Morphological evolution for pipe inspection using ROS

Inmanypracticalapplicationssuchaswater/oil/gasdistributionsystems(e.g. thoseusedinmetallurgyplants),continuous monitoringof thestatusof thepipes isneededtopreventleaksandotherkindsofmalfunctioningofthesystem.However,pipe systemsaretypicallydifficulttoaccess,andinspectingthemmightrequireacompleteinterruptionoftheoperationsofthepipes. An alternative solution is to usemm-sized passive sensor agents,whichareable to inspect thepipeswhile theyare still operating​. However, these agents are typically kinetically passive due to their size. Consequently, controlling the agent behaviour inaparticularenvironment ischallengingduetothecomplexityofthemathematicalrepresentationoftheproblemat hand (e.g. agent-environment interaction). Inthisregard,black-boxoptimizationtechniquessuchasEvolutionaryAlgorithms (EAs)suitetheproblem. Inthiswork,wepresentamethodologytoevolvethemorphologicalpropertiesofanagentforagiven behaviouralobjectiveinapipescenario.Asaproof-of-concept,wesetthemotionbehaviourobjectivetomovingacrossagiven pipeasfastaspossible, fordifferent insertionpoints inthepipe.Asfor theevolvablemorphologyaspect,weassumethatthe agenthaseightcompartments thatcanbefilledwithweights.Thus,ourcontributionisthreefold:1)Designinganddeveloping an evolutionary scheme for evolving the agent’s morphology tomove across a pipe as fast as possible. This includes genotype-phenotype representation, reproductionoperators, selectioncriteriaandfitnessfunction.2)Developingasimulation framework for evaluating theproposedsolution, usingRobotOperatingSystem(ROS), and3)Testingthewholeevolution scheme, includingsimulation,ontwodifferentpipeconfigurations.Thisproof-of-conceptworkisdesignedtotakepart inthe evaluation process of the Phoenix project, whichaims toexploreunknownenvironmentsviaevolvable sensor agents [1]. However,weenvisionthisapproachtobegenerallyhelpfulinwater/oil/gasdistributionsystems,whererealisticobjectivesother than speed must be set.