by date
Plan for Phil Carlisle
Plan for Phil Carlisle
| Name: | Phil Carlisle |  |
|---|---|---|
| Date Posted: | Jan 28, 2005 | |
| Rating: |  Not Rated | |
| Public: | YES | |
| Comments: | YES | |
| RSS Feed: |  or Subscribe with  . | |
| Profile Page: | View profile page for Phil Carlisle |
Blog post
Just thought I'd spread the word for my students..
Bolton Institute has finally been awarded our University title!!! we are now officially called
The University of Bolton
Now, this doesnt actually make any difference to me, but it sure as hell can make the difference to our students. It gets a bit hard explaining what an "institute" is :)
We've been doing degree and MSc courses for a while, so its not really new as an educational institute, but its at least easier to chat with people about.
On another note, I noticed something interesting testing the AI pack code the other night.
I have a 3vs3 bot CTF game going and I was basically laying out a simple level for use as the demo. What I was surprised initially by, was that the game was stalemate!
Basically, one player goes for the flag, one goes out to guard and one stays to defend.
But, I'd switched back in some code that allows for almost totally accurate firing. Now initially, there was some variance in the firing so I was getting flag runs and the occasional capture as expected.
Putting back in accurate firing led to this stalemate situation where both flag runners would get into the middle of the level, blast hell out of each other, one would win and carry on until he met the other teams scout.
The other teams scout would proceed to blast the flag runner and repeat...
It shows you that basically, the game is totally balanced, you then have to insert random elements (like innacurate firing, or changeable territory values or whatever) to allow for the balance to swing.
Of course introducing humans into the equation would obviously increase the complexity and hence the balance, but knowing that on an evenly matched 3vs3 game with only AI players, you SHOULD in theory reach a stalemate! which we did!!!
Ok, interesting to me at least :)
Bolton Institute has finally been awarded our University title!!! we are now officially called
The University of Bolton
Now, this doesnt actually make any difference to me, but it sure as hell can make the difference to our students. It gets a bit hard explaining what an "institute" is :)
We've been doing degree and MSc courses for a while, so its not really new as an educational institute, but its at least easier to chat with people about.
On another note, I noticed something interesting testing the AI pack code the other night.
I have a 3vs3 bot CTF game going and I was basically laying out a simple level for use as the demo. What I was surprised initially by, was that the game was stalemate!
Basically, one player goes for the flag, one goes out to guard and one stays to defend.
But, I'd switched back in some code that allows for almost totally accurate firing. Now initially, there was some variance in the firing so I was getting flag runs and the occasional capture as expected.
Putting back in accurate firing led to this stalemate situation where both flag runners would get into the middle of the level, blast hell out of each other, one would win and carry on until he met the other teams scout.
The other teams scout would proceed to blast the flag runner and repeat...
It shows you that basically, the game is totally balanced, you then have to insert random elements (like innacurate firing, or changeable territory values or whatever) to allow for the balance to swing.
Of course introducing humans into the equation would obviously increase the complexity and hence the balance, but knowing that on an evenly matched 3vs3 game with only AI players, you SHOULD in theory reach a stalemate! which we did!!!
Ok, interesting to me at least :)
Recent Blog Posts
| List: | 11/28/08 - GDC AI sessions 09/18/08 - Tell me I'm not going crazy!! 12/05/07 - The importance of good tools for productivity 11/17/07 - Using the way back machine. 09/21/07 - Juggling cats. 09/04/07 - End of Summer. 08/27/07 - Come work with me!! 08/14/07 - The changing nature of entertainment |
|---|
Submit your own resources!| dakz0rz (Jan 28, 2005 at 11:28 GMT) |
| Craig Fortune (Jan 28, 2005 at 12:01 GMT) |
As for the balance stuff thats so cool, in one way it shows just how good that pack can be if its producing results like that.
Well done :D
| Nauris Krauze (Jan 28, 2005 at 12:08 GMT) |
| Nick Zafiris (Jan 28, 2005 at 15:01 GMT) |
Nick
| Peter Dwyer (Jan 28, 2005 at 16:07 GMT) |
It appears to me that if the accuracy of the shots increases the AI should have been seeking cover or alternate routes home not simply getting blasted to pieces. Also played a few games where bot accuracy has been an issue. Enoguh to make me stop playing those games!
Such unfair senarios as being hit while running a zigzag pattern because the AI has the advantage of both response speed and pre-calculated trajectories for it's shots!
Still sounds like you are having fun.....
| Phil Carlisle (Jan 28, 2005 at 21:39 GMT) |
Of course, this is very subtle, but an interesting side effect.
