University of Limerick
Browse

Comprehensive Analysis of Learning Cases in an Autonomous Navigation Task for the Evolution of General Controllers

Download (2.2 MB)
journal contribution
posted on 2023-03-09, 09:55 authored by ENRIQUE NAREDOENRIQUE NAREDO, Candelaria Sansores, Flaviano Godinez, Francisco Lopez, Paulo C. M. Urbano, Leonardo Trujillo, Conor RyanConor Ryan

Robotics technology has made significant advancements in various fields in industry and society. It is clear how robotics has transformed manufacturing processes and increased productivity. Additionally, navigation robotics has also been impacted by these advancements, with investors now investing in autonomous transportation for both public and private use. This research aims to explore how training scenarios affect the learning process for autonomous navigation tasks. The primary objective is to address whether the initial conditions (learning cases) have a positive or negative impact on the ability to develop general controllers. By examining this research question, the study seeks to provide insights into how to optimize the training process for autonomous navigation tasks, ultimately improving the quality of the controllers that are developed. Through this investigation, the study aims to contribute to the broader goal of advancing the field of autonomous navigation and developing more sophisticated and effective autonomous systems. Specifically, we conducted a comprehensive analysis of a particular navigation environment using evolutionary computing to develop controllers for a robot starting from different locations and aiming to reach a specific target. The final controller was then tested on a large number of unseen test cases. Experimental results provide strong evidence that the initial selection of the learning cases plays a role in evolving general controllers. This work includes a preliminary analysis of a specific set of small learning cases chosen manually, provides an in-depth analysis of learning cases in a particular navigation task, and develops a tool that shows the impact of the selected learning cases on the overall behavior of a robot’s controller. 

Funding

14271.22-P

Automatic Design of Digital Circuits (ADDC)

Science Foundation Ireland

Find out more...

History

Publication

Mathematical Computational Applications 28(2), 35

Publisher

MDPI

Department or School

  • Computer Science & Information Systems

Usage metrics

    University of Limerick

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC