|
Andreas Nüchter - Publications |
| Jacobs University | EECS | COSYP | CASE | Legal |
|
|
Back to my short list! Check out my top 10 publications! A list of my major publications sorted by subject can be found here. There is also a list of my co-authors and a bibtex file. In Press
Books and Edited Proceedings
Book Chapters
Journal and Magazin Papers
Conference, Workshop and Symposium Papers
Extended Abstracts and Posters
Technical Reports, Thesis, Misc
|
| Last changed: 2011-11-25 |
Abstract:
Dieser Beitrag stellt eine neue Implementation der
Octree-Datenstruktur vor. Sie ermöglicht es, eine
Milliarde 3D-Punkte in 8 GB Hauptspeicher exakt zu
repräsentieren und effiziente Algorithmen zu
implementieren. Der Octree gibt eine Hierarchie vor, die
dazu verwendet werden kann, grosse Punktwolken zu
inspizieren und flüssig in ihnen zu navigieren. Des
Weiteren schlagen wir in diesem Artikel ein effizentes
binäres Dateiformat für den Austausch von 3D-Scans
vor.
About this book: The monograph
written by Andreas Nüchter is focused on acquiring spatial models
of physical environments through mobile robots. The robotic mapping
problem is commonly referred to as SLAM (simultaneous localization and
mapping). 3D maps are necessary to avoid collisions with complex
obstacles and to self-localize in six degrees of freedom (x-,
y-, z-position, roll, yaw and pitch angle). New
solutions to the 6D SLAM problem for 3D laser scans are proposed and a
wide variety of applications are presented. 
Zusammenfassung: Intelligentes
autonomes Roboterhandeln in Alltagsumgebungen erfordert den Einsatz
von 3D-Karten, in denen Objekte klassifiziert sind. 3D-Karten sind
u.a. zur Steuerung notwendig, damit der Roboter komplexen Hindernissen
ausweichen und sich mit 6 Freiheitsgraden (x-, y-,
z-Position,
Nick-, Gier-, und Rollwinkel) lokalisieren kann. Soll der Roboter mit
seiner Umgebung interagieren, wird Interpretation
unumgänglich. über erkannte Objekte kann der Roboter
Schlussfolgerungen ziehen, sein Wissen wird inspizier- und
kommunizierbar. Aus diesen Gründen ist die automatische und schnelle
semantische 3D-Modellierung der Umgebung eine wichtige Fragestellung
in der Robotik. 3D-Laserscanner sind eine junge Technologie, die die
Erfassung räumlicher Daten revolutioniert und Robotern das
dreidimensionale Abtasten von Objekten möglich macht. Die
vorliegende Arbeit untersucht und evaluiert mit Hilfe eines
3D-Laserscanners und des mobilen Roboters Kurt3D die zur automatischen
semantischen 3D-Kartenerstellung notwendigen Algorithmen.
A fundamental problem in
the design of autonomous mobile cognitive systems is the perception of
the environment. Robotics researches this field in order to build
reliable technical systems or to broaden the understanding of human
perception. Perception is therefore studied independently by many
researchers. On one hand, a basic part of the perception is to learn,
detect and recognize objects, which has to be done with the limited
resources of a mobile robot. The performance of a mobile system
crucially depends on the accuracy, duration and reliability of its
perceptions and the involved interpretation process. On the other
hand, automatic environment sensing and modeling is a fundamental
scientific issue in robotics, since the availability of maps is
essential for many robot tasks.
Abstract: Gibson's notion
of affordance seems to attract roboticists' attention. On a
phenomenological level, it allows functions, which have "somehow" been
implemented, to be described using a new terminology. However, that
does not mean that the affordance notion is of help for building
robots and their controllers. This paper explores viewing an
affordance as an abstraction from a robot-environment relation that is
of inter-individual use, but requires an individual
implementation. Therefore, the notion of affordance helps share
environment representations and theories among robots. Examples are
given for navigability, as afforded by environments of different types
to robots of different undercarriages and sensor configurations.
Abstract:
This paper presents a novel heuristic for correcting scan pose estimations
after loop closing in SLAM using 3D laser scans. Contrary to state of the
art approaches, the built SLAM graph is sparse, and optimization is done
without any iteration between the SLAM front and back end, yielding a
highly efficient loop closing method.
Abstract:
Acquisition and registration of terrestrial 3D laser scans is a fundamental
task in mapping and modeling of cities in three dimensions.
To automate this task marker-free registration methods are required.
Based on the existence of skyline features this paper proposes a novel method.
The skyline features are extracted from panoramic 3D scans and encoded as strings
enabling the use of string matching for merging the scans.
Initial results of the proposed method in the old city center of Bremen are
presented.
Abstract:
The Hough Transform is a well-known method for detecting para\-metrized
objects. It is the de facto standard for detecting lines and
circles in 2-dimensional data sets. For 3D it has attained little
attention so far. Even for the 2D case high computational costs have
lead to the development of numerous variations for the Hough
Transform. In this article we evaluate different
variants of the Hough Transform with respect to their applicability to detect
planes in 3D point clouds reliably.
Apart from computational costs, the main problem is the representation of the
accumulator. Usual implementations favor geometrical objects with certain
parameters due to uneven sampling of the parameter space.
We present a novel approach to design the accumulator focusing on
achieving the same size for each cell and compare it to existing designs.
Abstract:
В статье
представлена
система
визуальной
навигации
в
реальном
времени с
использованием
времяпролетной
камеры
без
априорных
знаний о
сцене.
Представлен
спо соб
комплексирования
времяпролетной
и
оптической
камеры. Рассмотрены
методы
повышения
робастности
локализации,
учитывающие
цветовую
и
пространственную
информацию.
Zusammenfassung: Seit Herbst
2004 existiert die Arbeitsgruppe "Wissensbasierte Systeme" am Institut
für Informatik der Universität Osnabrück. Eines ihrer
Arbeitsthemen ist der Bau von Roboterkarten basierend auf
3D-Laserscans bei 6-dimensionalen Scanposen. Wir geben einen
Überblick über die wichtigsten Ergebnisse dazu und über
die Perspektive dieses Themas für die Zukunft. 
Abstract:
The ICP (Iterative Closest Point) algorithm is the de facto
standard for geometric alignment of three-dimensional
models when an initial relative pose estimate is
available. The basis of the algorithm is the minimization
of an error function that takes point correspondences into
account. Four closed-form solution methods are known for
minimizing this function. This paper presents novel linear
solutions to the scan registration problem, i.e., to the
problem of putting and aligning 3D scans in a common
coordinate system. We extend the methods for registering
n-scans in a global and simultaneous fashion, such that
the registration of the n-th scan influences all
previous registrations in one step.
Abstract: We propose a new approach to appearance-based
loop detection for mobile robots, using 3D laser
scans. Loop detection is an important problem in the
SLAMdomain, and, because it can be seen as the problem of
recognizing previously visited places, it is an example of
the data association problem. Without a flat floor
assumption, 2D laser-based approaches are bound to fail in
many cases. Two of the problems with 3D approaches that we
address in this paper are how to handle the greatly
increased amount of data and how to efficiently obtain
invariance to 3D rotations. We present a compact
representation of 3D point clouds that is still
discriminative enough to detect loop closures without
false positives (i.e., detecting loop closure where there
is none). A low false positive rate is very important
because wrong data association could have disastrous
consequences in a SLAM algorithm. Our approach uses only
the appearance of 3D point clouds to detect loops and
requires no pose information. We exploit the NDT surface
representation to create feature histograms based on
surface orientation and smoothness. The surface shape
histograms compress the input data by two to three orders
of magnitude. Because of the high compression rate, the
histograms can be matched efficiently to compare the
appearance of two scans. Rotation invariance is achieved
by aligning scans with respect to dominant surface
orientations. We also propose to use expectation
maximization to fit a Gamma mixture model to the output
similarity measures in order to automatically determine
the threshold that separates scans at loop closures from
non-overlapping ones. We discuss the problem of
determining ground truth in the context of loop detection
and the difficulties in comparing the results of the few
available methods based on range information. Furthermore,
we present quantitative performance evaluations using
three real-world data sets, one of which is highly
self-similar, showing that the proposed method achieves
high recall rates (percentage of correctly identified loop
closures) at low false positive rates in environments with
different characteristics.
