Skip to main content
Fachgebiete
Automobil + Motoren Bauwesen + Immobilien Business IT + Informatik Elektrotechnik + Elektronik Energie + Nachhaltigkeit Finance + Banking Management + Führung Marketing + Vertrieb Maschinenbau + Werkstoffe Versicherung + Risiko
DE
EN
Bücher
Zeitschriften
Themenseiten
Organisationspsychologie Projektmanagement Marketing Smart Manufacturing
Weitere Formate
Podcasts Webinare Technik Webinare Wirtschaft Kongresse Veranstaltungskalender Awards
Jetzt Einzelzugang starten
Zugang für Unternehmen
Referenzkunden
SLX-Digitalkonferenz: Zukunftswerkstatt 2024

Mehr Umsatz mit Top-Kontakten

Am 7. Mai erfahren Sie, wie Vertriebsteams Leadgenerierung, Leadnurturing und Leadstrategien effizient steuern können.
Erweiterte Suche
Anmelden
nach oben

2013 | Buch

Introduction Kapitel lesen Erstes Kapitel lesen

Semantic 3D Object Maps for Everyday Robot Manipulation

verfasst von: Radu Bogdan Rusu

Verlag: Springer Berlin Heidelberg

Buchreihe : Springer Tracts in Advanced Robotics

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Einloggen, um Zugang zu erhalten
insite
SUCHEN

Über dieses Buch

The book written by Dr. Radu B. Rusu presents a detailed description of 3D Semantic Mapping in the context of mobile robot manipulation. As autonomous robotic platforms get more sophisticated manipulation capabilities, they also need more expressive and comprehensive environment models that include the objects present in the world, together with their position, form, and other semantic aspects, as well as interpretations of these objects with respect to the robot tasks.

The book proposes novel 3D feature representations called Point Feature Histograms (PFH), as well as a frameworks for the acquisition and processing of Semantic 3D Object Maps with contributions to robust registration, fast segmentation into regions, and reliable object detection, categorization, and reconstruction. These contributions have been fully implemented and empirically evaluated on different robotic systems, and have been the original kernel to the widely successful open-source project the Point Cloud Library (PCL) -- see http://pointclouds.org.

Inhaltsverzeichnis

Frontmatter
Introduction
Abstract
The population in Europe and many other countries is aging — probably causing the share of people aged 65 years or over to double within the next 40 years. As life expectancy increases, so does the chance of people becoming physically and cognitively limited or disabled, often leading to care-dependency. In the same way as the number of people requiring care is projected to increase is the number of people able to give care expected to decrease. This societal development forces us to develop new concepts, including technologies, for promoting independent living. Prolonging the independence of elderly people with minor disabilities and increasing their participation in daily life is expected to improve the well-being and the health-state of these people and thereby also mitigate the care-giving problem.
Radu Bogdan Rusu

Semantic 3D Object Mapping Kernel 15

Frontmatter
3D Map Representations
Abstract
The most primitive data representation unit in a three-dimensional Euclidean space is the 3D point itself, p i . This chapter gives a basic overview on acquisition techniques for 3D points, as well as on representation formats for collections of points in the context of mapping with mobile robots. Figure 2.1 presents examples of very accurate collections of 3D points representing objects found in indoor environments.
Radu Bogdan Rusu
Mapping System Architectures
Abstract
To use a well known metaphor, designing the architecture of a 3D perception system can be anything from a walk in the park to a day of fishing. This chapter takes on the challenge of proposing a comprehensive system architecture for the Semantic 3D Object Mapping kernel used in this book. Given the scenario laid out in the introduction, we are investigating the following computational problem: from a set of point cloud representations, each resulting from 3D laser scans acquired in an indoor environment, say a kitchen, automatically compute an integrated semantic representation of the data, with cupboards and drawers represented as cuboid containers with front doors and handles, tables and shelves as horizontal planes, and objects of interest classified and categorized based on their properties and the underlying geometric surface that they represent. Since this represents nothing else but the “holy grail” of 3D mapping systems from a data interpretation point of view (in the context of our application scenario), the resultant system architecture would need to comprise a big list of geometric and learning steps, where the output of a step requires the input of another, and so on.
Radu Bogdan Rusu
3D Point Feature Representations
Abstract
In their native representation, points as defined in the concept of 3D mapping systems are simply represented using their Cartesian coordinates x, y, z, with respect to a given origin. Assuming that the origin of the coordinate system does not change over time, there could be two points p 1 and p 2, acquired at t 1 and t 2, having the same coordinates. Comparing these points however is an ill-posed problem, because even though they are equal with respect to some distance measure (e.g. Euclidean metric), they could be sampled on completely different surfaces, and thus represent totally different information when taken together with the other surrounding points in their vicinity. That is because there are no guarantees that the world has not changed between t 1 and t 2. Some acquisition devices might provide extra information for a sampled point, such as an intensity or surface remission value, or even a color, however that does not solve the problem completely and the comparison remains ambiguous.
Radu Bogdan Rusu
From Partial to Complete Models
Abstract
This chapter presents mechanisms for the integration of partial datasets acquired from different views into a consistent global model. In particular, we will address the problems of: i) point cloud registration, to transform and align scans into the same coordinate system, and ii) data resampling, to create smooth datasets and filter low frequency point outliers.
Radu Bogdan Rusu
Clustering and Segmentation
Abstract
Storing and processing large point cloud datasets represents one of the main bottlenecks of a 3D perception system. Though algorithms that can process individual data frames with low computational resources can be written, their capabilities are hindered as larger quantities of data accumulate. A simple example in this direction can be given by the problem of estimating the best planar model that represents a set of points \(\mathcal{P}\).
Radu Bogdan Rusu

