PBEM vs. the PC

[Exer]

What's the update on the AI work? This thread died so suddenly.

-
t

[Pink Horror]

Well, I've been working on team analysis up until the last week or so. Then I spent the next few available days looking through a couple books and planning things out. My brother visisted for a few days, and now I'm at that period where I have to get myself to stare at the balnk screen and write the first line. Once I get started, it could be a couple weeks before I get my end of the interface squared out, and then it will take some time for SkiJunkie to do his end, and maybe by then I'll have a computer opponent made that simply blocks who it can block and moves guys towards the ball. I'm sorry I haven't started any code yet, but sometimes even those of us with no life have other things to do.



Pink Horror

[Guest]

What about the AI on the Microleague version of Bloodbowl for the PC, any way to use/pirate/copy/etc from that? It was a semi stupid AI but at least it's got some sort of a skeleton to work from. Too bad the game they put out is semi buggy and never had any follow up to it ie deathzone/patches/support......... it had the makings of greatness.

Frank

[Piepgrass]

Any nws on this or did it die ?

[Dragoonkin]

Death.

[Marc Vaughan]

If anyone is planning to do AI for a bloodbowl game then I'd be happy to lend advice - I've been a professional games programmer for over 7 years (nearly exclusively on sports titles).

My email addy's in my profile (marcv2@sigames.com).

PS. Sorry - but I don't have enough spare time to devote to such a project myself, much as I love BloodBowl ...

[mepmuff]

Being a huge BB and CM fan i really really really hope someone takes Marc up on his offer!

[GalakStarscraper]

Being a huge BB and CM fan i really really really hope someone takes Marc up on his offer!

I'd love to take someone up ... but I don't enough about AIs to build it and unless Marc knows Delphi ... its won't help much.

Just making the AI for interceptions and Side Step were big projects. Automating an entire turn ... I just cannot think how you would go about deciding where to dodge, who to blitz, how to plan 2 turns ahead for your defense. That's some intense programming.

Galak

[martynq]

Automating an entire turn ... I just cannot think how you would go about deciding where to dodge, who to blitz, how to plan 2 turns ahead for your defense.

Whoa... hold it... who said anything about thinking 2 turns ahead? I don't manage to do that when I'm playing as it is!!!

Martyn

[Marc Vaughan]

Being a huge BB and CM fan i really really really hope someone takes Marc up on his offer!

I'd love to take someone up ... but I don't enough about AIs to build it and unless Marc knows Delphi ... its won't help much.

Only bits and pieces I'm afraid - my mum took a computing course a couple of years ago which contained it (and guess which mug got to do her homework?).

Just making the AI for interceptions and Side Step were big projects. Automating an entire turn ... I just cannot think how you would go about deciding where to dodge, who to blitz, how to plan 2 turns ahead for your defense. That's some intense programming.
Galak

Best bet for that sort of thing (presuming you don't have the resources to spend months full-time on the AI) is to do functions to analyse the current board position generally and return warnings about strengths and weaknesses in defense/attack.

When writing AI for a board game simply try and break down the thought processes you go through when playing it yourself and put them into programming terms - hey presto AI (its pretty much what we do in CM, although we simulate the thought processes of each individual player/referee etc. in the game).

For example write a function to check for unmarked players who could recieve passes and return a list of them with scores indicating how dangerous their position is (ie. an unmarked player who has a clear run for a touchdown would score high, an unmarked player who has 3 players who could catch him easily won't score so high), etc.

Off the top of my head other defensive functions could include:
check_for_mismatched_defenders - ie. defenders who are going to 'go down' against stronger opposition
check_for_potential_sacking - ie. defenders who can reach the guy with the ball
check_for_easily_pummelled_player - ie. people who you can easily take down because they're surrounded by your players

Outside of this give players sensible defaults for them to use if they don't get triggered by any of these catches, eg. mark players, move towards ball, regroup with other players or fall back into a 'spread' formation where they (and other team-mates) cover the pitch.

If you do this analysing the results from the functions will give you a decent amount of pointers about what should be done tactically to defend a position.

If you get this analysis right then you won't need to try and 'predict' problems because your AI will always try and keep a tactically sound position and should be reasonably able to recover from most problems.

(for attack obviously reverse this psychology)

Obviously if someone wants to discuss this sort of thing in more detail rather than off the top of my head thoughts then feel free to email me as I said ...

[Perox]

There's a none-dead thread on fumbbl.com:

http://fumbbl.com/index.php?name=PNphpB ... 7777#27777

Happy to discuss it here too, if anyone is still interested..

In short, my idea is that a board evaluation program is needed first, and would be the basis of any AI people would be inclined to build. Once the puter can evaluate the board, it can decide which moves are better than others.

Perox.

[Mostly-Harmless]

I´ve been working on an AI Framework for BB for some time now...
There is lots of work to be done so everyone is invited to join the bbalc project.
Please visit http://auf.schalke.internnetz.de/bbalc/ and contribute your ideas, tactics and/or code to the wiki page.