Context-Aware Active Inference for Robot Navigation
Keywords
Loading...
Authors
Issue Date
2022-01-01
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
In neuroscience, perception-based pose estimation has given an insightful approach
to further discovering human functions used for localization and navigation
using sensory information. This paper seeks to draw inspiration from this by
implementing an algorithm that performs sensory-based localization and navigation
by minimizing variational free energy under the free energy principle. This
is done by minimizing the difference between predicted sensory observations, obtained
through a generative model of an environment, and actual observations.
We expand on this algorithm by adding a classifier that maps sensory observations
to environment types. This gives the ability to switch between different generative
models for different types of environments which increases overall adaptability
and flexibility. By implementing the algorithm in a virtual robot, we can evaluate
the localization and navigation capabilities of the suggested model. Results show
that the method posed in this paper is a proof-of-concept of context-aware robotic
localization and navigation under the free energy principle.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen