each other without worrying about my spectator position on some maps destroyed, especially with games involving VTOLs.

[NoQ]

a text command

I don't bother with AI chatting so far, because it won't be available in v3.1 JS engine anyway, and also #3199 (i will add chat support to the AI when it is run on master/3.2+ as soon as it starts working, but Per said that it's currently a lot of work to implement chat properly in network games) (hmm, but i think i can listen to messages already gonna try).

Probably porting this to 3.1+ instead is a good idea.

[aubergine]

Once the full set of chat features are working in the API I'm gonna go to town on rules.js

Could labels be a better way to define spectators in 3.1+?

[Iluvalar]

Note: Turtle AI behaves much better on master builds on maps with well-put gateways. It is even able to defeat the NullBot classic/hard on this map.

Iluvalar: I'd really like to listen to your opinion on some more important issues. For example: How should the AI decide how many labs and how many factories or vtol factories it should use? How should the AI decide how much it focuses on fundamental research vs. weapon branch research? Any ideas?

tank, vtol and cyborgs factory and defence are the same. They are millitary productor and you should treat it as the same entity in term of budget.

You end up with 3 big area of spending : Infrastructure, military, research. I think infrastructure is easily dealt with and i'll ignore it for now. So you have the remaining of your power that can be spent either in millitary or in research.

The main objective is to always have a tiny bit less military power than your opponent. Until you enter some "danger zone" were you absolutely need to react (he reached your base or some critical frontier/oil depot). This way you will always have a bit more budget for research and on long term, you'll gain technological advantage.

For focus : The standard mod have a very bad linear progression. Which mean that you are really encouraged to "square" all possible research instead of focusing on a single element asap.

[effigy]

So, what I didn't tell you is that I built a huge wall around MR nullbot base to prevent it's units getting to the enemies, waited for the turtle nullbots to build themselves up, then demolished the walls to let the MR nullbot go play. There was much carnage

I remember the first time someone did that to me with their AI teammate in a MP match. The horde was quite impressive

[NoQ]

v0.33.

Changes:

• Unit production and group size adaptation: don't produce much more units than the enemy has, and pack up the groups until its size is bigger than the biggest enemy unit cluster (not overdoing though ...); only on hard difficulty. Not sure it's worth it, but i think it didn't break too much things. That's not exactly what Iluvalar said (:
• Some AIs now pick propulsions based on weapon class. For example, MC AI uses halftracks for cannons and tracks for machineguns to keep the group speed more or less synchronized.
• MR and MC extra personalities now use adaptive research paths.
• Fix delays in early-game unit production (don't wait until a generator is up when low on power in early T1/no bases game)
• Fix broken scavengers in challenges included with the extra pack.

[NoQ]

v1.0 (:

• Fix: Don't produce the same truck multiple times (actually paying multiple times, but having only one truck at the end; this bug caused all those hangs without power when no generator was built yet; the AI became harder in early-game)
• A certain workaround for (and mentioned in) #3161 applied, so no more disappearing production queues! (the bug still remains, but it is much harder to reproduce) (note: the fix mentioned above is known to be unrelated to this).

[x1400]

Hi Guy's,
Sorry if this is a dumb question, but I'm new to WZ and loving it! How do I try out Nullbot? This AI sounds pretty cool. I downloaded the files, extracted them to /home/user/.warzone2100-3.1/mods/multiplay/skirmish, but when I start the game I don't see where to choose them as the AI in multiplayer or skirmish, only get Semperfi and Nexus. I tried re-compiling with those folders and files in place with no luck. What am I missing?

I'm using compiled 3.1 beta 4 on Linux Mint 12.

Thanks!

[aubergine]

@x1400: You don't need to extract the files, you can leave them in the .wz file. Put that .wz file in your mods/autoload folder and they'll get auto-loaded when the game starts.

@NoQ: Congrats on getting the 1.0 release!