search

Zannick / logic-graph

Tools for video game logic representation and analysis, particularly routing and beatability checks for speedruns and randomizers.
MIT License
3 stars 0 forks source link

Smooth existing solutions #66

Closed Zannick closed 1 year ago

Zannick commented 1 year ago

We frequently find greedy solutions from imperfect states. In AV2, these are like: "go to spot, deploy drone, go to other spot, recall drone, go to third spot" or "do long deploy action to move to spot, get map visited item, recall drone, (do other stuff...), do long deploy action again".

We ought to be able to examine such a string of history and reach a subsequent state faster than the original discovered, but the regular search has not obviated that solution yet. Our current version of this is to just insert into the queue the intermediate states, but this doesn't work very well if the states are never picked up. This would help us find better solutions earlier on.

Zannick commented 1 year ago

Oops, I just realized that the optimized sequence of states still has the skips of each location, so of course adding them to the queue just results in a ton of dead-ends. Either we have to go through and fix those, or use the history from the best one to recreate them all.

Zannick commented 1 year ago

We might revisit this later when a game graph is more complete, but for now having a route defined accomplishes the same original goal of having a good solution to start from (and use as a time limit).