AI rollcall

[AF]

Who here is currently developing AIs?

Aside from error, myself, kloot, hughperkins, cranphin? (Ive missed a few)

[Pendrokar]

AFAIK you are!


I crack myself up!!!

[LoidThanead]

I'm working on a Java AI. Still quite some ways from being usable though.

[Tobi]

me (The Lua AI C.R.A.I.G., that plays S44 and SWIW.)

(although very, very slow atm due to doing other stuff instead, but maintenance is definitely still happening )

[DJ]

I'm working on one. Nowhere near ready though:-

http://code.google.com/p/spring-djai/

[cranphin]

Well, for now more porting an AI, and as long as it doesn't do some amazing things compared to KAIK I don't see why anyone would want to use it, until/uneless we really polish up Java performance :)

It's a decent tool to work on Java performance tho, which was one of the thoughts, see if a Java AI is plausible :)

[Rafal99]

I have an unfinished AI on my hard drive which I was working on for some time, but haven't touched it for like 3 years (!)
I remember I stopped working on it at the same time when rotatable buildings were added to Spring, as support for them was the last feature I remember adding...
The funny part is that I still plan to continue it at some time, but it may as well never happen.

[SpliFF]

I'm working on several AI widgets which will become the basis of a Support AI (Group AI in Spring terminology) to help the player manage idle units. I'm currently focusing on automated scouts and engineers which work in squads under the control of a central army AI.

The AI will be modular so as long as you have the core widget you can enable and disable types of automation independently (ie, enable automated scouts but not automated builders).

The main purpose of this AI suite is to automate 'tactical' micro-management / high-CPS concerns (like repairing combat units, creeping artillery, jinking, scouting/harassment, building/repairing mex, etc). This will enable a new generation of games/mods where high-level strategic concerns form the core of the gameplay rather than manual unit control.

These AI widgets are comparable to existing helpers like 'Mex Upgrader' but with key difference being that my widgets share global data and utility functions with the core army AI widget and each other. This allows things like a shared cache and a common threat map that will improve automated behaviour in dangerous areas.

Because most mods lack complex high-level strategy (there are no missions in most games of BA/CA/XTA/etc) I'm concerned that these widgets may grossly unbalance standard MP mods or make them fairly dull (ie, BA relies heavily on manual CPS to determine player skill). For this reason it's my goal to only release these AI widgets for my upcoming Metalstorm mod (which will feature gameplay based around missions and diplomacy). This has the advantage that all players will have access to the same widgets, and certain automation tasks can be improved by using unit, cob or gadget scripts. Most importantly it allows me to sidestep the whole widgets=cheating debate since Metalstorm's gameplay expects extensive unit/squad automation rather than being an optional third-party afterthought.

Eventually I'll port these scripts to LuaAI to create a complete opponent AI but for now the widget interface is superior for testing and debugging.

[JohannesH]

Because most mods lack complex high-level strategy (there are no missions in most games of BA/CA/XTA/etc) I'm concerned that these widgets may grossly unbalance standard MP mods or make them fairly dull (ie, BA relies heavily on manual CPS to determine player skill).

Lol wtf

[SpliFF]

What I said makes sense if you understand the difference between tactics and strategy. TA is a largely tactical game since victory revolves around concepts of attacking and defending using combinations of troop types that will defeat the enemy troops. Positioning of units (high ground, correct range) is very important in tactics.

Wikipedia says: "Strategy is different from tactics. In military terms, tactics is concerned with the conduct of an engagement while strategy is concerned with how different engagements are linked. In other words, how a battle is fought is a matter of tactics: the terms that it is fought on and whether it should be fought at all is a matter of strategy."

Strategy is more about prioritizing goals, diplomacy, deceit, etc. My experience is that most RTS games are weak in this area as your time is largely consumed by tactical maneuvers. Something that seems simple (like a sneak attack using cloakers or capturing a base without damaging non-combat structures) cannot be implemented without a large amout of micromanagement and oversight.

Honestly, how often to you really see high-level strategies, attack synchronisation, coordination or diplomacy in a game of BA DSD? It doesn't happen very often because the "mission" is quite rigid and doesn't provide many strategic options beyond choosing your start position and deciding whether or not to porc.

