An Active Inference approach to the multi-agent lost target search problem

Keywords

Loading...
Thumbnail Image

Issue Date

2022-03-11

Language

en

Document type

Journal Title

Journal ISSN

Volume Title

Publisher

Title

ISSN

Volume

Issue

Startpage

Endpage

DOI

Abstract

Active Inference is nowadays more often used in a search-related environ- ment, thus this research tests whether it is as e ective as other algorithms such as a No Detection algorithm. This is tested in a 10 by 10 grid, where it tries to nd a target most optimally by minimizing the Variational Free Energy (VFE) and Expected Free Energy (EFE). Bayesian Inference will be applied to solve these minimization problems. The agent uses a momentum- cone in order to choose the most optimal action. Results show a de cit in information gain for the Active Inference compared to the No Detection algorithm. This could be due to an error made in computing the VFE for the agent, as well as an optimization error in the computation for the momentum-cone. Further research can be helpful in nding whether xing these errors will improve the algorithm.

Description

Citation

Faculty

Faculteit der Sociale Wetenschappen