Browse files

update README

  • Loading branch information...
1 parent c58c543 commit 0bc98cd99ed916b6a17aa4a9db7c8c5fd8f53348 @chriskiehl committed Feb 4, 2012
Showing with 45 additions and 0 deletions.
  1. +45 −0 README
@@ -0,0 +1,45 @@
+Burrito Bison bot.
+Below are some of the thoughts and failed ideas that I went through while wrtting this thing. All the problems I faced
+are most likely really simple and will likely seem trival as I learn more, but I leave them here for those who, like me,
+are just starting out.
+Simple little script to 'play' Burrito Bison (found here: The code is pretty
+weak, this was written just to learn python a little more, and these 'launch' games all seemed mindless enough that
+I could easily get a bot to control them.. (it ended up being trickier than I expected)
+The code is pretty crappy, and there are tons of things I'd do differently, and, in fact, if you look at the code you'll
+notice that calls start changing little by little and becoming more concise. Which was really the point of this, to
+get a little better at constructing a program. So the things I don't like about how it's written now, didn't occur to
+me as possible issues when I first started coding, but little by little, better ways of approaching a problem became
+It's now completely workable, and will play the game from start to finish. There's only one menu that it doesn't catch,
+I noticed it when I was recording the final start-to-finish playthrough, but I was too lazy to fix it.
+The bot works by gathering in about the game by taking screen shots and then checking specific pixel locations.
+It's a very fragil way to do things, as any difference in browser size ruins the expected pixel values.
+I tried all kinds of things to get around this problem. Being that all important events are different colors from the
+main stage and gummies (like the cop for instance), I tried all kinds of funniness with averaging the colors of a
+snapshot, assuming that when a new color appeared on screen the average color disrtibution would change. I still think
+it's a neat idea for flexibly checking what's on screen, but in practice I couldn't get it to work due to speed issues
+(and probably poor code issues). Several variations of this idea were tried in an effort to reduce the number of
+calculations needed. A few attempts were made at grayscalling the image first and then getting the colors, others
+included taking a full snapshot of the playarea, thumbnailing it to something like 25x25, and then averaging the colors
+that way.
+Of all the things I tried (revalting to the getcolors idea) this proved the most reliable way
+of noticing when a cop was on screen. Reliable in my case being about (maybe)10% success rate. Even the current
+iteration is quite poor, altough, when the bison is moving at slower speeds I'd say it's around a 40% success rate.
+Which I was happy enough with to move on. There tons of things with the current implementation that could raise this
+number quite a bit (I think). Such as multiprocessing some of the search events so there's not so much time inbetween
+calls, turning off all of the other search features that aren't needed anymore (like dialog boxes) would free up a ton
+of processing room. They're the old version that takes a unique screen shot instead of 'sharing' one between all

0 comments on commit 0bc98cd

Please sign in to comment.