Seite - 131 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

(a) Overview ofhardware setup (b) Human-robot collaboration tobuild aTangram puzzle Figure1.Overviewofhardwaresetupforpuzzle solving (e.g., industrial applications, entertainment applications, service applications, assistive technology applications). Concerning the applications of choice, it was decided that at least one ”set of use cases” should be chosen for which it was possible to carry them throughout all possible categories (A to D plus optionally E and F) by continuously extending and adding ”human-robot collaboration features”. Theusecasesetofchoicefor thispurpose is tosolveaTangrampuzzlewhile therobotand thehumanarecooperating toward thegoal (seeFigure1(b)). Furtherusecases, incollaborationwith industrialpartners, address industrialapplicationssuchashuman-robot jointassemblyandinspection tasks. 4. AddressedResearchChallenges 4.1. DynamicWorkingEnvironmentMonitoringandSafety In order to perform a safe and reliable collaborative task in complex and dynamic environments, a robust monitoring system, which provides a real-time status of the target area, is required. Methods from2D[14]and3D[24]quality inspection (andcombined [29]) canbeapplied to robotic scenarios as long as sensors are lightweight and compact enough to be used on moving parts of the robot. 2D inspectionissufficient incaseofclose-to-planarobjects. 3Dshapecomparisone.g.,byIterativeClos- est Point (ICP) [23], is a robust method in industrial inspection for irregular objects. Texture-based methods are complementary in case of smooth 3D surfaces that do not allow a precise 3D alignment due to shape ambiguities [5]. One major current application for such techniques is the inspection of correct 3D shape, the automatic planning of grasping is still a very rare application under real pro- duction conditions. Within CollRob, we use various optical sensors for localization, detection, 3D reconstruction, and depth estimation. The overall system has to cope with challenges such as sub- millimeter position accuracy for robot grasping, occlusion, surface reflections and complex shapes. 131