As Nauris says, it does tend to show up issues of imbalance (for instance, making a specific path home "safer" for one side).
| Ben Garney (Jan 29, 2005 at 01:24 GMT) |
| Benjamin Bradley (Jan 29, 2005 at 04:46 GMT) |
Edited on Jan 29, 2005 06:07 GMT
| Peter Dwyer (Jan 29, 2005 at 09:44 GMT) |
This is why I am puzzled as to how there is a stalemate. Do you just have a flat map with no cover areas or does the AI have no decision tree for seeking cover, clearing obstacles (in this case that would equate to the enemy teams AIs) etc. etc.
It sounds like you just have goal led AI such as the flag carriers instructions are get flag, return flag to base. The scounts instructions are move forward and eliminate any enemies. The base guard is simply area guarding and eliminates any enemy in the trigger area.
Didn't mean to dismiss your AI, I guess I'm working on an AI level where the decision trees are in the order of.
Scout objectives - Sweep outwards. Seek high ground if possible, report positions, engage enemy only if enemy engages you. Snipe from a distance (targetting offset by small random amounts to allow for hand wobble).
Guard objectives - cover base entrances, favour quick kill weaponry (slower fire rate but effective in confined base areas). pursue fleeing enemy if enemy leaves trigger area (personal score goals I call this). Return to base if more than 30 - 50 meters from base (hey think of the team man!) etc. etc.
| Nick Zafiris (Jan 29, 2005 at 10:53 GMT) |
Game/AI balance is something that must be adjusted and tweaked many times during the development of the game. And since no two games are the same, it's just impossible to just use the default values that come with any AI pack.
Just my thoughts:)
Nick
| Phil Carlisle (Jan 29, 2005 at 12:55 GMT) |
We arent trying to do a tactical AI but just show how to do the basics. I'm sure that there are many many layers of much more game-specific AI we can incorporate down the road, but right now we're covering the basics so that everyone has a head-start.
| Peter Dwyer (Jan 29, 2005 at 13:21 GMT) |
Thanks for clarifying. I guess it was just hard to visualise without images of the level.
Do you use visual cues for your AI?
I have mine set-up so that they "See" where they are going but, will turn around to face the direction of incoming fire and attempt to spot where it is coming from before laying down fire of their own.
I'm going more for tactical AI where enemies will allocate roles for each other. This allows the player to take out a command AI and have the others revert to their pre-defined personalities, be that run away screaming (drop weapon and just plain run for it), death rush (take you head on), pin down (stick head out every now and then and shoot at your position. Side note: you can fool these guys by moving when they duck down. They shoot where they think you still are...well I think it's clever anyway ;o) ) or back stab (attempt to flank and well shoot you in the back).
Some of my command AI are proving to be quite difficult to defeat due to their self preservation level i.e. they sit in the back ordering heavy gunners to pin you down while ordering grunts to flank your position and snipers to cover the middle ground in case the heavies run out of ammo or need to re-load :oP
I introduced finite ammo for all AI so that both the player and the AI can end up knife fighting!
| Phil Carlisle (Jan 29, 2005 at 20:52 GMT) |
Yeah, we have visual cue's for some of the ops the AI is doing. Ive got an iconic system that basically gives the player an indication of what senses the player is using (essentially wether theyre guard, attacker, scout looking about etc).
Obviously there are loads of things you CAN do :) but we've got to keep in mind that our pack is essentially for everyone, so we are trying to keep things usable at a multitude of levels.
I am sure we will work on a more tactical AI as a later update. I've always wanted to do a tactical squad based AI.
Knife fighting sounds fun though :)
Personally, I'm interested in having a go at will van der sterrens grenade throwing and tactical positioning ideas.
Phil.
| Lobo the Duck (Jan 29, 2005 at 23:59 GMT) |
i dunno weather i should be askin this here but are you doing a summer school at bolton, doing game development, cos im tryin ta get a place for a game dev summer school course at university of bolton and would be honored if i was being taught by you.
i was just wonderin if you were doing the teaching.
regards
Mozel
Edited on Jan 30, 2005 00:00 GMT
| Phil Carlisle (Jan 31, 2005 at 09:24 GMT) |
No, that will be someone else I believe. Not sure who right now though. During the summer I expect to be in asia somewhere.
| Tom Bampton (Jan 31, 2005 at 11:13 GMT) |
T.
| Lobo the Duck (Jan 31, 2005 at 16:33 GMT) |
You must be a member and be logged in to either append comments or rate this resource.
