A Perceived Environment Design using a Multi-Modal Variational Autoencoder for learning Active-Sensing

I. INTRODUCTION Active sensing (AS) is one of the most fundamental problems and challenges in mobile robotics which seeks to maximize the efficiency of an estimation task by actively controlling the sensing parameters [1]. AS can be divided into two sub-tasks: the identification of a point of interest (PoI) to achieve (e.g. In teams of heterogeneous robots that employ different sensory modalities, AS is of particular interest as it can be used to resolve observation ambiguities. Friston [2] states that minimizing free energy is equivalent to maximizing model evidence, which is equivalent to minimizing the complexity of accurate explanations for observed outcomes. Following this principle, if one could directly obtain an estimation of free energy through the current observation, a controller for sensing parameters can be learned that minimizes free energy.