[Neddie]

Modeling strategy is not very enjoyable, unless you're planning the actions of intelligent, unpredictable entities... in other words, actual groups of people. Even then, you're primarily watching things happen as a consequence of orders, rather than doing things in the moment - your hands are tied without access to the tactical or task level. This will seldom if ever, as a result, be enjoyable. Games are active entertainment. Without continuity of interaction, as a purely strategic game would exhibit, the experience becomes punctuated passive entertainment and will prove unable to attract and retain most people in most moods.

Even the mightiest warlord wants to get out there and crack some heads.

Strategy as a genre of games is centered upon the attempt to use all the options in the provided toolkit to achieve a provided goal... in Spring, for the most part this takes the form of complete annihilation of your foes or the negation and subjugation of your opponent. The vast majority of the activity occurs at a tactical or task level... as it does in any strategy game, from Chess on up.

[SpliFF]

Couldn't agree more. Last comment, since this is getting OT. My AI / Mod plan allows for manual tactics in every way spring supports. It only concerns itself with idle units: that being any unit with no current task or queue, that isn't set to 'hold position' or 'wait', and isn't currently selected or controlled by the player. Player orders always override AI orders unless the player specifically locks a unit for exclusive AI use (in which case player orders are ignored).

Anyway, should be interesting to see how a widget AI stacks up to LuaAI in terms of performance/ease of use.

[hoijui]

if i understand LuaAI right, then the only thing you had to do to create one, is activating all your AI widgets for the AI controlled teams, so could be 10 lines of code or so.

[JohannesH]

Theres deep level strategy in rts playing even if youre not skilled enough to see it. BA DSD a bit less because who takes that kind of game seriously. And yes bad unit control is a barrier of entry to many strategies, but doesnt mean that those strategies were always superior to the easier ones even with great execution.

And some things widgets can possibly automate, and be decent at it to an extent, but that doesnt mean that itd unbalance the game in the sense that bad players would suddenly have a chance against the good players. You dont need that great micro to win in BA, good strategy and multitasking is much more important.

but yeah bit OT maybe

[1v0ry_k1ng]

Modeling strategy is not very enjoyable, unless you're planning the actions of intelligent, unpredictable entities... in other words, actual groups of people. Even then, you're primarily watching things happen as a consequence of orders, rather than doing things in the moment - your hands are tied without access to the tactical or task level. This will seldom if ever, as a result, be enjoyable. Games are active entertainment. Without continuity of interaction, as a purely strategic game would exhibit, the experience becomes punctuated passive entertainment and will prove unable to attract and retain most people in most moods.

Even the mightiest warlord wants to get out there and crack some heads.

Strategy as a genre of games is centered upon the attempt to use all the options in the provided toolkit to achieve a provided goal... in Spring, for the most part this takes the form of complete annihilation of your foes or the negation and subjugation of your opponent. The vast majority of the activity occurs at a tactical or task level... as it does in any strategy game, from Chess on up.

actually im pretty down with stratergy

[initram]

I am part of a group at Aalborg University(Denmark), that is working on an AI using the new C++ interface. The project we are doing is in "Opponent" modeling", so we are currently using a Bayesian Network to try to classify the opponent.

The AI (called brAIn) is currently not available to the public, but we are working on replacing some commercial libraries, so that it can be used in the public domain.

As it is now brAIn plays a subset of BA, but so far it is able to beat RAI in most cases using only ROCOs and LLTs (RAI does not protect is commander very well).

[Petah]

I am working on BAI, currently about 50% complete and beats all other AIs the vast majority of the time.

[Error323]

I am part of a group at Aalborg University(Denmark), that is working on an AI using the new C++ interface. The project we are doing is in "Opponent" modeling", so we are currently using a Bayesian Network to try to classify the opponent.

The AI (called brAIn) is currently not available to the public, but we are working on replacing some commercial libraries, so that it can be used in the public domain.

As it is now brAIn plays a subset of BA, but so far it is able to beat RAI in most cases using only ROCOs and LLTs (RAI does not protect is commander very well).

That sounds interesting, can you post the bayesian network graph? Give some more information on the project? :)

[casperjeff]

I am working on a java AI called COCAI (Chain of Command).
I will be attempting to implement some very military techniques for asset/unit management (including a TO&E established by initial map analysis) and chain of command/orders mechanism for distributing efforts to groups of units.

Dunno if it will work out but I have a January 31st deadline to get something playable.

[zwzsg]

Who here is currently developing AIs?

Does having maintained and expanded KDR_11k's Kernel Panic Lua AI count?