Abstract: This article investigates the use of
Time-of-Flight (ToF) cameras in mapping tasks for
autonomous mobile robots, in particular in simultaneous
localization and mapping (SLAM) tasks. While ToF cameras
are in principle an attractive type of sensor for 3D
mapping owing to their high rate of frames of 3D data, two
features of them make them difficult as mapping sensors,
namely, their restricted field of view and influences on
the quality of range measurements by high dynamics in
object reflectivity; in addition, currently available
models suffer from poor data quality in a number of
aspects. The paper first summarizes calibration and
filtering approaches for improving accuracy, precision and
robustness of ToF camera independent of their intended
usage. Then, several ego motion estimation approaches are
applied or adapted, respectively, in order to provide a
performance benchmark for registering ToF camera data. As
a part of this, an extension to the Iterative Closest
Point (ICP) algorithm has been developed that increases
the robustness under restricted field of view and under
larger displacements. Using an indoor environment, the
paper provides results from SLAM experiments using these
approaches in comparison. It turns out that the
application of ToF cameras is feasible to SLAM tasks,
although this type of sensor has a complex error
characteristic.
Abstract:
In this paper, we present a two-step approach for object
recognition based on principles of human perception:
Attentive Classification. First, regions of interest are
detected by a biologically motivated attention
system. Second, these regions are analyzed by a fast
classifier based on the Adaboost learning technique. Thus,
the classification effort is restricted to a small data
subset. The approach has two advantages over normal
classification: First, the system becomes considerably
faster, which is an important factor for real-time
systems. Second, since the attention system is able to
make use of top-down target-information, the combination
of the systems yields a significant reduction of false
detections for objects which are usually difficult to
discriminate from the surrounding. We show the performance of
the system in several experiments in robotic
scenarios. The presented attentive classification system
represents an important step towards effective general
object recognition which is fast, robust and flexibly
adaptable to a current task.
Abstract: Intelligent
autonomous acting of mobile robots in unstructured environments
requires 3D maps. Since manual mapping is a tedious job,
automatization of this job is necessary. Automatic, consistent
volumetric modeling of environments requires a solution to the
simultaneous localization and map building problem (SLAM problem). In
3D this task is computationally expensive, since the environments are
sampled with many data points with state of the art sensing
technology. In addition, the solution space grows exponentially with
the additional degrees of freedom needed to represent the robot
pose. Mapping environments in 3D must regard six degrees of freedom to
characterize the robot pose. This paper summarizes our 6D SLAM
algorithm and presents novel algorithmic and technical means to reduce
computation time, i.e., the data structure cached k-d tree and
parallelization. The availability of multi-core processors as well as
efficient programming schemes as OpenMP permit the parallel execution
of robotics tasks. 
Abstract: Intelligent autonomous
action in ordinary environments calls for 3D maps. 3D geometry is
necessary for avoiding collision with complex obstacles and to self
localize in six degrees of freedom (6 DoF) (x, y,
z positions, roll, yaw, and pitch angles). Meaning, in
addition to geometry, becomes inevitable if the robot is supposed to
interact with its environment in a goal-directed way. A semantic
stance enables the robot to reason about objects; it helps
disambiguate or round off sensor data; and the robot knowledge becomes
reviewable and communicable.
Abstract: Quite a number of
approaches for solving the simultaneous localization and mapping
(SLAM) problem exist by now. Some of them have recently been extended
to mapping environments with six degrees of freedom (DoF) poses,
yielding 6D SLAM approaches. To demonstrate the capabilities
of the respective algorithms, it is common practice to present
generated maps and successful loop closings in large outdoor
environments. Unfortunately, it is non-trivial to compare different
6D SLAM approaches objectively, because ground truth data about the
outdoor environments used for demonstration is typically unavailable.
We present a novel benchmarking method for generating this ground
truth data based on reference maps. The method is then demonstrated
by comparing the absolute performance of some previously existing 6D
SLAM algorithms which build a large urban outdoor map.
Abstract: A globally consistent
solution to the simultaneous localization and mapping (SLAM) problem
in 2D with three degrees of freedom (DoF) poses was presented by Lu
and Milios [F. Lu, E. Milios, Globally consistent range scan alignment
for environment mapping, Autonomous Robots 4 (April) (1997)
333-349]. To create maps suitable for natural environments it is
however necessary to consider the 6DoF pose case, namely the three
Cartesian coordinates and the roll, pitch and yaw angles. This article
describes the extension of the proposed algorithm to deal with these
additional DoFs and the resulting non-linearities. Simplifications
using Taylor expansion and Cholesky decomposition yield a fast
application that handles the massive amount of 3D data and the
computational requirements due to the 6DoF. Our experiments
demonstrate the functionality of estimating the exact poses and their
covariances in all 6DoF, leading to a globally consistent map. The
correspondences between scans are found automatically by use of a
simple distance heuristic.
Abstract: 6D SLAM (Simultaneous
Localization and Mapping) or 6D Concurrent Localization and Mapping of
mobile robots considers six dimensions for the robot pose, namely, the
x, y and z coordinates and the roll, yaw and
pitch angles. Robot motion and localization on natural surfaces, e.g.,
driving outdoor with a mobile robot, must regard these degrees of
freedom. This paper presents a robotic mapping method based on locally
consistent 3D laser range scans. Iterative Closest Point (ICP)
scan matching, combined with a heuristic for closed loop detection and
a global relaxation method, results in a highly precise mapping
system. A new strategy for fast data association, cached kd tree
search, leads to feasible computing times. With no ground-truth data
available for outdoor environments, point relations in maps are
compared to numerical relations in uncalibrated aerial images in order
to assess the metric validity of the resulting 3D maps.
Abstract: 6D SLAM
(Simultaneous Localization and Mapping) or 6D Concurrent Localization
and Mapping of mobile robots considers six dimensions for the robot
pose, namely, the x, y and z coordinates
and the roll, yaw and pitch angles. Robot motion and localization on
natural surfaces, e.g., when driving with a mobile robot outdoor, must
regard these degrees of freedom. 3D (6 DOF) scan matching, combined
with a heuristic for closed loop detection and a global relaxation
method, results in a highly precise mapping system for outdoor
environments. The mobile robot Kurt3D is capable to run the mapping
process with its on-board sensors and computers and is used to
digitalize different environments. This paper summarizes our previous
research.
Abstract: The RoboCup Rescue
competition aims at boosting research in robots and infrastructure
able to help in real rescue missions. The task is to find and report
victims in areas of different grades of roughness, which are currently
indoor. It challenges to some extreme the mobility of robot platforms
as well as the autonomy of their control and sensor interpretation
software. In the 2004 competition, the Kurt3D robot was introduced,
the first participant capable of mapping its environment in 3D and
self-localizing in all six degrees of freedom, i.e., x, y, z positions
and roll, yaw and pitch angles. In 2005, we have upgraded the system
with more sensors, with a focus on speeding up the algorithms, and we
have started to develop a tracked robot platform to cooperate with
Kurt3D. This paper gives an introduction to the competition in general
and presents main contributions of our Deutschland1 RoboCup Rescue
team. 
Abstract: We discuss online
strategies for visibility-based searching for an ob ject hidden behind
a corner, using Kurt3D, a real autonomous mobile robot. This task is
closely related to a number of well-studied problems. Our robot uses a
threedimensional laser scanner in a stop, scan, plan, go fashion for
building a virtual three-dimensional environment. Besides planning
trajectories and avoiding obstacles, Kurt3D is capable of identifying
objects like a chair. We derive a practically useful and
asymptotically optimal strategy that guarantees a competitive ratio of
2, which differs remarkably from the well-studied scenario without the
need of stopping for surveying the environment. Our strategy is used
by Kurt3D, documented in a separate video. 
Abstract: Bei dem Verfahren zur
Ermittlung der Position und Orientierung eines autonom navigierenden
Systems, beispielsweise eines Roboters, in einer Umgebung werden die
Entfernung des in Fahrtrichtung vor den navigierenden System liegenden
Bereichs der Umgebung bei Bewegung des navigierenden Systems
abgetastet und die abgetasteten Entfernungspunkte mindestens zweier
aufeinanderfolgender Abtastvorgänge als Entfernungsmesskurven in
der Polardarstellung gespeichert. Anschliessend werden die
Entfernungsmesskurven der beiden aufeinander folgenden
Abtastvorgänge auf charakteristische merkmale wie
z.B. Extremwerte untersucht. Danach werden die parameter von einender
getrennt durchführbaren Translations- und
Rotationstransformationen der einen Entfernungsmesskurve zur
Ermittlungder zuordnung der charakteristischen Merkmale der
trnasformierten Entfernungsmesskurve zu solchen der anderen
Entfernungsmesskurve und die Position und Orientierung des
navigierenden Systems anhand von dessen Position und dessen
Orientierung zum Zeitpunkt des zeitlich ersten von zwei
aufeinanderfolgenden Abtastungen sowie der ermittelten Parameter der
Translation- und Rotationstransformation bestimmt.
