Meta-behavioral extraction of Atari shooters

Abstract

The purpose of this study is to investigate the learning of meta-modules in symbolization in game BOTs. To extract information common to similar games, we considered similar games and trained a hierarchical structure with a network that outputs meta actions and a network that outputs actions based on game-specific features. In Atari’s Assault, the learning results for the meta-module alone were lower than those for random as the learning progressed. Control by the intrinsic module is necessary.