xurxodiz / MarioLevels

My entry for the Mario AI Championship 2012, implementing adaptive procedural level generation. Holds 1st place at the all-time points and percentage tally.
MIT License
2 stars 1 forks source link

Design the core automaton #16

Closed xurxodiz closed 12 years ago

xurxodiz commented 12 years ago

As per #7, we now will try to model the level building as a Markov process. We will, therefore, create an automaton in which each state is a different chunk, and upon entering them the map building function is called. The transitions between them represent the probabilities of a certain chunk to sequentially follow another certain chunk.

We must design the different states needed, and the transitions between them (links only, not values).

Absolutely required before proceeding to #17.

xurxodiz commented 12 years ago

This has become the most important issue right now. Figure out what states to implement and connect them accordingly.

xurxodiz commented 12 years ago

Momentarily on hold while we complete the clustering part of the system. To be picked up again as soon as we have a transversally functioning platform.

xurxodiz commented 12 years ago

The automaton works like this: we start at initial, move to start to create starting platform, and finally to flat for some buffer. Then we reach hub. hub redirects to all the other parts, according to the probabilities. We have coins, blocks, gaps, cannons, pipes and enemies. (We are missing hillsteps and background hills).

pipe can lead to pipepiranha. coins, blocks, flat and enemies can loop over themselves.

enemies is still in an early stage. It will comprise several more states, per its possible combination with blocks, double blocks, coins and all the available types of enemies.

xurxodiz commented 12 years ago

We consider it designed enough for the paper. Moving it to thesis, since it may also be affected by issue #38, on evolving the transitions (but we'd need nice starting points) and issue #39, on evolving the automata themselves.

xurxodiz commented 12 years ago

Closing it since too many things have changed since its inception (we have several automata, for starters).