Abstract: In this paper, we present a
new bimodal attention system for robotic applications capable of
processing data from different sensor modes
simultaneously. Considering several sensor modalities is an obvious
approach to regard a variety of ob ject properties. Nevertheless,
conventional att ention systems only regard the processing of camera
images. In contrast to these systems, the input data to our system is
provided by a bimodal 3D laser scanner, mounted on top of an
autonomous mobile robot. In a single 3D scan pass, the scanner yields
range as well as reflectance data. Both dat a modes are illumination
independent, yielding a robust approach that enables all day
operation. Data from both laser modes are fed into our attention
system built on principles of one of the standard models of visual
attention by Koch & Ullman. The system computes conspicuities of both
modes in parallel and fuses them into one saliency map. The focus of
attention is directed to the most salient points in this map
sequentially. We present results on recorded scans of indoor and
outdoor scenes showing the respective advantages of the sensor
modalities enabling the mode-specific detectio n of different ob ject
properties. Furthermore, we show as an application of the attention
system the recognition of ob jects for building semantic 3D maps of
the robot's environment. Key words: visual attention, saliency
detection, bimodal sensor fusion, 3D laser scanner
Abstract: This paper describes a
novel, laser-based approach for tracking the pose of a high-speed
mobile robot. The algorithm is outstanding in terms of accuracy and
computation time. The efficiency is achieved by a closed-form solution
for the matching of two laser scans, the use of natural scan features
and fast linear filters. The implemented algorithm is evaluated with
the high-speed robot Kurt3D (4 m/s), and compared to standard scan
matching methods in indoor and outdoor environments. Keywords:
Localization; Pose tracking; Autonomous mobile robots; Scan matching;
High-speed robotics.
Abstract: Digital 3D models of the
environment are needed in rescue and inspection robotics, facility
managements and architecture. This paper presen ts an automatic system
for gaging and digitalization of 3D indoor environments. It consists
of an autonomous mobile robot, a reliable 3D laser range finder and
three elaborated software modules. The first module, a fast variant of
the Iterative Closest Points algorithm, registers the 3D scans in a
common coordinate system and relocalizes the robot. The second module,
a next best view planner, computes the next nominal pose based on the
acquired 3D data while avoiding complicated obstacles. The third
module, a closed-loop and globally stable motor controller, navigates
the mobile robot to a nominal pose on the base of odometry and avoids
collisions with dynamical obstacles. The 3D laser range finder
acquires a 3D scan at this pose. The proposed method allows one to
digitalize large indoor environments fast and reliably without any
intervention and solves the SLAM problem. The results of two 3D
digitalization experiments are presented using a fast octree-based
visualization method. Keywords: Autonomous mobile robots; 3D laser
range finder; Scan matching; Next best view planning; 3D
digitalization; 3D gaging; Robot relocalization; SLAM
Abstract:
This work introduces a novel solution for localizing
objects based on search strings and freely available Google SketchUp
models. To this end we automatically download and
preprocess a collection of 3D models to obtain equivalent
point clouds. The outdoor scan is segmented into
individual objects, which are sequentially matched with
the models by a variant of iterative closest points
algorithm using seven degrees of freedom and resulting in
a highly precise pose estimation of the object. An error
function evaluates the similarity level. The approach is
verified using various segmented cars and their
corresponding 3D models.
Abstract:
This paper presents a novel technique for detecting
vegetation of virtually all forms in terrestrial laser
scanning data of urban environments. We make use of a
modern laser range finder capability to measure multiple
echoes per laser pulse via Full Wave Analysis. The
algorithm is able to efficiently, i.e., less than
acquisition time, identify vegetation to a high degree of
accuracy (more than 99 percent). We present and evaluate
three alternatives to classify candidate regions as either
vegetation or non-vegetation.
Abstract:
Autonomous robots equipped with laser scanners acquire
data at an increasingly high rate. Registration, data
abstraction and visualization of this data requires the
processing of a massive amount of 3D data. The increasing
sampling rates make it easy to acquire Billions of spatial
data points. This paper presents algorithms and data
structures for handling this data. We propose an efficient
octree to store and compress 3D data without loss of
precision. We demonstrate its usage for fast 3D scan
matching and shape detection algorithms. We evaluate our
approach using typical data acquired by mobile scanning
platforms.
Abstract:
The computational complexity of SLAM is large
and constitutes a challenge for real-time processing of a huge
amount of sensor data with the limited resources of a mobile
robot. Often, notebooks are used to control a mobile system and
even these computing devices have nowadays graphics cards
which allow general purpose computation using many cores.
SLAM à la carte (graphique) exploits these
capabilities and carries out 3D scan registrations on the
GPU. A speed-up of more than one order of magnitude for
precise 3D mapping is reported.
Abstract:
Acquisition and registration of terrestrial 3D laser scans is a fundamental
task in mapping and modeling of cities in three dimensions.
To automate this task marker-free registration methods are required.
Based on the existence of skyline features this paper proposes a novel method.
The skyline features are extracted from panoramic 3D scans and encoded as strings
enabling the use of string matching for merging the scans.
Initial results of the proposed method in the old city center of Bremen are
presented.
Zusammenfassung:
Dieser Beitrag stellt eine neue Implementation der Octree-Datenstruktur vor,
die es ermöglicht, eine Milliarde 3D-Punkte in 8 GB Hauptspeicher exakt zu
repräsentieren und effiziente Algorithmen zu implementieren. Der Octree gibt
eine Hierarchie vor, die dazu verwendet werden kann, große Punktewolken zu
inspizieren und flüssig in ihnen zu navigieren.
Abstract:
To benchmark the efficiency of exploration strategies one has
to use robot simulators. In an exploration task, the robot faces an
unknown environment. Of course one could test the algorithm in different
real-world scenarios, but a competitive strategy must have good
performance in any environment that can be systematically constructed inside
a simulator. This paper presents an evaluation of exploration strategies
we developed for a specific sensor. A continously rotating 3D laser
scanner that scans only into one direction at a time moves through the
environment sampling the surrounding. Our evaluation framework
features an efficient scanning and robot simulator for kinematic feasible
trajectories. We will show that shorter trajectories do not necessarily
imply quicker exploration. A simple simulator framework is sufficient for
evaluating these properties of path planning algorithms.
Abstract:
Three dimensional point clouds acquired by range scanners often do not
represent the environment precisely due to noise and errors in
the acquisition process. These latter systematical errors manifest as
deformations of different kinds in the 3D range image. This paper
presents a novel approach to correct deformations by an analysis of
the structures present in the environment and correcting them by
non-rigid transformations. The resulting algorithms are used for
creating high-accuracy 3D indoor maps.
Abstract:
The recently introduced Minimum Uncertainty
Maximum Consensus (MUMC) algorithm for 3D scene registration
using planar-patches is tested in a large outdoor urban
setting without any prior motion estimate whatsoever. With
the aid of a new overlap metric based on unmatched patches,
the algorithm is shown to work successfully in most cases. The
absolute accuracy of its computed result is corroborated for the
first time by ground-truth obtained using reflective markers.
There were a couple of unsuccessful scan-pairs. These are
analyzed for the reason of failure by formulating two kinds of
overlap metrics: one based on the actual overlapping surfacearea
and another based on the extent of agreement of rangeimage
pixels.We conclude that neither metric in isolation is able
to predict all failures, but that both taken together are able to
predict the difficulty level of a scan-pair vis-a-vis registration
by MUMC.
Abstract:
Geodesy and surveying are the sciences for monitoring the earth. In
recent years traditional surveying equipment has been pushed aside
by the emerging technology of laser scanners, that automate the
precise measurement of points in the environment. A further step of
automation is achieved by operating the surveying equipment
automatically and the usage of robotic mapping. Thus, robotic mapping
will become a key component in monitoring and surveillance tasks. This
paper evaluates a 3D laser scanner from surveying for its usage in
robotic monitoring tasks. We examine how seasonal changes and weather
conditions impact the data of the 3D scanner and how to deal with
these changes. For this analysis 3D scans of various predetermined
locations on the campus of the Jacobs University Bremen were taken on
a weekly basis over a period of 13 weeks using the RIEGL VZ-400 3D
laser range finder. The scans have been registered by means of
conventional surveying markers, SIFT features and using our point
cloud based 6D SLAM framework. An analysis of the changes in the
environment over the course of the scanning period and an evaluation
of the matching results complete the analysis.
