next »

[EmmerichSpiel]

Hey fellow developers,

I was wondering what kind of AI would you guys recommend for creating an enemy in a fighting game, like Mortal Kombat?

For example, I was thinking that the enemy could have a raycast vector to detect where the player is, then use RNG to decide the enemy's next move.

Do you all think that this type of AI is effective?

[droqen]

Raycasting makes sense in a first-person shooter, but not in a fighting game - in a 1v1 fighting game, the spaces are generally very simple (often they are a featureless rectangular space) and you can always see exactly where your opponent/s is/are.

Randomly selecting moves will probably create an interesting and unpredictable opponent for a beginner to start out with, as long as the AI can understand spacing correctly (i.e. you can start with making sure it does not use short-range moves when at long range), and from there you can hopefully continue developing solutions to easy tactics that exploit the ai's edge cases.

[raigan]

You could also consider researching how similar games have tackled this -- for example, here's an overview of how SF2's AI works, which ought to be perfect for you: https://sf2platinum.wordpress.com/2017/01/20/the-ai-engine/

This looks to be very data-heavy, ie they authored little scripts for every possible combination of fighters, and annotated every attack of character with a "yoke" which instructs the AI which counter-attack to throw, but the general idea (a small state machine or even just a set of if/else rules) seems workable: "can any attacks hit the opponent? Y: throw one of them; N: move closer".