279 research outputs found
Multimodal Manipulation Under Uncertainty (Dagstuhl Seminar 15411)
No full textThis report documents the program and the outcomes of Dagstuhl Seminar
15411 "Multimodal Manipulation Under Uncertainty". The seminar was
organized around brief presentations designed to raise questions and
initiate discussions, multiple working groups addressing specific
topics, and extensive plenary debates. Section 3 reproduces abstracts of brief presentations, and Section 4 summarizes the results of the working groups
Modeling Human Motor Skills to Enhance Robots’ Physical Interaction
No full textThe need for users’ safety and technology acceptability has incredibly increased with the deployment of co-bots physically interacting with humans in industrial settings, and for people assistance. A well-studied approach to meet these requirements is to ensure human-like robot motions and interactions. In this manuscript, we present a research approach that moves from the understanding of human movements and derives usefull guidelines for the planning of arm movements and the learning of skills for physical interaction of robots with the surrounding environment.Green Open Access added to TU Delft Institutional Repository 'You share, we take care!' - Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.Learning & Autonomous Contro
A Visual Intelligence Scheme for Hard Drive Disassembly in Automated Recycling Routines
No full textAs the state-of-the-art deep learning models are taking the leap to generalize and leverage automation, they are becoming useful in real-world tasks such as the disassembly of devices by robotic manipulation. We address the problem of analyzing the visual scenes on industrial-grade tasks, for example, automated robotic recycling of a computer hard drive with small components and little space for manipulation. We implement a supervised learning architecture combining deep neural networks and standard point cloud processing for detecting and recognizing hard drives parts, screws, and gaps. We evaluate the architecture on a custom hard drive dataset and reach an accuracy higher than 75% in every component used in our pipeline. Additionally, we show that the pipeline can generalize on damaged hard drives. Our approach combining several specialized modules can provide a robust description of a device usable for manipulation by a robotic system. To our knowledge, we are the pioneers to offer a complete scheme to address the entire disassembly process of the chosen device. To facilitate the pursuit of this issue of global concern, we provide a taxonomy for the target device to be used in automated disassembly scenarios and publish our collected dataset and code
A Simple Ontology of Manipulation Actions Based on Hand-Object Relations
No full textHumans can perform a multitude of different actions with their hands (manipulations). In spite of this, so far there have been only a few attempts to represent manipulation types trying to understand the underlying principles. Here we first discuss how manipulation actions are structured in space and time. For this we use as temporal anchor points those moments where two objects (or hand and object) touch or un-touch each other during a manipulation. We show that by this one can define a relatively small tree-like manipulation ontology. We find less than 30 fundamental manipulations. The temporal anchors also provide us with information about when to pay attention to additional important information, for example when to consider trajectory shapes and relative poses between objects. As a consequence a highly condensed representation emerges by which different manipulations can be recognized and encoded. Examples of manipulations recognition and execution by a robot based on this representation are given at the end of this study.European Community [270273, 269959
Sampling-based multiview reconstruction without correspondences for 3D edges
No full textThis paper introduces a novel method for featurebased 3D reconstruction using multiple calibrated 2D views. We use a probabilistic formulation of the problem in the 3D, reconstructed space that allows using features that cannot be matched one-to-one, or which cannot be precisely located, such as points along edges. The reconstructed scene, modelled as a probability distribution in the 3D space, is defined as the intersection of all reconstructions compatible with each available view. We introduce a method based on importance sampling to retrieve individual samples from that distribution, as well as an iterative method to identify contiguous regions of high density. This allows the reconstruction of continuous 3D curves compatible with all the given input views, without establishing specific correspondences and without relying on connectivity in the input images, while accounting for uncertainty in the input observations, due e.g. to noisy images and poorly calibrated cameras. The technical formulation is attractive in its flexibility and genericity. The implemented system, evaluated on several very different publicly-available datasets, shows results competitive with existing methods, effectively dealing with arbitrary numbers of views, wide baselines and imprecise camera calibrations.Damien Teney, Justus Piate
Continuous pose estimation in 2D images at instance and category levels
No full textWe present a general method for tackling the related problems of pose estimation of known object instances and object categories. By representing the training images as a probability distribution over the joint appearance/pose space, the method is naturally suitable for modeling the appearance of a single instance of an object, or of diverse instances of the same category. The training data is weighted and forms a generative model, the weights being based on the informative power of each image feature for specific poses. Pose inference is performed through probabilistic voting in pose space, which is intrinsically robust to clutter and occlusions, and which we render tractable by treating separately the least interdependent dimensions. The scalability of category-level models is ensured during training by clustering the available image features in the joint appearance/pose space. Finally, we show how to first efficiently use a category-model, then possibly recognize a particular trained instance to refine the pose estimate using the corresponding instance-specific model. Our implementation uses edge points as image features, and was tested on several existing datasets. We obtain results on par with or superior to state-of-the-art methods, on both instance- and category-level problems, including for generalization to unseen instances.Damien Teney, Justus Piate
Safe Autonomous Reinforcement Learning
No full textTechnika posilovaného učení již nesčetněkrát prokázala svou užitečnost v robotice a dalších aplikacích strojového učení. Dovoluje učit strategie řízení robotů bez přesné znalosti, jaká akce je ve kterém stavu ideální. K nalezení optimální strategie stačí dodat funkci užitku a několikrát systém spustit.Reinforcement Learning is a technique proven by uncountable use-cases in the robotics community and many other machine-learning fields. It allows training optimal decision policies without knowing precisely which actions are the best at any given moment. A reward function and some number of policy rollouts suffice to estimate the best decision policy
Generalized exemplar-based full pose estimation from 2D images without correspondences
No full textThis paper addresses the problem of full pose estimation of objects in 2D images, using registered 2D examples as training data. We present a general formulation of the problem, which departs from traditional approaches by not focusing on one specific type of image features. The proposed algorithm avoids relying on specific model-to-scene correspondences, allowing using similar-looking and generally unmatchable features. We effectively demonstrate this capability by applying the method to edge segments. Our algorithm uses successive histogrambased and probabilistic evaluations, which ultimately recover a complete description of the probability distribution of the pose of the object, in the 6 degree-of-freedom 3D pose space, thereby accounting for the inherent ambiguities in the 2D input data. Furthermore, we propose, in a rigorous framework, an efficient procedure for fusing multiple sources of evidence, such as multiple registered 2D views of the same scene. The proposed method is evaluated qualitatively and quantitatively on synthetic and real test images. It shows promising results under challenging conditions, including occlusions and heavy clutter, while being capable of handling objects with little texture and detail.Damien Teney, Justus Piate
- …