Abstract:
The Hough Transform is a well-known method for detecting
parametrized objects. It is the de facto standard for the
detection of lines and circles in 2-dimensional data
sets. For 3D it has attained little attention so far.
Apart from computational costs, the main problem is the
representation of the accumulator: Usual implementations
favor geometrical objects with certain parameters due to
uneven sampling of the parameter space. In this paper we
present a novel approach to design the accumulator
focusing on achieving the same size for each cell. The
proposed accumulator is compared to previously known
designs.
Abstract:
This paper reports a method and results for solving the
following problem: Given a 3D polygonal indoor map and a
mobile robot equipped with a 3D time of flight (ToF)
camera, localize at frame rate the 6D robot pose with
respect to the map. To solve the problem, the polygonal
map is represented for efficient usage as a solid-leaf BSP
tree; at each control cycle, the 6D pose change is
estimated a priori from odometry or IMU, the expected ToF
camera view at the prior pose sampled from the BSP tree,
and the pose change estimation corrected a posteriori by
fast ICP matching of the expected and the measured ToF
image. Our experiments indicate that, first, the method is
in fact real-time capable; second, the 6D pose is tracked
reliably in a correct map under regular sensor conditions;
and third, the tracking can recover from some faults
induced by local map inaccuracies and transient or local
sensing errors.
Abstract:
This paper presents a novel approach to create polygonal
maps from 3D point cloud data. The gained map is augmented
with a interpretation of the scene. Our procedure shows to
be fast and reliable and produces accurate maps in indoor
environments. The created maps are used with different
kinds of sensors for reliable self localization.
Abstract:
The ICP (Iterative Closest Point) algorithm is the de
facto standard for geometric alignment of
three-dimensional models when an initial relative pose
estimate is available. The basis of the algorithm is the
minimization of an error function that takes point
correspondences into account. While four closed-form
solution methods are known for minimizing this function,
linearization seems necessary for solving the global scan
registration problem. This paper presents such linear
solutions for registering n-scans in a global and
simultaneous fashion. It studies parameterizations for the
rigid body transformations of the n-scan registration
problem.
Abstract:
Laserscanner sind präzise Messgeräte zur Ermittlung von
Distanzwerten. Dennoch weist eine jede mit einem
Laserscanner akquirierte Punktwolke Fehler auf, die auch
durch Kalibrierung nicht vollständig verhindert werden
können. Neben Sensorrauschen kommt es auch zu
systematischen Fehlern. Die meisten künstlich geschaffenen
Umgebungen bestehen aus einer großen Anzahl ebener
Flächen, die helfen können, die Qualität von
Laserscan-Karten zu verbessern. Dieser Aufsatz stellt
Methoden vor, die unter Zuhilfenahme von ebenen
Umgebungsstrukturen die Fehler in den Laserscans durch
nicht-rigide Verformungen verringern.
Einleitung:
Terrestrisches Laserscanning (TLS) hat die in den letzten
Jahren die Vermessungstechnik revolutioniert. Durch die
Entwicklung des kinematischen terrestrischen
Laserscannings (k-TLS) oder Mobile Mapping, das die
Aufnahme von geometrische Umgebungsinformation von einer
bewegten Plattform aus erlaubt, wurde ein wesentlicher
Schritt zur weiteren Automatisierung in der
Vermessungstechnik getan. Leider ist k-TLS nicht überall
einsetzbar, da neben den Daten des Laserscanners
hochgenaue Informationen über die Pose (Position und
Orientierung) der mobilen Plattform vorliegen müssen.
Nearest Neighbor Search (NNS) is employed by many computer
vision algorithms. The computational complexity is large
and constitutes a challenge for real-time capability. The
basic problem is in rapidly processing a huge amount of
data, which is often addressed by means of highly
sophisticated search methods and parallelism. We show that
NNS based vision algorithms like the Iterative Closest
Points algorithm (ICP) can achieve real-time capability
while preserving compact size and moderate energy
consumption as it is needed in robotics and many other
domains. The approach exploits the concept of general
purpose computation on graphics processing units (GPGPU)
and is compared to parallel processing on CPU. We apply
this approach to the 3D scan registration problem, for
which a speed-up factor of 88 compared to a sequential CPU
implementation is reported.
Abstract:
Simultaneous Localization and Mapping (SLAM) is the problem of
building a map of an unknown environment by a mobile robot while at
the same time navigating the environment, using the unfinished
map. For SLAM, two tasks have to be solved: First reliable feature
extraction and data association, second the optimal estimation of
poses and features. These two parts are often referred to as SLAM
frontend and backend. Algorithms that solve SLAM by using laser scans
commonly rely on matching closest points in the frontend part. Then
the SLAM front- and backend have to be iterated to ensure that the map
converges.
Abstract: Time-of-Flight
cameras constitute a smart and fast technology for 3D robotic
perception but lack in measurement precision and robustness. We
present a comprehensive approach for 3D environment mapping based on
this technology. Imprecision of depth measurements are properly
handled by calibration and application of several filters. Robust
registration is performed by a novel extension to the Iterative
Closest Point algorithm. Remaining registration errors are refined by
global relaxation after loop-closure and surface smoothing. A
laboratory ground truth evaluation is provided as well as 3D mapping
experiments in a larger indoor environment.
Abstract:
To advance robotic science it is important to perform experiments that
can be replicated by other researchers to compare different
methods. However, these comparisons tend to be biased, since
re-implementations of reference methods often lack thoroughness and do
not include the hands-on experiences obtained during the original
development process. This paper presents the results of a field
experiment, carried out by two research groups that are leading in the
field of 3D robotic mapping. The iterative closest points algorithm
(ICP) is compared to the normal distributions transform (NDT).
Abstract: We propose a new
approach to appearance based place recognition from metric 3D maps,
exploiting the NDT surface representation. Locations are described
with feature histograms based on surface orientation and smoothness,
and loop closure can be detected by matching feature histograms. 
Zusammenfassung: Dieser
Artikel beschreibt eine Linearisierung in geschlossener Form für
die Minimierung der Fehlerfunktion, die beim ICP-Algorithmus
auftaucht. Die Linearisierung approximiert die tatsächliche
Lösung und nutzt die Annahme aus, dass die auftretenden Winkel klein
sind. Weiterhin zeigt der Artikel die Möglichkeit auf, die
Linearisierung für global konsistentes Scanmatching zu
verwenden. Global konsistentes Scanmatching minimiert den
Gesamtfehler, wenn mehr als zwei 3D-Punktwolken vorliegen.
Abstract: The following paper
evaluates a time-of-flight 3D camera with regards to its usability for
RoboCup Rescue. This includes an evaluation of the influence of outer
conditions to the camera data, as well as its usage for automatic 3D
mapping by scan registration. A color camera is calibrated with
respect to the 3D camera in order to gain colored texture information
for the acquired measurements. 
Abstract: Over ten years ago,
Lu and Milios presented a probabilistic scan matching algorithm for
solving the simultaneous localization and mapping (SLAM) problem with
2D laser range scans, a standard in robotics. This paper presents an
extension to this GraphSLAM method. Our iterative algorithm uses a
sparse network to represent the relations between several overlapping
3D scans, computes in every step the 6 degrees of freedom (DoF)
transformation in closed form and exploits efficient data association
with cached k-d trees. Our approach leads to globally
consistent 3D maps, precise 6D pose and covariance estimates, as
demonstrated by various experimental results.
Abstract: Das Einpassen
bzw. Registrieren von Punktmengen unter starren Transformationen ist
eines der Grundprobleme in der Bildverarbeitung. Hierbei können
die 3D-Daten aus Laserscannern, Stereokameras
u.ä. stammen. Für zwei 3D-Punktmengen bildet der
ICP-Algorithmus (Iterative Closest Points) einen de facto Standard
für das Registrieren. Weitet man diesen Algorithmus jedoch auf
viele 3D-Scans aus, akkumulieren sich Registrierungsfehler. Der
vorliegende Beitrag skizziert einen neuen Algorithmus für das
global konsistente 3D-Scanmatching (Borrmann, Elseberg, Lingemann,
Nüchter und Hertzberg 2008) und zeigt eine Anwendung für das
Erfassen von Gebäuden. Die Genauigkeit des Algorithmus wird
bezüglich geodätischer Methoden evaluiert.
