BACS: integrating behavioral sequences to ACS2

In many real-world environments, only partial observations are provided, thus presenting challenges for Anticipatory Learning Classifier Systems (ALCS). The perceptual aliasing issue occurs when systems cannot differentiate situations that are truly distinct. To tackle the perceptual aliasing issue, ALCS classifiers can be chained in order to build Behavioral Sequences. Those sequences permit ALCS to deal with this issue, but they have never been implemented within ACS2 (Anticipatory Classifier System 2), although this is one of the most advanced ALCS. This paper introduces a novel learning classifier system, BACS, that integrates Behavioral sequences to ACS2. This integration required the adaptation of the action selection policy, the integration of an aliasing detection algorithm that let the system build behavioral classifiers, and the adaptation of the anticipatory learning process. The results obtained over a maze environment benchmark show that behavioral sequences are a promising approach to address the perceptual aliasing issue.