Page - 27 - in Joint Austrian Computer Vision and Robotics Workshop 2020

AutomaticOntology-basedPlanGeneration foranIndustrialRoboticsSystem TimonHoebert,WilfriedLepuschitz,MunirMerdan PracticalRobotics InstituteAustria {hoebert,lepuschitz,merdan}@pria.at Abstract. Programming and re-configuration of robots are associated with high costs, especially for small- and medium-sized enterprises. We present an ontology-driven solution that can automate the con- figuration as well as the generation of process plans and schedules thereby significantly lowering the ef- forts in thecaseof changes. Thepresentedapproach is demonstrated in a laboratory environment with an industrialpilot test case. 1. Introduction Robotics technology, which can prove high effi- ciency, precision, and repeatability, is regarded as a viable solution to cope with the increasing number of individualized products. However, robot systems still often do not meet the demands of small- and medium-sized enterprises (SMEs) [8]. Especially, since the programming of industrial robots is com- plex and time-consuming. To be able to dynami- cally adapt to new products, robotic systems need to work autonomously. Autonomous systems, in this context,meansthatrobotssystemscanperformhigh- level taskspecificationswithoutexplicitlybeingpro- grammed [2]. To reach specific goals, such systems shouldbeable to receivegoalsandautomatically se- quenceplansandexecute themconsidering theircur- rent state. In our previous work, we presented the control architecture for industrial robots, which can generate actions based on an product model by link- ing product model, manufacturing process, and pro- ductionenvironmentinanontology[7]. Inthispaper, we focus on the automated plan generation from the ontology and present an approach for flexibly cou- pling of the decision-making mechanism and ontol- ogy. In section 2, we will detail the architecture and im- plementation. Finally, Section 3 concludes the paper with a summary and an outlook on further research issues. 2.Architecture The industrial robot control layer responsible for the management of the robotics systems consists of a World-Model and a Decision-Making component. The decision-making mechanism (Planner) acts as a link between the semantic model of the produc- tion environment and the available robot system ca- pabilities. The World Model contains the semantic representation of the relevant objects in the robotics system including their properties and relations. The Planning Domain Definition Language (PDDL) is used for decision-making and the world model is conceptually defined using the Web Ontology Lan- guage(OWL)standard. In thiscontext,wetransform robotics domain knowledge represented in OWL to PDDL as a targeted mechanism for planning. Multi- ple applied robotic systems use PDDL for task plan- ning and a lot of work has been done in combining ontologies and AI planning base [5, 1]. Especially ROSPlan[4], aROSimplementation, is acommonly used implementation for this purpose. Based on ROSPlan, OWL-ROSPlan [3] extends this approach using a specialized OWL-Ontology as knowledge- base instead of traditional databases. The disadvan- tage of these approaches is the implementation ef- fort for application. Even OWL-ROSPlan requires a predefined data format of the ontology. Our work extends this research by automating the translation of the input required by PDDL from the ontology as well as fromthePDDLback to theontologywithout anypredefinedontology formats. 2.1.OWL-PDDLMappingscheme ThebasicbuildingblocksofOWL,aretriplescon- sisting of subject, predicate, and object. The ba- sic building blocks of PDDL are actions and PDDL- predicates. To avoid confusion of the two different 27