Publications Search Results

Now showing 1 - 10 of 52
  • Publication
    Autonomere und flexiblere Robotik. Maschinelles Lernen ermöglicht matrixfähige Kommissionierzelle
    Ein Kunde ruft an und möchte sobald wie möglich ein neues Ersatzteil produzieren lassen. An sich eine erfreuliche Nachricht - aber natürlich wissen Sie, dass die Maschinen die zugehörigen Werkstücke noch gar nicht kennen. Bisher ist es sehr aufwendig, die Produktionslinie entsprechend umzustellen. Die Matrixproduktion ermöglicht mehr Flexibilität und vereinfacht eine solche Umstellung deutlich. Das Fraunhofer IPA zeigt an einem Demonstrator, wie sich die Vereinzelung, Zuführung und Kommissionierung von Bauteilen matrixfähig umsetzen lässt.
  • Publication
    Automatic Grasp Pose Generation for Parallel Jaw Grippers
    This paper presents a novel approach for the automatic offline grasp pose synthesis on known rigid objects for parallel jaw grippers. We use several criteria such as gripper stroke, surface friction, and a collision check to determine suitable 6D grasp poses on an object. In contrast to most available approaches, we neither aim for the best grasp pose nor for as many grasp poses as possible, but for a highly diverse set of grasps distributed all along the object. In order to accomplish this objective, we employ a clustering algorithm to the sampled set of grasps. This allows to simultaneously reduce the set of grasp pose candidates and maintain a high variance in terms of position and orientation between the individual grasps. We demonstrate that the grasps generated by our method can be successfully used in real-world robotic grasping applications.
  • Publication
    Real-time Instance Detection with Fast Incremental Learning
    Object instance detection is a highly relevant task to several robotic applications such as automated order picking, or household and hospital assistance robots. In these applications, a holistic scene labeling is often not required whereas it is sufficient to find a certain object type of interest, e.g. for picking it up. At the same time, large and continuously changing object sets are characteristic in such applications, requiring efficient model update capabilities from the object detector. Today's monolithic multi-class detectors do not fulfill this criterion for fast and flexible model updates.This paper introduces InstanceNet, an ensemble of efficient single-class instance detectors capable of fast and incremental adaptation to new object sets. Due to a dynamic sampling-based training strategy, accurate detection models for new objects can be obtained within less than 40 minutes on a consumer GPU while only a small percentage of the existing detection models needs to be updated in a very efficient manner. The new detector has been thoroughly evaluated on the basis of a novel dataset of 100 grocery store objects.
  • Publication
    Automatic Grasp Generation for Vacuum Grippers for Random Bin Picking
    In random bin picking, grasps on a workpiece are often defined manually, which requires extensive time and expert knowledge. In this paper, we propose a method that generates and prioritizes grasps for vacuum and magnetic grippers by analyzing the CAD model of a workpiece and gripper geometry. Using projections of these models, heatmaps such as the overlap of gripper and workpiece, the center of gravity, and the surface smoothness are generated. To get a combined heatmap, which estimates the probability for a successful grip, all individual heatmaps are fused by means of a weighted sum. Grid-based sampling generates prioritized grasps and suggests the most suitable gripper automatically. This approach increases the autonomy of bin picking significantly.
  • Publication
    Separating Entangled Workpieces in Random Bin Picking using Deep Reinforcement Learning
    ( 2021) ;
    Kulig, Marco
    ;
    ; ;
    Roggendorf, Simon
    ;
    Petrovic, Oliver
    ;
    ;
    Entangled workpiece situations often occur in random bin picking of chaotically stored objects and are a common source of problem in the bin picking process. Previous methods for averting this problem, such as randomly shaking the gripper over the bin, lead to decreasing production efficiency and an increase in cycle time. A promising new strategy uses supervised learning and deep neural networks to learn the separation. However, this approach requires a large amount of labeled data. To overcome this issue, this paper proposes a deep reinforcement learning approach to separate entangled workpieces and to minimize the setup effort.
  • Publication
    Mobiler Serviceroboter für die Getränkelogistik
    Im Forschungsprojekt ""Luka-Beverage" entsteht ein Kl-basierter Handhabungsassistent, der beim Verräumen von Leergut und Getränkekisten unterstützen soll.
  • Publication
    Using Deep Neural Networks to Separate Entangled Workpieces in Random Bin Picking
    Entanglements can cause robots to pick multiple parts within random bin picking applications. Previous approaches cope with this problem by shaking the gripped workpiece above the bin. However, these methods increase the cycle time and may decrease the robustness of the application. Therefore we propose a new method to separate entangled workpiece situations by using deep supervised learning. To generate annotated training data for a convolutional neural network we set up a simulation scene. In this scene, bins are filled with different amounts of sorted workpieces in several entangled situations. Each workpiece is then moved into different directions to path poses which are evenly distributed along the surface of a hemisphere. The emerging dataset consists of cropped depth images of entangled workpiece situations and several path poses. A serial connection of convolutional neural networks is trained on this dataset and proposes a sequence of poses yielding the general departure path. Finally, the performance of this method is validated on simulated data. To the best of our knowledge, our proposed method is the first systematic approach to find the best extraction strategy to separate entangled workpieces in a pile while decreasing the effective cycle time for gripping entangled workpieces and increasing the robustness significantly.
  • Publication
    Investigations on Output Parameterizations of Neural Networks for Single Shot 6D Object Pose Estimation
    Single shot approaches have demonstrated tremendous success on various computer vision tasks. Finding good parameterizations for 6D object pose estimation remains an open challenge. In this work, we propose different novel parameterizations for the output of the neural network for single shot 6D object pose estimation. Our learning-based approach achieves state-of-the-art performance on two public benchmark datasets. Furthermore, we demonstrate that the pose estimates can be used for real-world robotic grasping tasks without additional ICP refinement.
  • Publication
    Kognitive Roboter für Logistik-Automation
    Ein Erfolgsfaktor zur weiteren Automation in Lager und Logistik sind kognitive Fähigkeiten für Roboter. Diese ermöglichen eine flexible und intelligente (Teil-)Automatisierung.
  • Publication
    Model-Free Grasp Learning Framework based on Physical Simulation
    ( 2020)
    Riedlinger, Marc A.
    ;
    ; ;
    Khalid, Muhammad Usman
    ;
    The work at hand presents a generic framework to build classifiers that allow to predict the quality of 6-DOF grasp candidates for arbitrary mechanical grippers based on the depth data captured by a depth sensor. Hereby, the framework covers the whole process of setting up a deep neural network for a given mechanical gripper by making use of synthetic data resulting from a new grasp simulation tool. Furthermore, a new extended convolutional neural network (CNN) architecture is introduced that estimates the quality of a suggested grasp candidate based on local depth information and the pose of the corresponding grasp. As a result, robust grasp candidates can be detected in a model-free fashion.