Page - 28 - in Joint Austrian Computer Vision and Robotics Workshop 2020

:vGripper2 r:isFree “true”^^xsd:boolean. :part23 ro:movable :vGripper2. :part23 r:connects :part54, :part64. (r:connects :part23 :part54 :part64) (r:movable :part23 :vGripper2) (r:isFree :vGripper2) subject predicate object1 subject(predicate [object1] [, objectN]. [objectN]) 3... ... Figure 1. Mapping OWL triples with PDDL predicates. Also, three examples of different parameter length are shown. types of predicates, the latter ones are only referred to as PDDL-Predicates. The general idea of this ap- proachis theequalizationofbothbuildingblocks, re- lating triples with PDDL-predicates. Using a similar approach like WebPDDL [6], OWL-IRIs are used as PDDL-predicatenamestoidentifythedatadistinctly. OWL-predicates relate subjects and objects, as verbs do in sentences, but PDDL-predicates are only binarystatements relating tomultipleobjectparame- ters. In practice, PDDL-predicates usually only have one or two object parameters, which can be seen as subjectandobject. Thecompletemappingschemeis illustrated with three examples in Figure 1. PDDL- predicates with only one parameter are mapped to boolean-valued objects triples. In practice, PDDL- predicates with more than two parameters are rare because of their complexity (only 4 percent of all predicates from all IPC (1998-2018) domains. But, even these PDDL-predicates can be simplified to multiple PDDL-predicateswith twoparameters. 2.2.SemanticPDDLGeneration The system automatically generates the PDDL3- problem for the planner based on the information in the ontology and PDDL-domain. This enables easy and extensible programming of the system. Theuser only has to specify the PDDL-domain with IRIs as PDDL-predicate names and add the goal as triples into a separate part (separate graph) of the ontol- ogy database. The system automatically queries all triples of NamedIndividuals regarding this pred- icates, maps them to PDDL-predicates as mentioned earlierandaddsthemtotheinitsectioninthePDDL- problem. These queries are executed in parallel, and the particular subjects are recorded. After query- ing the triples, the OWL-types of the recorded sub- jectsaresearchedintheontologyandwrittenintothe PDDL-problem. Since each NamedIndividual can have multiple parent-classes, but not all are relevant forplanning,only theoneswhicharespecified in the PDDL-domainareused. 3.Conclusion The proposed knowledge-driven approach simpli- fies the programming efforts of the industrial robot. The code for the industrial implementation is gener- ated automatically based on the defined rules, states and actions. A system engineer only needs to de- scribe the functionality of the assembly line or char- acteristics of the product to be assembled, without having to consider further engineering issues. In our application, we successfully used the developed mechanism for planning pick-and-place operations ofanindustryrobotbyKukaaswellas theFestopor- talrobot,whenjointlyappliedforassemblingofPCB boards. As future work, we aim to consider product assemblytasksinvolvingmorecomplexproductsand production layouts. References [1] S. Balakirsky and Z. Kootbally. An Ontology Based Approach to Action Verification for Agile Manufac- turing, pages 201–217. Springer International Pub- lishing,Cham,2014. [2] G. A. Bekey. Autonomous Robots: From Biologi- cal Inspiration to Implementation and Control (Intel- ligent Robotics and Autonomous Agents). The MIT Press, 2005. [3] L. Buoncompagni, A. Capitanelli, and F. Mastrogio- vanni. Arosmulti-ontologyreferencesservices: Owl reasoners and application prototyping issues. arXiv preprintarXiv:1706.10151, 2017. [4] M. Cashmore, M. Fox, D. Long, D. Magazzeni, B. Ridder, A. Carrera, N. Palomeras, N. Hurtos, and M. Carreras. Rosplan: Planning in the robot operat- ingsystem. InTwenty-Fifth InternationalConference onAutomatedPlanningand Scheduling, 2015. [5] M.Crosby,R.P.A.Petrick,F.Rovida,andV.Kru¨ger. Integrating mission and task planning in an industrial robotics framework. In ICAPS, 2017. [6] D. Dou. The formal syntax and semantics of web- pddl. Technical report, Technical Report, Technical report, UniversityofOregon,2008. [7] T. Hoebert, W. Lepuschitz, E. List, and M. Merdan. Cloud-baseddigital twinforindustrial robotics. pages 105–116, Cham, 2019. Springer International Pub- lishing. [8] A. Perzylo, N. Somani, S. Profanter, I. Kessler, M. Rickert, and A. Knoll. Intuitive instruction of industrial robots: Semantic process descriptions for small lotproduction. 102016. 28