Navigational algorithms evaluation and benchmarking

One of the fundamental problems in mobile robotics is navigating unexplored environments safely and efficiently. Efforts to address this issue are classified into three categories: reactive-based approaches, which make instantaneous decisions; map-based approaches, involving grid or topological representations; and learning-based approaches. Evaluating and comparing approaches is essential to better understand them, particularly in how they perform in different problem environments and in relation to each other. This information serves to guide the development of further approaches, highlight problem environments, and provide a clear mapping between approaches and environments. However, current comparative studies within a single category have been limited by the existence of a degree of similarity between the different approaches. There has not yet been a comparative framework across different categories in navigational robotics. Thus, this work aims to develop an evaluation method to compare a variety of different approaches in the same environment to achieve a better understanding of navigational algorithms. To this end, a framework has been proposed that simulates these approaches in a common set of problem environments and evaluates them with the same set of metrics to compare their effectiveness and efficiency. The most common reactive and map-based approaches are implemented and a generic, precise, and empirical way to evaluate their performance to the set of environments they are in and compared to the other different approaches is demonstrated. The resulting analysis shows that methods like RRT* don’t improve on the RRT when benchmarked and the evaluation of the problem areas of the Potential Field approach led to the development of the novel Pheromone Potential Field approach. This work opens the doors to more in-depth research into benchmarking across the different navigational categories in static and dynamic environments, which will result in a better understanding and significantly impact the future development of navigational approaches. This research is a step toward dynamic navigational planners that match the different approaches to a set of problems or environments.

Grantor dc:publisher

Oxford Brookes University

Year dc:date

2023

Author dc:creator

• Idries, Mohamed Osama

Contributors dc:contributor

• Olde Scheper, Tjeerd V.
• Rolf, Matthias

Statement dc:rights

• All rights reserved

Language dc:language

en

DOI dc:identifier

https://doi.org/10.24384/w8me-g696

OAI identifier oai:identifier

tle:bab75dda-c9ba-4356-a8e4-de2af3d67d21:d6bd9758-527a-46cd-bfe2-c433766e8fca:1