Joshua

Come in here to talk about your sky-net style world-destroying super bots!

Moderator: Defcon moderators

User avatar
martin
level5
level5
Posts: 3210
Joined: Fri Nov 19, 2004 8:37 pm
Location: ::1
Contact:

Joshua

Postby martin » Mon Jan 19, 2009 6:16 pm

Well, since we're derailing the main thread, I thought I'd better start a new thread for my bot, called Joshua becuase I couldn't think of any other incredibly non subtle movie references :roll:
So, at the moment the bot does nothing, however I'm going to base it on the design I partly implemented in my C++ bot before getting bored.

Defcon 5) Survey the globe, to account for different maps, the bot surveys the globe, basically moving through every point on the globe querying territory/ocean/land etc. Once this is done, do nothing
Defcon 4) Place units, you can see enemy buildings during defcon 4, so place radars first and if the enemy were stupid enough place in defcon5 you get an advantage over them. I don't know how I'll do the actual placement of buildings, probably broadly a star all the time, however I'm open to other suggestions.
Defcon 3... do stuff, I'm not much of a defcon player, so I'll need suggestions, or to make it up as I go along here

Advantages of AIs:
    They know stuff exactly, so for example they can fly planes just outside the range of your RADARs/Silos (once they've found them, of course).

    They can micromanage, for example they can have constant fighter patrols around their territory, just outside enemy silo ranges and always making sure the planes land before they crash (this also applies to carriers ofc)

    They can look at large data sets, for example if you sent in a few fighters which didn't see the enemy silo, you could probably pretty accurately work out its position from time for missiles to appear/direction of missiles etc.


Despite what I said in the other thread about real time tactics not beating strategy, I think a combination of these advantages, with good (if not excellent) strategy will be a reliable winner (prepares to eat his words in a few months time)
GENERATION 22:The first time you see this, copy it into your sig on any forum and add 1 to the generation. Social experiment.
User avatar
tllotpfkamvpe
level5
level5
Posts: 1768
Joined: Fri Aug 31, 2007 12:04 am
Contact:

Postby tllotpfkamvpe » Mon Jan 19, 2009 6:22 pm

This bot made properley, will be almost unbeatable :shock:
User avatar
martin
level5
level5
Posts: 3210
Joined: Fri Nov 19, 2004 8:37 pm
Location: ::1
Contact:

Postby martin » Mon Jan 19, 2009 6:27 pm

You really think so, all those micromanagement things are on the easy end of the scale of complexity for programming a bot. Personally I thought they wouldn't have much impact on the game except making it a little harder to beat - then again I'm not much of a defcon player.
I guess I'd better stop writing interpreters for esoteric languages, and write this bot :P
Last edited by martin on Mon Jan 19, 2009 6:29 pm, edited 1 time in total.
GENERATION 22:The first time you see this, copy it into your sig on any forum and add 1 to the generation. Social experiment.
User avatar
rus|Mike
level5
level5
Posts: 2750
Joined: Thu Jun 28, 2007 5:52 pm
Location: Russia, St. Petersburg

Re: Joshua

Postby rus|Mike » Mon Jan 19, 2009 6:29 pm

martin wrote:Defcon 5) Survey the globe, to account for different maps, the bot surveys the globe, basically moving through every point on the globe querying territory/ocean/land etc. Once this is done, do nothing
Defcon 4) Place units, you can see enemy buildings during defcon 4, so place radars first and if the enemy were stupid enough place in defcon5 you get an advantage over them. I don't know how I'll do the actual placement of buildings, probably broadly a star all the time, however I'm open to other suggestions.

Sometimes placing some ships right at Defcon 5 is good. For example, being Africa against Europe, you can place a carrier in the most northern possible place of your waters and head it to Iceland. This way you both have a chance to scout Iceland and have a very good info on enemy's fleet at Defcon 4.

For protecting from such things it's vice to place some battleships in the most close to you parts of enemy waters so that he can't get scout ships deep in your waters (thus making you huge troubles with placing your fleet).

Tell me in case such deep aspects of gameplay are not required so far ;)

EDIT:
About micromanagement. Micro is EVERYTHING if you look on any situation from a tactical angle (ofcoirse, better strategy will beat better micromanagement). Micro should be bot's main and ultimate power because CPU can be at it times and times better than any human (instant reaction, any number of moves at the same time, etc).
User avatar
martin
level5
level5
Posts: 3210
Joined: Fri Nov 19, 2004 8:37 pm
Location: ::1
Contact:

Postby martin » Mon Jan 19, 2009 6:34 pm

That's a good point, Hmm. Ok, fleets should be deployed in defcon 5 then?

And it may seem somewhat deep, but I need to work out what tasks I'm going to do in the setup stage of the game, and how exactly I'm going to do these tasks, before coding anything interesting :P

Also, a question to any programmers out there who know more about these things, how hard would it be to write a version of the API for C# - I far prefer the language, and it would make writing this more fun (visual studio + C# > Notepad + Lua) :lol:
GENERATION 22:The first time you see this, copy it into your sig on any forum and add 1 to the generation. Social experiment.
User avatar
rus|Mike
level5
level5
Posts: 2750
Joined: Thu Jun 28, 2007 5:52 pm
Location: Russia, St. Petersburg

Postby rus|Mike » Mon Jan 19, 2009 6:49 pm

martin wrote:That's a good point, Hmm. Ok, fleets should be deployed in defcon 5 then?

That depends of the setup.

Close setup:
Very few ships right at Defcon 5, all the rest based on recon data at Defcon 4.

Far setup:
All at Defcon 5 (because there will be no recon data at all at Defcon 4, no reason to wait). This way the ships will get into desired positions faster.
User avatar
martin
level5
level5
Posts: 3210
Joined: Fri Nov 19, 2004 8:37 pm
Location: ::1
Contact:

Postby martin » Mon Jan 19, 2009 6:56 pm

That's a good point. Don't forget, that ships will all be deployed in fleets of one by Joshua, since he can manually micromanage fleets and will get a greater flexibility this way :)
GENERATION 22:The first time you see this, copy it into your sig on any forum and add 1 to the generation. Social experiment.
User avatar
Ace Rimmer
level5
level5
Posts: 10803
Joined: Thu Dec 07, 2006 9:46 pm
Location: The Multiverse

Postby Ace Rimmer » Mon Jan 19, 2009 7:05 pm

Well, in talking scout ships, that only matters if the bot is involved in 1v1, 2v2, or perhaps 3v3. If you're talking six player free for all, scout ships are a waste. Probably in most diplomacy settings as well.

As for micromanagement, I agree with rus|Mike. That should be the bot's greatest tool as it can do it everywhere at the same time no matter the actual speed of the game and with every unit. Trying to overcome that would be a challenge in and of itself. In vs games, the only way I could see defeating a micromanaging bot would be in it's lack of intuition. Or, to put it in other terms, as rus|Mike said; good strategy can overcome excellent micromanagement, but not without effort.

There should be other tactics as well, like using single ship fleets. I think the bot should only ever use single ship fleets as that's another aspect of micromanagement where it can potentially outshine ever human player. In naval combat always keeping bombers just behind the line of bb's, flying routes parallel to the bb line and never extending beyond the line at either end. Allowing bombers to break beyond the bb line opens them up to unnecessary fire as they are now unprotected.

I've been thinking of creating short videos for basic tactics that most don't consider or value. Perhaps that would help.

I think the best way to improve the bot is to go about this much in the same way you would improve anything else; come up with a defined list of faults (call them "bugs") and give a matching defined list of corrections (call them "fix/solutions").

Edit: Ninja'd!
Smoke me a kipper, I'll be back for breakfast...
Blackbeard
level4
level4
Posts: 930
Joined: Fri Aug 31, 2007 1:05 pm

Postby Blackbeard » Tue Jan 20, 2009 12:11 pm

Ace Rimmer wrote:I've been thinking of creating short videos for basic tactics that most don't consider or value. Perhaps that would help.


/me wonders if he(Martin) would get some good ideas from watching a few "classic" games, using Dedcon. I know there are plenty to choose from, but without having the game installed at the moment, I wouldn't know where to begin with suggestions.

MOR and Mojo play a bit like (ro)bots, so I wonder if they would be good examples to watch (Mike has a panic attack at the thought of the AI using Super(BOR) tactics :lol: :P ).

Ps. If this has already been suggested/discarded etc. just type /ignore loli :roll:
User avatar
martin
level5
level5
Posts: 3210
Joined: Fri Nov 19, 2004 8:37 pm
Location: ::1
Contact:

Postby martin » Tue Jan 20, 2009 3:08 pm

a specific video of tactics would probably be more useful, and less time consuming - I'm afraid I do still have uni work limiting my time :P
GENERATION 22:The first time you see this, copy it into your sig on any forum and add 1 to the generation. Social experiment.
User avatar
Ace Rimmer
level5
level5
Posts: 10803
Joined: Thu Dec 07, 2006 9:46 pm
Location: The Multiverse

Postby Ace Rimmer » Tue Jan 20, 2009 4:19 pm

I'll see what I can do with my limited time. Of course, I'll have to put my underused tutorial intro on every one of them. :P
Smoke me a kipper, I'll be back for breakfast...
elDiablo
level5
level5
Posts: 3111
Joined: Thu Mar 14, 2002 12:23 pm
Location: London, UK

Postby elDiablo » Tue Jan 20, 2009 4:28 pm

Blackbeard wrote:.../me wonders if he(Martin) would get some good ideas from watching a few "classic" games, using Dedcon...


Now what you want here, see, is one of them Neural Networks. Using the old feedback learning, you can run games into it using orders (from both teams) in some way as input neurons, and the final game score as the output value. Then it should do some learning on which game moves are best!

Obviously it would need more then just orders. Like, current world state. And stuff. Still, YARR.
We dont stop playing cos we get old... We get old cos we stop playing.
User avatar
martin
level5
level5
Posts: 3210
Joined: Fri Nov 19, 2004 8:37 pm
Location: ::1
Contact:

Postby martin » Tue Jan 20, 2009 7:23 pm

Feel free to program that elDiablo, I'll stick to more "conventional" techniques. I'm only in my first year of comp-sci XD
GENERATION 22:The first time you see this, copy it into your sig on any forum and add 1 to the generation. Social experiment.
User avatar
trickser
level5
level5
Posts: 1823
Joined: Thu Mar 06, 2008 2:15 pm
Location: The Senate ; GMT+1
Contact:

Postby trickser » Tue Jan 20, 2009 8:24 pm

Blackbeard wrote:/me wonders if he(Martin) would get some good ideas from watching a few "classic" games, using Dedcon.


I wonder if bots can spec games themself to obtain data. Anybody knows if the API allows that?
elDiablo
level5
level5
Posts: 3111
Joined: Thu Mar 14, 2002 12:23 pm
Location: London, UK

Postby elDiablo » Tue Jan 20, 2009 8:42 pm

martin wrote:Feel free to program that elDiablo, I'll stick to more "conventional" techniques. I'm only in my first year of comp-sci XD


I did a similar thing for my 3rd year project, and Computer Go. Having only 4 weeks to implement it due to implementing other algorithms meant that it didn't "work" as such... :)
We dont stop playing cos we get old... We get old cos we stop playing.

Return to “AI Bots”

Who is online

Users browsing this forum: No registered users and 1 guest