Robo Home | Changes | Preferences | AllPages

Restart working on my bot "Melody" after a long time
Part of the movement code is borrowed from Voidious's WaveSurfing/Tutorial
i = argmax_i P( S = i | O )

which, according to Bayesian rule, is equivalent to

i = argmax_i P( O | S = i ) P( S = i )

Note that P ( S = i ) can be estimated easily, and is alway done by any basic GF targeting algorithms. The first factor may be calculated in many ways. If we do VQ in the feature space, O will become a discrete random variable, and the statistics can be collected via counting. In that case, it would be equivalent to many Segmentation algorithms. It is also possible, though I'm not sure whether it will be better, that P ( O | S = i ) is modeled using continuous distributions. In the simpliest case, this conditional probability can be just ignored, which gives a basic GF gun. This flexiblity might be useful when data is not sufficient, or when we're not confident enough about the first factor.

Comments, questions, feedback:

You pushed me (GrubbmGrb) out of the top-50!! When I can find the time to work on the movement of my new bot, I will beat you again ;-) Quite a nice entry for basic GF-targeting and basic WaveSurfing -- GrubbmGait

Robo Home | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited July 14, 2006 13:58 EST by GrubbmGait (diff)