Mapping of Indoor Environments

Frontmatter
Static Scene Interpretation
Abstract
This chapter reports on experiences regarding the acquisition of hybrid Semantic 3D Object Maps for indoor household environments, in particular kitchens, out of sensed 3D point cloud data. The proposed approach includes a comprehensive pipeline, with geometric mapping and learning mechanisms, for processing large input datasets and for extracting relevant objects useful for a personal robotic assistant to perform complex manipulation tasks. The type of objects modeled are those which perform utilitarian functions in the environment such as kitchen appliances, cupboards, tables, and drawers. The resultant model is accurate enough to use it in physics-based simulations, where doors of 3D containers can be opened based on their hinge position. The resultant map is represented as a hybrid concept and is comprised of hierarchically classified objects and triangular meshes used for collision avoidance in manipulation routines.
Radu Bogdan Rusu
Surface and Object Class Learning
Abstract
Segmenting and interpreting the surrounding environment that a personal robot operates in from sensed 3D data is an important research topic. Besides recognizing a certain location on the map for localization or map refinement purposes, obtaining accurate and informative object models is essential for precise manipulation and grasping. Although the acquired data is discrete and represents only a few samples of the underlying scanned surfaces, it quickly becomes expensive to store and work with. It is therefore imperative to find ways to address this dimensionality problem and group clusters of points sampled from surfaces with similar geometrical properties together, or in some sense try to “classify the world”. Achieving the latter annotates sensed 3D points and geometric structures with higher level semantics and greatly simplifies research in the aforementioned topics (e.g. manipulation and grasping).
Radu Bogdan Rusu
Parametric Shape Model Fitting
Abstract
This chapter tackles the problem of scene interpretation, including object reconstruction for everyday manipulation activities in domestic human living environments, in particular kitchens, out of point cloud data. The proposed framework takes raw point cloud datasets as input, and segments horizontal planar areas that support objects used in manipulation scenarios (see Figure 9.1).
Radu Bogdan Rusu

Applications

Frontmatter
Table Cleaning in Dynamic Environments
Abstract
Personal robots are coming to age, enabling a multitude of problem scenarios that were unsolvable until recently, to be tackled. One of the remaining challenges however is to design systems that can function in the presence of humans without harming them, that is, in general operate safely in the presence of environment dynamics.
Radu Bogdan Rusu
Identifying and Opening Doors
Abstract
An important challenge for autonomous personal robots is to be able to enter a human living environment and function in it, that is, find ways of navigating and interacting with the world in an effective manner. This means the robot should be capable of going anywhere where it can physically fit, where it would be able to find energy sources and recharge its batteries when their capacity is running low, and in general it must be able to do useful things such as cleaning tables. These behaviors need the support of complex perception routines which can recognize power plugs, certain structures and objects in the world, etc. Some of these environment structures, such as fixtures (handles and knobs) on doors and pieces of furniture, are of key importance for the robot’s performance. Any robot that will operate indoors must be able to correctly locate these fixtures and open doors to be able to better carry out different tasks. Since the robot may have to function in a wide variety of environments under varying lighting conditions, a robust door detection ability is essential for the robot.
Radu Bogdan Rusu
Real-Time Semantic Maps from Stereo
Abstract
Creating rich, meaningful map representations of the world from stereo data poses two interesting questions: i) given that stereo depth data is traditionally not as precise as laser measurements, what is the level of detail that a stereo-based semantic map could have; and ii) how can a mapping system able to process the large quantities of data that stereo cameras produce (e.g. frame rates of 30 fps) in an efficient real-time manner should be designed.
Radu Bogdan Rusu
Conclusion
Abstract
As autonomous personal robotics systems come to age, they need to find ways of functioning in human living environments, that is, navigating and interacting with the surrounding world in an effective manner. This means a robot should be capable of going anywhere where it can physically fit, find energy sources and recharge its batteries when their capacity is running low, and perform useful chores such as cleaning tables. These behaviors require support from complex perception routines which can recognize power plugs, door handles, and certain structures and objects in the world. Therefore one of the most important aspects deciding on the applicability of personal robotics in real world environments is represented by the robot’s ability to perceive and understand the semantic meanings of the objects present in the world. The key challenge in achieving that is being able to build semantic 3D perception architectures that can enable complex mobile manipulation scenarios.
Radu Bogdan Rusu
Backmatter
Metadaten
Titel
Semantic 3D Object Maps for Everyday Robot Manipulation
verfasst von
Radu Bogdan Rusu
Copyright-Jahr
2013
Verlag
Springer Berlin Heidelberg
Electronic ISBN
978-3-642-35479-3
Print ISBN
978-3-642-35478-6
DOI
https://doi.org/10.1007/978-3-642-35479-3

Neuer Inhalt

    Bildnachweise