Grod
Joined: Sep 30, 2003
|
 Posted:
Oct 18, 2010 - 10:21 |
 
|
| TuernRedvenom wrote: | | Flix wrote: | | the FFB Projekt took years, a AI Projekt would take centuries |
To let it beat at CR 150+ coach? Yes.
|
To be honest, I am a little hurt. I thought that the GrodBot (not even complete as it is) would already be challenging for a CR 150 coach ... |
_________________
I am so clever that sometimes I don't understand a single word of what I am saying.
Oscar Wilde |
|
peikko
Joined: Aug 02, 2003
|
| Posted:
Oct 18, 2010 - 10:41 |
|
| Grod wrote: | | To be honest, I am a little hurt. I thought that the GrodBot (not even complete as it is) would already be challenging for a CR 150 coach ... |
Well, atleast its win ratio vs Cyanides AI even with its hardest settings would be crushing for sure. Though I bet my cat could beat that too. |
|
|
Grod
Joined: Sep 30, 2003
|
| Posted:
Nov 02, 2010 - 18:37 |
|
Latest update:
Firstly, check my Blog for details of the latest AIBowl release, it is updated pretty frequently (look at the release notes for details).
Next
I got myself side tracked...
Since I was developing AI, I decided to look into something called "ChatBots". It took me a bit of playing around to work it all, but essentially I have got the (famous) open source chat Bot ALICE up and running. If you want to test it out,
Download and expand the following file:
http://www.ahd.tudelft.nl/~peterm/BotChat.zip
Execute the "run.bat" file.
It is not yet connected to a BloodBowl client, but I have the vague idea of (in the future) allowing the Bots to chat with the players during the game (and vice versa). The software is based on something called "Program D" (if you want to Google it), with the ALICE aiml files. The next step would be to update the ALICE personality so that it knows something about BloodBowl by editing the aiml files.
It is pretty funny to download and try to chat to (although it is limited), so let me know what you think. If you have any problems getting it up and running, let me know (I did test it in Linux and windows, so it should work). |
_________________
I am so clever that sometimes I don't understand a single word of what I am saying.
Oscar Wilde |
|
Grod
Joined: Sep 30, 2003
|
| Posted:
Nov 12, 2010 - 11:38 |
|
Hey dedicated Bloodbowlers.
I need some suggestions about what is missing in the GrodBot to make it a perfect Bot. Now that I have ironed out all the bugs (at least I don't receive any bug reports any more), and the AI project appears to be rock-hard stable, it is time to develop things further.
At the moment I have implemented leap moves and using the secret weapons on my development version, which will be released when it is fully tested. At this point, it is time to improve the AI.
So I am open to suggestions. What would you try to get the AI to do, to make it better? For example, would you have the AI foul more? Place defensive sweepers? Be more aggressive?
I am open to all kinds of ideas. |
_________________
I am so clever that sometimes I don't understand a single word of what I am saying.
Oscar Wilde |
|
Garion
Joined: Aug 19, 2009
|
| Posted:
Nov 12, 2010 - 11:39 |
|
Could you not make a slide bar which controls the propensity for the bot to do these things? |
_________________
 |
|
Glorin
Joined: Aug 18, 2009
|
| Posted:
Nov 12, 2010 - 11:47 |
|
Another issue is skills. It'd be boring to make an AI with TR200 teams that are all just Block, Tackle, Dodge/DP on linemen and Block, Guard, Tackle on blitzers etc. etc. One of the best things about bloodbowl is I can make a skaven team built to have a blitzing GR, a one turner, and two catchers for Hail Mary plays, and utilize them for the other jobs. I think it'd be very hard for an AI to work with a lot of different skills and playstyles built into one team. |
|
|
Mr_Foulscumm
Joined: Mar 05, 2005
|
| Posted:
Nov 12, 2010 - 11:47 |
|
| Grod wrote: | | . The next step would be to update the ALICE personality so that it knows something about BloodBowl by editing the aiml files.. |
So... you're aiming to make the GrodBot a whiner? Just what we need, coz we don't get that playing BB anyway  |
_________________
Everybody's favorite coach on FUMBBL |
|
SillySod
Joined: Oct 10, 2006
|
| Posted:
Nov 12, 2010 - 12:06 |
|
I dont think that you need the GrodBot to do things more or less often. You need the GrodBot to improve its technical skills, stuff like effective blocking and screen formation would go a long way towards improving the win rate. Programming the bot to score touchdowns using a specific style would help too. |
_________________
Putting the "eh?" back into Sexeh.
"There are those to whom knowledge is a shield. There are those to whom it is a weapon. Neither view is balanced." |
|
Grod
Joined: Sep 30, 2003
|
| Posted:
Nov 12, 2010 - 13:01 |
|
| SillySod wrote: | | I dont think that you need the GrodBot to do things more or less often. You need the GrodBot to improve its technical skills, stuff like effective blocking and screen formation would go a long way towards improving the win rate. Programming the bot to score touchdowns using a specific style would help too. |
Yes, that is what I am asking for suggestions on. Which technical skills the Bot needs to have.
Screen formations is high on the list so far. |
_________________
I am so clever that sometimes I don't understand a single word of what I am saying.
Oscar Wilde |
|
Wietse
Joined: Oct 09, 2003
|
| Posted:
Nov 13, 2010 - 09:46 |
|
pretty nice work grod.
I tried a game of bot vs bot, but neither team payed any attention to the ball during the first half. That was kind of weird. |
|
|
Grod
Joined: Sep 30, 2003
|
| Posted:
Nov 16, 2010 - 13:10 |
|
| Wietse wrote: | pretty nice work grod.
I tried a game of bot vs bot, but neither team payed any attention to the ball during the first half. That was kind of weird. |
Hmm. If the ball is on the sideline, or too far away to reach it in one move, the players may be ignoring it. This needs some updating  |
_________________
I am so clever that sometimes I don't understand a single word of what I am saying.
Oscar Wilde |
|
Grod
Joined: Sep 30, 2003
|
| Posted:
Nov 16, 2010 - 14:13 |
|
By the way, I have a challenge for anyone who can help me.
I included the use of the Leap skill in the pathfinding on the version I am testing, but the computation became very heavy (but still, just, playable).
I did did the simplest method I could think of (always the best way to start):
For a give player
* Execute a pathfinder to all possible squares.
* From each ending point, if there was movement left over on the path to reach it, perform a leap to all possible leap squares.
* Execute a new pathfinder from the leaped-to square (with corresponding reduced movement)
* For each square on the secondary path finding:
- If the total cost (probability based) of the new path is smaller than any existing path that arrived there (or if no other path arrived there), replace existing path with new compound path involving leap.
It seems to work properly, but I won't release it before I am sure it is completely stable. However, the method outlined above is fairly brute-forcish. I am open to some inspiration to improve this if anyone has some good ideas?
The trouble was, I couldn't think of a natural way to include a single leap within a single AStar Path Finding algorithm, but the solution above is probably not the best...
Cheers. |
_________________
I am so clever that sometimes I don't understand a single word of what I am saying.
Oscar Wilde |
|
Christer
Joined: Aug 02, 2003
|
I'm not entirely sure what the problem is for you Grod, but here's what I'd do:
- For each square, keep track of two path-costs, one for leap and the other for non-leap. I'll call these two variables Gn and Gl. I'll call the heuristic cost F. Also keep two sets of "came from" references, one for leap and the other for no leap
- Process the nodes from the open set in order of min(F+Gn, F+Gl)
- When processing neighbour nodes, update costs for both leap and non-leap in all target nodes. Note that there will be considerably more neighbour nodes due to the leap (20 extra ones).
In fact, you should most likely generalise this into having any number of sets of costs in order to support strategical decisions like "find the best path that doesn't require rerolls to be used". More than likely, you also want this to be able to be computed in a "lazy init" way (ie, don't perform the calculation until something actually requests it). |
|
|
Grod
Joined: Sep 30, 2003
|
| Posted:
Nov 16, 2010 - 16:14 |
|
Thanks Christer.
There were two different, but very good tips there. I will take a look at it tonight when I get home. I guess I need two "closed lists" etc, I will need to check the details.
I also like the the "lazy init" idea |
_________________
I am so clever that sometimes I don't understand a single word of what I am saying.
Oscar Wilde |
|
Macavity
Joined: Nov 23, 2004
|
| Posted:
Nov 16, 2010 - 17:06 |
|
| Grod wrote: | | Wietse wrote: | pretty nice work grod.
I tried a game of bot vs bot, but neither team payed any attention to the ball during the first half. That was kind of weird. |
Hmm. If the ball is on the sideline, or too far away to reach it in one move, the players may be ignoring it. This needs some updating |
Yeah, as far as technical skills, the ability to move to nowhere in particular because it's near/in the path of the ball carrier and could help next turn would do leaps and bounds for win ratio, I would think. |
_________________
When I became a man I put away childish things, including the fear of childishness and the desire to be very grown up. -C.S. Lewis |
|
|
|
| |
Finishing the 60 Gam...
Search
