Page - 137 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

IndustrialGrasping -AnAutonomousOrderPicking System∗ Julia NitschandGeraldSteinbauer Institute forSoftwareTechnology Graz University ofTechnology,Austria {jnitsch,steinbauer}@ist.tugraz.at Abstract Automated storing, retrieving, anddelivering items is an important part of Industry 4.0 application. For low-volume this task is done usually manual. In this paper we present an architecture and a proof-of-concept implementation fororderpickingusing therobotBaxter fromRethinkRobotics. The maincontributionbesidesproviding full functioningprototype isadependablecontrolarchitecture. 1. Introduction Industry 4.0 is one of the keywords, when we talk about the next level of production. Industry 4.0 represents the4th industrial revolutionandpromises improvementofproductivity throughautomated, self-organizing and self-optimizing processes. It addresses the needs of high-quality products which arealsohighly customized but still ready formassproduction. This work contributes to the field of Industry 4.0 by developing an assistant robot for order picking. Such robots share the environments with humans. In a typical warehouse system items can be stored inlarger transportboxes. Thetransportboxesagaincanbestoredinshelvestosavespace. Ifaspecific itemneeds tobepickedthe transportboxfirstneeds tobepulledoutof theshelfandthenthe itemcan bepickedanddelivered. Thisprocedure iscalledorderpicking. For itemswithamoderate frequency this type of picking is usually done by hand which is a monotonic and time consuming task. In our scenariowe tend to automatize that task. The system we propose is based on a 3-TIER architecture. The planning layer uses an artificial intelligence(AI)planner togeneratea listof skills the robothas toexecute. Theplanneroutputsa list ofskills, therobotneedstoexecuteinorder toachieveitsgoal. Skillsarecomposedofskillprimitives. These primitives can perform perception, manipulation, grasping tasks or any combination of those. Failures are already detected at the level of the primitives where local recoveries can be performed. If these recoveries fail too, these errors are reported to the executive layer. This architecture ensures the detection and recognition of failures. Together with appropriate steps for recovery dependable execution is achieved. The proposed architecture was realized as a proof-of-concept implementation using the twoarmrobotBaxter fromRethinkRobotics. Fordetailsabout the realizedsystemwerefer the interested reader to [10]. The reminder of this paper is organized as follows. In the next sections we briefly discuss related research and the target environment. In Section 4. the proposed system architecture is presented. Due to the space constraints we focus on skill primitives. In the next section we briefly present an evaluation focused on the skill primitives. In section6.wedrawsomeconclusions. ∗This workwas supportedby incubed IT GmbH. 1 137