Abstract: In the past many
solutions for simultaneous localization and mapping (SLAM) have been
presented. Recently these solutions have been extended to map large
environments with six degrees of freedom (DoF) poses. To demonstrate
the capabilities of these SLAM algorithms it is common practice to
present the generated maps and successful loop closing. Unfortunately
there is often no objective performance metric that allows to compare
different approaches. This fact is attributed to the lack of ground
truth data. For this reason we present a novel method that is able to
generate this ground truth data based on reference maps. Further on,
the resulting reference path is used to measure the absolute
performance of different 6D SLAM algorithms building a large
urban outdoor map.
Abstract: Robotic 3D Mapping of
environments is computationally expensive, since 3D scanners sample
the environment with many data points. In addition, the solution space
grows exponentially with the additional degrees of freedom needed to
represent the robot pose. Mapping environments in 3D must regard six
degrees of freedom to characterize the robot pose. This paper extends
our solution to the 3D mapping problem by parallelization. The
availability of multi-core processors as well as efficient programming
schemes as OpenMP permit the parallel execution of robotics task with
on-board means. 
Abstract: In this paper we
present a reliable approach for real-time outdoor trail following and
obstacle avoidance. The trail classification is done using an
off-the-shelf webcam and a pitched 2D laser scanner on a KURT2 robot
equipped with an Intel Centrino laptop. This simple setup enables us
to follow given pathways of different kinds using a GPS receiver for
rough orientation.
Abstract: 6D SLAM (Simultaneous
Localization and Mapping) or 6D Concurrent Localization and Mapping of
mobile robots considers six degrees of freedom for the robot pose,
namely, the x, y and z coordinates and the roll, yaw and pitch
angles. In previous work we presented our scan matching based 6D SLAM
approach [cite deleted], where scan matching is based on the well
known iter ative closest point (ICP) algorithm [3]. Effcient
implementations of this algorithm are a result of a fast computation
of closest points. The usual approach, i.e., using k-d trees is
extended in this paper. We describe a novel search strategy, that lead
s to significant speed-ups. Our mapping system is real-time capable,
i.e., 3D maps are computed using the resources of the used robotic
hardware.
Abstract: The focus of this paper
is on the performance comparison of two simultaneous localization and
mapping (SLAM) algorithms namely 6D Lu/Milios SLAM and Force
Field Simulation (FFS). The two algorithms are applied to a 2D
data set. Although the algorithms generate overall visually comparable
results, they show strengths and weaknesses in different regions of
the generated global maps. The question we address in this paper is,
if different ways of evaluating the performance of SLAM algorithms
project different strengths and how can the evaluations be useful in
selecting an algorithm. We will compare the performance of the
algorithms in different ways, using grid and pose based quality
measures.
Abstract:
The ICP (Iterative Closest Point) algorithm is the de facto standard
for geometric alignment of three-dimensional models when an initial
relative pose estimate is available. The basis of ICP is the search
for closest points. Since the development of ICP, k-d trees have been
used to accelerate the search. This paper presents a novel search
procedure, namely cached k-d trees, exploiting iterative behavior of
the ICP algorithm. It results in a significant speed-up of about 50%
as we show in an evaluation using different data sets.
Abstract: The great majority of
path following control laws for either kinematical or dynamical mobile
robot models are designe d assuming ideal actuators, i.e. assuming
that any commanded velocity or torque (in the kinematical and
dynamical ca ses respectively) will be instantly implemented
regardless of its value. Real actuators are far from being ideal. In
particular, only bounded velocities and torques can be realized for
any given command. With reference to the kinemati cal model of a
differential drive mobile robot, a known path following control law is
modified to account for actuator velocity saturation. The proposed
solution is experimentally shown to be particularly useful for high
speed applica tions where accounting for actuator velocity saturation
may have a large influence on performance.
Abstract:
One fundamental problem in mobile robotics research is
Simultaneous Localization and Mapping (SLAM): A mobile
robot has to localize itself in an unknown environment, and
at the same time generate a map of the surrounding area. One fundamental
part of SLAM algorithms is loop closing: The robot
detects whether it has reached an area that has been visited before, and
uses this information to improve the pose estimate in the next step.
In this work, visual camera features are used to assist closing the
loop in an existing 6 degree of freedom SLAM (6D SLAM) architecture.
For our robotics application we propose and evaluate several detection
methods, including salient region detection and maximally stable
extremal region detection. The detected regions are encoded using
SIFT descriptors and stored in a database. Loops are detected by
matching of the images' descriptors.
A comparison of the different feature detection methods shows that the
combination of salient and maximally stable extremal regions suggested by
performs moderately.
Zusammenfassung:
3D-Laserscanner sind eine junge Technologie, die die Erfassung
räumlicher Daten revolutioniert und Robotern das dreid imensionale
Abtasten von Objekten möglich macht. Roboter, die ihre Umgebung
dreidimensional kartieren können, eignen sich zum automatischen
Erstellen virtuelle 3D-Welten. Eine 3D-Welt, oder 3D-Umgebungskarte
muss mit der wirklichen Umgebung übereinstimmen, also korrekt und
konsistent sein. Ist die Position des Roboters genau bekannt, können
die loka len Karten auf der Grundlage dieser Position zusammengefügt
werden. Leider ist die Selbstlokalisation eines Roboters stets
fehlerbehaftet. Daher darf der Kartenbau nicht nur auf der
Roboterposition basieren, sondern muss auch auf der Grundlage der
Sensorwerte geschehen. In diesem Zusammenhang spricht man vom
simultanen Lokalisations- und Kartierungs problem (SLAM, simultaneous
localization and mapping problem). Korrekte, global konsistente
Modelle entstehen durch e inen 6D-SLAM Algorithmus. Hierbei werden 6
Freiheitsgrade in der Roboterpose berücksichtigt, geschlossene Kreise
erka nnt und der globale Fehler minimiert. Die Basis des 6D-SLAM ist
ein sehr schneller ICP-Algorithmus.
Abstract: 6D SLAM (Simultaneous
Localization and Mapping) of mobile robots considers six dimensions
for the robot pose, namely, the x, y and z coordinates and the roll,
yaw and pitch angles. Robot motion and localization on natural
surfaces, e.g., when driving with a mobile robot outdoor, must regard
these degrees of freedom. This paper presents a robotic mapping method
based on locally consistent 3D laser range scans. Scan matching,
combined with a heuristic for closed loop detection and a global
relaxation method, results in a highly precise mapping system for
outdoor environments. The mobile robot Kurt3D was used to acquire data
of the Schloss Birlinghoven campus. The resulting 3D map is compared
with ground truth, given by an aerial photograph.
Abstract: USARSIM is a
worldwide used robot simulator deployed in Urban Search and Rescue
(USAR) and in the context of the RoboCup Rescue Real Robot
contest. This paper describes the USARSIM simulator for KURT2 and
Kurt3D robot platforms, which we are using in both education and
research. As it simulates on the device level, a seamless integration
of real robot control software with the simulations becomes
possible. We evaluate the performance for simulating laser range scans
and the camera system. In addition, we show a simulation of the rescue
robots.
Abstract: This paper presents
a novel object recognition approach based on range images. Due to its
insensitivity to illumination, range data is well suited for reliable
silhouette extraction. Silhouette or contour descriptions are good
sources of information for object recognition. We propose a complete
object recognition system, based on a 3D laser scanner, reliable
contour extraction with floor interpretation, feature extraction using
a new, fast Eigen-CSS method, and a supervised learning algorithm. The
recognition system was successfully tested on range images acquired
with a mobile robot, and the results are compared to standard
techniques, i.e., Geometric features, Hu and Zernike moments, the
Border Signature method and the Angular Radial Transformation. An
evaluation using the receiver operating characteristic analysis
completes this paper. The Eigen-CSS method has proved to be
comparable in detection performance to the top competitors, yet faster
than the best one by an order of magnitude in feature extraction time.
Abstract: A basic issue of
mobile robotics is generating environment maps automatically. Outdoor
terrain is challenging since the ground is uneven and the surrounding
is structured irregularly. In earlier work, we have introduced 6D
SLAM (Simultaneous Localization and Mapping) as a method to taking all
six DOF of robot poses (x, y, z
translation; roll, pitch, yaw angles) into account. This paper adds
to 6D SLAM a method for extracting drivable surfaces in the 3D maps
while they are being generated. Experiments have been made in a
Botanical Garden, with drivable surfaces consisting of gravel paths or
lawn, both involving significant slope.
Abstract: This paper describes the
control algorithms of the high speed mobile robot Kurt3D. Kurt3D
drives up to 4 m/s autonomously and reliably in an unknown office
environment. We present the reliable hardware, fast control cycle
algorithms and a novel set value computation scheme for achieving
these velocities. In addition we sketch a real-time capable laser
based position tracking method that is well suited for driving with
these velocities.
Abstract: In der Lehre zum Thema
Wissensbasierte Robotik verwenden wir seit Kurzem den Robotersimulator
USARSIM, der weltweit im Kontext der RoboCup Rescue Real Robot Liga
eingesetzt wird. Wir stellen den Lehr-Kontext vor, in dem wir
arbeiten, skizzieren den Simulator und beschreiben seine Einbindung in
unsere Lehre. Unsere Erfahrungen bezuglich der Motivation der
Studierenden und ihrer Leistungen der Verwendung des Simulators sind
sehr positiv.
Abstract: 6D SLAM (Simultaneous
Localization and Mapping) or 6D Concurrent Localization and Mapping of
mobile robots considers six dimensions for the robot pose, namely, the
x, y and z coordinates and the roll, yaw and pitch angles. Robot
motion and localization on natural surfaces, e.g., driving with a
mobile robot outdoor, must regard these degrees of freedom. This paper
presents a robotic mapping method based on locally consistent 3D laser
range scans. Scan matching, combined with a heuristic for closed loop
detection and a global relaxation method, results in a highly precise
mapping system for outdoor environments. The mobile robot Kurt3D was
used to acquire data of the Schloss Birlinghoven campus. The resulting
3D map is compared with ground truth, given by an aerial photograph.
Abstract: This paper presents a
novel method for object detection and classification in 3D laser range
data that is acquired by an autonomous mobile robot. Unrestricted
objects are learned using classification and regression trees (CARTs)
and using an Ada Boost learning procedure. Off-screen rendered depth
and reflectance images serve as an input for learning. The performance
of the classification is improved by combining both sensor modalities,
which are independent from external light. This enables highly
accurate, fast and reliable 3D object localization with point
matching. Competitive learning is used for evaluating the accuracy of
the object localization.
Abstract: This paper provides a new
solution to the simultaneous localization and mapping (SLAM) problem
with six degrees of freedom. A fast variant of the Iterative Closest
Points (ICP) algorithm registers 3D scans taken by a mobile robot into
a common coordinate system and thus provides relocalization. Hereby,
data association is reduced to the problem of searching for closest
points. Approximation algorithms for this searching, namely,
approximate kd-trees and box decomposition trees, are presented and
evaluated in this paper. A solution to 6D SLAM that considers all free
parameters in the robot pose is built based on 3D scan matching.
Abstract: A basic task of rescue
robot systems is mapping of the environment. Localizing injured
persons, guiding rescue workers and excavation equipment requires a
precise 3D map of the environment. This paper presents a new 3D laser
range finder and novel scan matching method for the robot Kurt3D
[9]. Compared to previous machinery [12], the apex angle is enlarged
to 360 . The matching is based on semantic information. Surface
attributes are extracted and incorporated in a forest of search trees
in order to associate the data, i.e., to establish
correspondences. The new approach results in advances in speed and
reliability.
Andreas Nüchter, Kai Lingemann, Joachim
Hertzberg, Hartmut Surmann, Kai Pervölz, Matthias Hennig,
K. R. Tiruchinapalli, Rainer Worst, and Thomas Christaller,
Mapping of Rescue Environments with Kurt3D, in
Proceedings of the International Workshop on Safty, Security and
Rescue Robotics (SSRR '05), ISBN 0-7803-8946-8, pages 158 - 163,
Kobe, Japan, June 2005, (best paper award) [
Abstract: Deploying rescue workers in
an urban setting is often a perilous, time-, power-, and
force-consuming job, and systems to assist in this effort are
needed. A fundamental task for rescue is to localize injured
persons. To this end, robotic systems are used for mapping a site and
for remote inspection of suspicious objects. The mobile robot Kurt3D
is the first rescue robot that is capable of mapping its environment
in 3D and self localize in all six degrees of freedom, i.e.,
considering its x, y and z positions and the roll, yaw and pitch
angles.
Abstract: In this paper, we present
a new combination of a biologically inspired attention system (VOCUS
Visual Object detection with a Computational attention System) with a
robust object detection method. As an application, we built a reliable
system for ball recognition in the RoboCup context. Firstly, VOCUS
finds regions of interest generating a hypothesis for possible
locations of the ball. Secondly, a fast classifier verifies the
hypothesis by detecting balls at regions of interest. The combination
of both approaches makes the system highly robust and eliminates false
detections. Furthermore, the system is quickly adaptable to balls in
different scenarios: The complex classifier is universally applicable
to balls in every context and the attention system improves the
performance by learning scenario-specific features quickly from only a
few training examples. Index Terms - visual attention, object
classification.
Abstract: This paper presents a
robust and real-time capable recognition system for the fast detection
and classification of objects in spatial 3D data. Depth and reflection
data from a 3D laser scanner are rendered into images and fed into a
saliency-based visual attention system that detects regions of
potential interest. Only these regions are examinated by a fast
classifier. The time saving of classifying objects in salient regions
rather than in complete images is linear with the number of trained
object classes. Robustness is achieved by the fusion of the bi-modal
scanner data; in contrast to camera images, this data is completely
illumination independent. The recognition system is trained for two
different object classes and evaluated on real indoor data.
Abstract: This paper describes a
novel, laser-based approach for tracking the pose of a high-speed
mobile robot. The algorithm is outstanding in terms of accuracy and
computational time, being 33 times faster than real time. The
efficiency is achieved by a closed form solution for the matching of
two laser scans, the use of natural landmarks and fast linear
filters. The implemented algorithm is evaluated with the high-speed
robot Kurt3D (4 m/s), and compared to standard scan matching methods
in indoor and outdoor environments.
Abstract: This paper presents a novel
scheme for fast color invariant ball detection in the RoboCup
context. Edge filtered camera images serve as an input for an Ada
Boost learning procedure that constructs a cascade of classification
and regression trees (CARTs). Our system is capable to detect
different soccer balls in the RoboCup and other environments. The
resulting approach for object classification is real-time capable and
reliable.
Abstract:
We discuss online strategies for visibility-based searching for an ob
ject hidden behind a corner, using Kurt3D, a real autonomous mobile
robot. This task is closely related to a number of well-studied
problems. Our robot uses a threedimensional laser scanner in a stop,
scan, plan, go fashion for building a virtual three-dimensional
environment. Besides planning tra jectories and avoiding obstacles,
Kurt3D is capable of identifying objects like a chair. We derive a
practically useful and asymptotically optimal strategy that guarantees
a competitive ratio of 2, which differs remarkably from the
well-studied scenario without the need of stopping for surveying the
environment. Our strategy is used by Kurt3D, documented in a separate
video.
Abstract: To create 3D volumetric maps
of scenes with autonomous mobile robots it is necessary to gage
several 3D scans and to merge them into one consistent 3D model. This
paper provides a new solution to the simultaneous localization and
mapping (SLAM) problem with six degrees of freedom. Robot motion on
natural surfaces has to cope with yaw, pitch and roll angles, turning
pose estimation into a problem in six mathematical dimensions. A fast
variant of the Iterative Closest Points (ICP) algorithm registers the
3D scans in a common coordinate system and relocalizes the
robot. Finally, consistent 3D maps are generated using closing loop
detection and a global relaxation. Keywords: autonomous mobile robots,
3D laser scanner, 3D scan matching, simultaneous localization and
mapping (SLAM), closing loop.
Abstract: We demonstrate how one
of the classical areas of computational geometry has reached practical
application, which in turn gives rise to new, fascinating geometric
problems. In particular, we discuss the problem of developing a good
online strategy for an autonomous mobile robot to locate an object
that is hidden behind a corner or door.
Abstract: A basic issue of mobile
robotics is the automatic generation of environment maps. This paper
presents novel results for the reconstruction of textured 3D maps with
an autonomous mobile robot, a 3D laser range finder and two pan-tilt
color cameras. Building 3D maps involves a number of fundamental
scientific issues. This paper adresses the issue of how to fuse the
geometry data of the 3D laser range finder with camera images. The
proposed algorithm allows to texturize geometrical 3D scenes-models.
Abstract: In this paper, we present a
new recognition system for the fast detection and classification of
objects in spatial 3D data. The system consists of two main
components: A biologically motivated attention system and a fast
classifier. Input is provided by a 3D laser scanner, mounted on an
autonomous mobile robot, that acquires illumination independent range
and reflectance data. These are rendered into images and fed into the
attention system that detects regions of potential interest. The
classifier is applied only to a region of interest, yielding a
significantly faster classification that requires only 30% of the time
of an exhaustive search. Furthermore, both the attention and the
classification system benefit from the fusion of the bi-modal data,
considering more object properties for the detection of regions of
interest and a lower false detection rate in classification.
Abstract: To create with an
autonomous mobile robot a 3D volumetric map of a scene it is necessary
to gage several 3D scans and to merge them into one consistent 3D
model. This paper provides a new solution to the simultaneous
localization and mapping (SLAM) problem with six degrees of freedom.
Robot motion on natural surfaces has to cope with yaw, pitch and roll
angles, turning pose estimation into a problem in six mathematical
dimensions. A fast variant of the Iterative Closest Points algorithm
registers the 3D scans in a common coordinate system and relocalizes
the robot. Finally, consistent 3D maps are generated using a global
relaxation. The algorithms have been tested with 3D scans taken in the
Mathies mine, Pittsburgh, PA. Abandoned mines pose significant
problems to society, yet a large fraction of them lack accurate 3D
maps.
Abstract: 3D models of the skin
surface of patients are created by ultra-fast holography and automatic
scan matching of synchronously recorded holograms. By recording with a
pulsed laser and continuous-wave optical reconstruction of the
holographic real image, motion artifacts are eliminated. Focal analys
is of the real image yields a surface relief of the patient. To
generate a complete 360 patient model, several synchronously recorded
reliefs are registered by automatic scan matching. We find the
transformation consisting of a rotation and a translation that
minimizes a cost function containing the Euclidian distances between
points pairs from two surface relief maps. A variant of the ICP
(Iterative Closest Points) algorithm2 is used to compute such a
minimum. We propose a new fast approximation based on kDtrees for the
problem of creating the closest point pairs on which the ICP algorithm
spends most of its time.
Abstract: This paper presents a new method for object detection and
classification in 3D laser range data that is acquired by an
autonomous mobile robot. Off-screen rendered depth and reflectance
images serve as an input for an Ada Boost learning procedure that
constructs a cascade of classifiers. The performance of the
classification is improved by combining both sensor modalities, which
are independent from external light. The resulting approach for
object classification is real-time capable and reliable. It combines
recent results in computer vision with the emerging technology of 3D
laser scanners.
Abstract: Precise digital 3D models
of indoor environments are needed in several applications, e.g.,
facility management, architecture, rescue and inspection
robotics. This paper presents a new method that transforms a 3D
volumetric model, acquired by a mobile robot equipped with a 3D laser
scanner, into a very precise compact 3D map and generates semantic
descriptions. The scanned 3D scene is matched against a coarse
semantic description of general indoor environments. The matching is
done by a Prolog program compiled from the scanned 3D scene and
combined with clauses from the coarse semantic description. The
generated scene specific knowledge produced by the unification in the
Prolog program is used to refine the 3D model.
Abstract: Precise digital 3D models
of indoor environments are needed in several applications, e.g.,
facility management, architecture, rescue and inspection
robotics. This paper presents a new algorithm that transforms a 3D
volumetric model into a very precise compact 3D map and generates
semantic descri ptions. Our system is composed of a robust, autonomous
mobile robot for the automatic data acquisition and a precise, cost
effective, high quality 3 D laser scanner to gage indoor
environments. The reconstruction method consists of reliable scan
matching and feature detection algorithms. The 3D scene is matched
against a coarse semantic description of general indoor environments
and the generated knowledge is used to refine the 3D model.
Abstract: Digital 3D models of the
environment are needed in facility management, architecture, rescue
and inspection robotics. To create 3D volumet ric models of scenes,
rooms or buildings, it is necessary to gage several 3D scans and to
merge them into one consistent 3D model. This paper presen ts a
system, composed of a fast and robust, autonomous mobile robot, a
precise, cost effective, high quality 3D laser scanner, and reliable
scan mat ching algorithms for measuring and reconstructing
environments, capable of matching two 3D scans within a fraction of a
second. The proposed new sof tware modules for scan matching are fast
variants of the iterative closest point algorithm (ICP) for consistent
alignment. Two applications are presented: First, the reconstruction
of an office environment, second, the fitting of sewer pipes into 3D
data to detect deviations from the spatial geometry.
Abstract: 3D digitalization of
environments without occlusions requires multiple 3D scans. Autonomous
mobile robots equipped with a 3D laser scanner are wel l suited for
the gaging task. They need an efficient exploration scheme for the
digitalization. We present a new approach for planning the next scan
pose as well as the robot motion. Our approach calculates a collision
free trajectory regarding complicated objects, e.g., with jutting out
edges. A closed loop and global ly stable motor control ler navigates
the mobile robot. The results of a 3D digitalization experiment in the
main hall of castle Birlinghoven is presented.
Abstract: In this paper we present
experimental results on a novel application of visual attention
mechanisms for the selection of points of interes t in an arbitrary
scene. The imaging sensor used is a multi-modal 3D laser scanner. In a
single 3D scan pass, it is capable of providing range data as well as
a gray-scale intensity image. The scanner is mounted on top of an
autonomous mobile robot and serves control purposes. We present
results achieved by applying the visual attention system of Itti et
al. [8] to recorded scans of indoor and outdoor scenes. The vast ma
jority of the prima ry attended locations pointed to scene ob jects of
potential interest for navigation and ob ject detection
tasks. Moreover, both sensor modalities c omplement each other,
resulting in a greater variety of points of interest than one modality
alone can provide.
Abstract: In this paper we present
experimental results on a novel application of visual attention
mechanisms for the selection of points of interest in an ar bitrary
scene. The imaging sensor used is a multi-modal 3D laser scanner. In a
single 3D scan pass, it is capable of providing range data as well as
a gray-scale intensity image. The scanner is mounted on top of an
autonomous mobile robot and serves control purposes. We present
results achieved by applying the visual attention system of Itti et
al. [8] to recorded scans of indoor and outdoor scenes. The vast
majority of the primary attended l ocations pointed to scene objects
of potential i nterest for navigation and object detection
tasks. Moreover, both sensor modalities complement each other,
resulting in a greater variety of points of interest than one modality
alone can provide.
Zusammenfassung: Autonome mobile
Roboter müssen in der Lage sein, sicher durch ihre Umgebung zu
navigieren, um anwendungsspezifische Aufgaben ausführen zu
können. Gelingen kann dies nur durch den Einsatz von 3D-Sensoren
und 3D-Karten. Daher ist die automatische und schnelle 3D-Modellierung
der Umgebung eine wichtige Fragestellung in der
Robotik. 3D-Laserscanner sind eine junge Technologie, die die
Erfassung räumlicher Daten revolutioniert und Robotern das
dreidimensionale Abtasten von Objekten mölich macht. Die
vorliegende Arbeit untersucht und evaluiert die zur autonome n
3D-Kartenerstellung notwendigen Algorithmen mit Hilfe des AIS
3D-Laserscanners, der sich auf einer geeigneten Roboterplattform
befindet. Das entwickelte System ermöglicht das berhrungslose
Abtasten der gesamten Umgebung. Dafür werden mehrere 3D-Scans zu
einer konsistenten Szene zusammengefügt sowie Scanpositionen
generiert. 
Abstract: This paper presents a
precise (1cm), lightweight (5kg) and low cost 3D laser range finder
for the fast gaging (1.4 sec) of environments. Real-time algorithms
for the data reduction, 3D-object segmentation are also presented. A
special designed suspension unit, a standard servo motor and a stan
dard 2D range finder are used to build the 3D scanner. Maximal
resolutions e.g. 180 (h) 90 (v) degree with 194400 points are grabbed
in 8.1 seconds and low resolutions with 16200 points are grabbed in
1.4 seconds. While scanning, different online algorithms for line and
surface detection are applied to the data. 3D-Object segmentation and
detection are done offline after the scan. With the proposed approach,
a precise, reliable, mobile, low cost, and real-time capable 3D sensor
for the contact-less measuring of environments without additional
landmarks is available.
