Industrial robot manipulators are widely used for repetitive applications that require high precision, like pick-and-place. In many cases, the movements of industrial robot manipulators are hard-coded or manually defined, and need to be adjusted if the objects being manipulated change position. To increase flexibility, an industrial robot should be able to adjust its configuration in order to grasp objects in variable/unknown positions. This can be achieved by off-the-shelf vision-based solutions, but most require prior knowledge about each object tobe manipulated. To address this issue, this work presents a ROS-based deep reinforcement learning solution to robotic grasping for a Collaborative Robot (Cobot) using a depth camera. The solution uses deep Q-learning to process the color and depth images and generate a greedy policy used to define the robot action. The Q-values are estimated using Convolutional Neural Network (CNN) based on pre-trained models for feature extraction. Experiments were carried out in a simulated environment to compare the performance of four different pre-trained CNNmodels (RexNext, MobileNet, MNASNet and DenseNet). Results showthat the best performance in our application was reached by MobileNet,with an average of 84 % accuracy after training in simulated environment.
Anticipatory force planning during grasping is based on visual cues about the object’s physical properties and sensorimotor memories of previous actions with grasped objects. Vision can be used to estimate object mass based on the object size to identify and recall sensorimotor memories of previously manipulated objects. It is not known whether subjects can use density cues to identify the object’s center of mass (CM) and create compensatory moments in an anticipatory fashion during initial object lifts to prevent tilt. We asked subjects (n=8) to estimate CM location of visually symmetric objects of uniform densities (plastic or brass, symmetric CM) and non-uniform densities (mixture of plastic and brass, asymmetric CM). We then asked whether subjects can use density cues to scale fingertip forces when lifting the visually symmetric objects of uniform and non-uniform densities. Subjects were able to accurately estimate an object’s center of mass based on visual density cues. When the mass distribution was uniform, subjects could scale their fingertip forces in an anticipatory fashion based on the estimation. However, despite their ability to explicitly estimate CM location when object density was non-uniform, subjects were unable to scale their fingertip forces to create a compensatory moment and prevent tilt on initial lifts. Hefting object parts in the hand before the experiment did not affect this ability. This suggests a dichotomy between the ability to accurately identify the object’s CM location for objects with non-uniform density cues and the ability to utilize this information to correctly scale their fingertip forces. These results are discussed in the context of possible neural mechanisms underlying sensorimotor integration linking visual cues and anticipatory control of grasping.
LINK
Studying images in social media poses specific methodological challenges, which in turn have directed scholarly attention towards the computational interpretation of visual data. When analyzing large numbers of images, both traditional content analysis as well as cultural analytics have proven valuable. However, these techniques do not take into account the circulation and contextualization of images within a socio-technical environment. As the meaning of social media images is co-created by networked publics, bound through networked practices, these visuals should be analyzed on the level of their networked contextualization. Although machine vision is increasingly adept at recognizing faces and features, its performance in grasping the meaning of social media images is limited. However, combining automated analyses of images - broken down by their compositional elements - with repurposing platform data opens up the possibility to study images in the context of their resonance within and across online discursive spaces. This paper explores the capacities of platform data - hashtag modularity and retweet counts - to complement the automated assessment of social media images; doing justice to both the visual elements of an image and the contextual elements encoded by networked publics that co-create meaning.
MULTIFILE
In greenhouse horticulture harvesting is a major bottleneck. Using robots for automatic reaping can reduce human workload and increase efficiency. Currently, ‘rigid body’ robotic grippers are used for automated reaping of tomatoes, sweet peppers, etc. However, this kind of robotic grasping and manipulation technique cannot be used for harvesting soft fruit and vegetables as it will cause damage to the crop. Thus, a ‘soft gripper’ needs to be developed. Nature is a source of inspiration for temporary adhesion systems, as many species, e.g., frogs and snails, are able to grip a stem or leave, even upside down, with firm adhesion without leaving any damage. Furthermore, larger animals have paws that are made of highly deformable and soft material with adjustable grip size and place holders. Since many animals solved similar problems of adhesion, friction, contact surface and pinch force, we will use biomimetics for the design and realization of the soft gripper. With this interdisciplinary field of research we aim to model and develop functionality by mimicking biological forms and processes and translating them to the synthesis of materials, synthetic systems or machines. Preliminary interviews with tech companies showed that also in other fields such as manufacturing and medical instruments, adjustable soft and smart grippers will be a huge opportunity in automation, allowing the handling of fragile objects.
