MazeGen: A Low-Code Framework for Bootstrapping Robotic Navigation Scenarios for Smart Manufacturing Contexts

Ivan Hugo Guevara, Tiziana Margaria

Research output: Contribution to journalArticlepeer-review

Abstract

In this research, we describe the MazeGen framework (as a maze generator), which generates navigation scenarios using Grammatical Evolution for robots or drones to navigate. The maze generator uses evolutionary algorithms to create robotic navigation scenarios with different semantic levels along a scenario profile. Grammatical Evolution is a Machine Learning technique from the Evolutionary Computing branch that uses a BNF grammar to describe the language of the possible scenario universe and a numerical encoding of individual scenarios along that grammar. Through a mapping process, it converts new numerical individuals obtained by operations on the parents’ encodings to a new solution by means of grammar. In this context, the grammar describes the scenario elements and some composition rules. We also analyze associated concepts of complexity, understanding complexity as the cost of production of the scenario and skill levels needed to move around the maze. Preliminary results and statistics evidence a low correlation between complexity and the number of obstacles placed, as configurations with more difficult obstacle dispositions were found in the early stages of the evolution process and also when analyzing mazes taking into account their semantic meaning, earlier versions of the experiment not only resulted as too simplistic for the Smart Manufacturing domain, but also lacked correlation with possible real-world scenarios, as was evidenced in our experiments, where the most semantic meaning results had the lowest fitness score. They also show the emerging technology status of this approach, as we still need to find out how to reliably find solvable scenarios and characterize those belonging to the same class of equivalence. Despite being an emerging technology, MazeGen allows users to simplify the process of building configurations for smart manufacturing environments, by making it faster, more efficient, and reproducible, and it also puts the non-expert programmer in the center of the development process, as little boilerplate code is needed.

Original languageEnglish
Article number2058
JournalElectronics (Switzerland)
Volume12
Issue number9
DOIs
Publication statusPublished - May 2023

Keywords

  • evolutionary computing
  • machine learning
  • navigation scenarios
  • smart manufacturing

Fingerprint

Dive into the research topics of 'MazeGen: A Low-Code Framework for Bootstrapping Robotic Navigation Scenarios for Smart Manufacturing Contexts'. Together they form a unique fingerprint.

Cite this