Abstract: This paper presents a high
quality, low cost 3D laser range finder designed for autonomous mobile
systems. The 3D laser is built on the bas e of a 2D range finder by
the extension with a standard servo. The servo is controlled by a
computer running RT-Linux. The scan resolution (5 cm) f or a complete
3D scan of an area of 150 (h) 90 (v) degree is up to 115000 points and
can be grabbed in 12 seconds. Standard resolutions e.g. 150 (h) 90 (v)
degree with 22500 points are grabbed in 4 seconds. While scanning,
different online algorithms for line and surface detection are applied
to the data. Object segmentation and detection are done offline after
the scan. The implemented software modules detect overhanging objects
blocking t he path of the robot. With the proposed approach a cheap,
precise, reliable and real-time capable 3D sensor for autonomous
mobile robots is availabl e and the robot navigation and recognition
in real-time is improved. 1. Introduction Prognoses at the beginning
of the nineties claimed for the new millennium a number of about
50.000 independently operating autonomous service robots in different
areas of production and service sectors [1]. The reality is
different. In industrial environments guided vehicles, i.e. vehicles
guided by a magnetic or optical track are standard [2]. Autonomous
mobile service systems, i.e. systems not restricted by a track, are
used extremely rarely although many research groups are working on
this since sev eral years particularly with mobile systems for
transportation tasks, e.g. [3, 4, 5, 6]. One of several reasons for
the gap between prognoses and r eality is the lack of good, cheap and
fast sensors that allow the robots to sense the environment in
real-time and to act on the basis of the acquir ed data. This paper
presents a 3D laser range finder designed for autonomous mobile
systems (fig. 1). A large number of today's autonomous robots us e 2D
laser range finders as a proximity sensor. They are very fast
(processing time 30 ms), precise ( 1 cm, ) and becoming cheaper
($3000) since th ere are at least two competing products [7, 8].
Abstract: After scoring
second in RoboCup Rescue 2004 with Kurt3D and participating in 2005
with two robots, namely Kurt3D and RTS Crawler, we concentrated on the
development of interaction between both vehicles. Kurt3D is able to
autonomously explore and map the environment in 3D and search for
victims. The RTS Crawler is designed to access more demanding
territory (i.e., red arena), working either remote controlled or
semi-autonomous. The key innovation of this system lies in the
capability for autonomous 6D SLAM (simultaneous localization and
mapping) and 3D map generation of natural scenes, combined with the
intelligent cooperation between robots that enables one operator to
efficiently map and explore the whole arena.
Abstract: The mobile robot Kurt3D
is the first robot that is capable of mapping its environment in 3D
and self localize in all six degrees of freedom, i.e., considering its
x, y and z positions and the roll, yaw and pitch angles. Robot
motion on natural surfaces has to cope with yaw, pitch and roll
angles, turning pose estimation into a problem in six mathematical
dimensions. Kurt3D creates a consistent volumetric scene in a common
coordinate frame from multiple 3D laser scans. To create 3D volumetric
maps it is necessary to gage several 3D scans and register them into
one consistent 3D model. A fast variant of the Iterative Closest
Points (ICP) algorithm is used for registration and relocalization.
Abstract: After the second place in
RoboCup Rescue 2004, a new version of the mobile robot Kurt3D was
developed in our groups during the last year [1]. The key innovation
of this system lies in the capability for autonomous or
operator-assisted 6D SLAM (simultaneous localization and mapping) and
3D map generation of natural scenes. Hence, Kurt3D already meets the
basic requirement regarding urban search and rescue. For the rescue
robot league competition, it is additionally configured with dedicated
state-of-the-art equipment e.g. infrared camera and CO2sensor. The
robot and the operator station are rather compact and easy to set
up. The operator uses a joystick as a remote control for the robot and
can watch a live video of the scene where the robot drives. Data are
transmitted via wireless LAN. A 3D laser scanner, which is mounted on
an outdoor variant of Kurt3D, is used as the main sensor for map
generation as well as for navigation and localization. The whole
system has been used with a proven record of success for different
tasks of map building, so that we are confident of managing the rescue
robot league competition, too.
Abstract: A mobile robot named
KURT3D was developed at the Fraunhofer Institute for Autonomous
Intelligent Systems during the last three years. The key innovation of
this system lies in the capability for autonomous or operatorassisted
6D SLAM (simultaneous localization and mapping) and 3D map generation
of natural scenes. Hence, KURT3D already meets the basic requirement
regarding urban search and rescue. For the rescue robot league
competition, it is additionally configured with dedicated
state-of-the-art equipment. The robot and the operator station are
rather compact and easy to set up. The operator uses a joystick as a
remote control for the robot and can watch a live video of the scene
where the robot drives. Data are transmitted via wireless LAN. A 3D
laser scanner, which is mounted on an outdoor variant of KURT3D, is
used as the main sensor for map generation as well as for navigation
and localization. The whole system has been used with a proven record
of success for different tasks of map building, so that we a
reconfident of managing the rescue robot league competition, too.
Zusammenfassung: Autonome mobile Roboter
müssen in der Lage sein, sicher durch ihre Umgebung zu
navigieren, um anwendungsspezifische Aufgaben ausführen zu
können. Gelingen kann dies nur durch den Einsatz von 3D-Sensoren
und 3D-Karten. Daher ist die automatische und schnelle Modellierung
der Umgebung eine wichtige Fragestellung in der
Robotik. 3D-Laserscanner sind eine junge Technologie, die die
Erfassung räumlicher Daten revolutioniert und Robotern das
dreidimensionale Abtasten von Objekten möglich macht. Die
vorliegende Arbeit untersucht und evaluiert die zur autonomen
3D-Kartenerstellung notwendigen Algorithmen mit Hilfe des AIS
3D-Laserscanners, der sich auf einer geeigneten Roboterplattform
befindet. Das entwickelte System ermöglicht dabei das
berührungslose Abtasten der gesamten Umgebung. Der erste Teil der
Arbeit beschäftigt sich mit der Aufgabe, 3D-Scans in einem
globalen Koordinatensystem zu registrieren. Die von der Odometrie des
Roboters geschätzte Pose (Position a und Orientierung) wird dabei
korrigiert. Verschiedene Variationen des iterativen Algorithmus der
nächsten Punkte (ICP) kommen zum Einsatz. Im zweiten Teil geht es
darum, eine möglichst optimale nächste Scanposition zu
bestimmen, von der aus unbekanntes Terrain erforscht werden kann. Ein
randomisierter Approximationsalgorithmus plant die neue Posi tion des
Scanners. Anschliessend ist diese Position durch eine geeignete
Motorregelung anzufahren, wobei Kollisionsvermeidung berucksichtigt
wird. Schliesslich werden die Ergebnisse in geeigneter Weise, unter
anderem durch Gittermodelle, visualisiert. Schlagwörter:
3D-Laserscanner, 3D-Modellierung, 3D-Kartenerstellung, Scanmatching,
iterativer Algorithmus der nchsten Punkte, autonome Exploration,
simultanes Lokalisationsa und Kartierungsproblem, Approximation der
nächsten optimalen Scanposition, Robotersteuerung,
Oberflächenreprsentation durch Gittermodelle.
Zusammenfassung: Diese Ausarbeitung stellt das
Konzept und die Realisierung eines 3D-Laserscanners vor. Ziel der
Realisierung war es, das Dreieck aus Kosten, Geschwindigkeit und
Qualität so zu optimieren, dass der 3D-Scanner auf autonomen mobilen
Robotern sinnvoll zur Exploration und Navigation eingesetzt werden
kann. Ein auf autonomen mobile Fahrzeugen häufig verwendeter
2D-Laserscanner wurde dazu mittels einer selbst konstruierten
Aufhängung und eines Servomotors aufgerüstet. Mittels des auf einem
Standardrechner laufenden Echtzeitbetriebssystem RT-Linux steuert der
Servomotor den 3D Laserscanner direkt an. Unterschiedliche
Online-Algorithmen zur Linienerkennung und Flächendetektion bilden die
Software-Basis des 3D Scanners. Offline-Algorithmen zur
Objektsegmentierung und Erkennung sowie ein 3D-Visualisierungsprogramm
runden das Softwarepaket ab.