Machine Learning for procedurally generated levels.

Started by
2 comments, last by wodinoneeye 9 years, 2 months ago

If I could parse a phrase or a sentence to the algorithm, how easy would it be to generate a level based on the semantics extracted from the sentence. Especially for a puzzle/adventure game.

Imagine parsing the words: "latent ball in slope" and generate a level that contains a slope where a ball rotates and menaces squashing the player.

boulder.png

Intel Core 2 Quad CPU Q6600, 2.4 GHz. 3GB RAM. ATI Radeon HD 3400.
Advertisement
Extremely difficult.

I think it would sort of depend on how varied you wanted it to be. Provided you had your level editor pretty much set up, doing what you ask there wouldn't be terribly difficult in and of itself, I wouldn't think. I mean, if you only provided a small set of options, say, ball, square / slope and plane, setting that up wouldn't be too hard.

If, on the other hand, you want to account for anything that might be typed in, that gets much, much more difficult, I imagine. I think if you limit the options to a set of pre-defined things, such as a set of geometric shapes, and a set of different inclines, etc. something like this wouldn't be terribly difficult to implement.

You're posting in the AI section though, so I imagine you're wanting a more robust handling of things, and I couldn't begin to offer a good solution there. But, I think if you considered scaling down the scope, and had a defined set of things the user knew they could type, you'd be in much simpler territory.

Beginner here <- please take any opinions with grain of salt

Depends on how complex/simple.

If the scene definition is simple (no conflicting elements) then its just building the situation from a limited list of objects/terrain/actions/results mapping from the limited description words.

IF you include that word 'AND', and you allow elements to conflict or act in combinations thene there will be a great deal more logic that has to be created to handle all the possibilities. You might also need a 'validator' which has yet more logic which will tell the user IF the situation being defined is nonsensical (incompatible as specified) withing your world mechanics.

Another level (Ive looked at auto-genration alot) is to have Templates where withing one named definition is actually a whole sub scenario of multiple objects/terrain/actions/results (like 'waterfall' which imples a height with falling water (and maybe a flow away of the water)). The grouping of associated entities making up that template may then have additional rules/mechnics dealing with THAT grouping's behaviors/etc...

Again, if you can spell out the situation using multiple templates (or one template AND'd with other simple objects) then there might need to be additional validation code for the special cases for that Template.

--------------------------------------------[size="1"]Ratings are Opinion, not Fact

This topic is closed to new replies.

Advertisement