/* attempt to deal with prototype, bootstrap, jquery conflicts */ /* for dropdown menus */

Jump to content

Image of the Day

Day 17 -- a bunch more tools join the table :) #blender3d #screenshotsaturday #itssaturdaysomewhere #lowpoly #3D https://t.co/CYJucn0Lfv
IOTD | Top Screenshots

The latest, straight to your Inbox.

Subscribe to GameDev.net's newsletters to receive the latest updates and exclusive content.

Sign up now

How does enemy AI work in fighters?

4: Adsense

Old topic!

Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
3 replies to this topic

#1 Bamcee   Members   


Posted 10 April 2013 - 06:18 PM

Well I've been wondering how the enemies in fighting games work, what I mean is the coding for their movements as in blocking,dodging etc.

#2 BCullis   Members   


Posted 10 April 2013 - 08:12 PM

The result you see in most fighters (I'll use a 3d fighter as an example) is the result of a physics system, a visual animation system, and the AI making attack choices.  The AI itself can be as simple as a state machine with various conditions to check for (like "is the player in the air?"  "Are they far away or close?")  Once the AI makes its choice (there are a large number of solutions for "making choices"), it picks an attack or action, and that is tied to an animation sequence for the fighter character.  So once the AI fighter chooses "low kick", it starts the low-kick animation that has been rigged by a character animator.  If the collision surface of the fighter's leg collides with the player character, that's when a "reaction" animation is played on the victim, or perhaps is done with ragdoll physics.


Though that's just one potential solution, there's probably as many solutions as there are fighting games.

Hazard Pay :: FPS/RTS in SharpDX (gathering dust, retained for... historical purposes)
DeviantArt :: Because right-brain needs love too (also pretty neglected these days)

#3 RedBaron5   Members   


Posted 11 April 2013 - 09:49 AM

I'm currently working on a fighting game and this is how I have my AI set up.


For enemy defense:

Enemy has a "chance to block number" that gets checked every time the player attacks.  If the check passes the enemy blocks.  I modify this number for certain situations.  For example, immediately after the player parries an attack, the chance to block for the enemy becomes zero, this allows a "free" counter attack.  Another example is setting the chance to block to 100 after the player performs 2 identical attacks to prevent attack spamming.  The next attack that is performed will be blocked.  For more complexity, you could have a chance to block number for each type of player attack and vary those for different situations.


For enemy offense:

This is still a work in progress for me.  I wanted the enemies to be reactive to player commands but I also wanted them to not sit and wait for the player to do something.  I gave each enemy an aggressiveness number that I vary based on certain situations (health remaining, consecutive hits by player).  I use this number to determine the length of time before an enemy attacks.  Other than that, I have the enemy react to player attacks.  If they block an attack they have a chance to immediately counter attack, if an attack connects they have a chance to perform the same attack again.


For enemy "special" attacks, I have a few situations where they are performed.  The first is simply a random timer (with a minimum time) that resets after they use a special.  The next is player/enemy health thresholds.  And the third is consecutive hits by the player.


My system isn't at all perfect and I literally did no research on it at all, its just what I came up with.  Hopefully, it gives you some ideas.

#4 redw0lf   Members   


Posted 11 April 2013 - 11:37 AM

I have not much knowledge of gaming AIs but as far as I think, what an AI basically does as mentioned in RedBaron5 example is calculating some kind of possibility either for defensive situations or offensive situations. Like e.g. how high is the possibility to lose health during the next turn, if it is high you either either try to dodge or block depending on whether the chance for blocking or dodging is higher. To model these states and dependencies you can use a state machine or some kind of tree. But this is calculated only for one move, the next may look totally different. Finding a good strategy here may be difficult


If you take a look at chess, a computer calculates x moves ahead, giving each move a score of 1 for a move where his situation will get better or a zero for where the situations gets worse. He can now sum up the scores for each path and choose the path with the highest score. This an easy task because you know each position and the possible next tasks for each figure. If you want to increase/decrease the difficulty of the AI in a chess game you could easily limit the moves which are calculated ahead.  In a fighter you could also think of such a way to calculate a strategy, e.g. using the elements move, fight, dodge, block and then search for the path yielding the maximum of hp loss for the player.

do! develop! create! - visit my scripters-corner.net

Old topic!

Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.