Research Training Group 1194 "Self-organizing Sensor-Actuator-Networks"
search
Research Training Group 1194 "Self-organizing Sensor-Actuator-Networks"
Research Training Group 1194 "Self-organizing Sensor-Actuator-Networks"
Mainpage
Lectures

*VORLESUNG*

  • type:

Ringvorlesung: Algorithmic Models for Wireless Networks

  • Date:

    03.02.2014, 17:30 Uhr

  • Speaker:

    Prof. Magnus Halldorsson, School of Computer Science, Reykjavik University, Island

  • Abstract:
    Wireless communication continues to have tremendous impact. From an algorithmic point of view, a major challenge in wireless computing is to find general models of interference that are both realistic and analytically feasible. Recent years have seen significant interest in studying algorithms in the so-called SINR model, where interference adds up and fades with distance.
    The basic SINR model still makes unrealistic assumptions that hold only in an idealistic situation. In particular, a major open issue has been to model environments with walls and obstacles. We introduce very recent approach that allows for arbitrary static environments while maintaining the performance guarantees that have been obtained for the basic SINR model. This might be the first algorithmic model of wireless computing that captures reality with high fidelity while maintaining generality and analytic feasibility.
     
  • Place:

    Geb. 50.34, R -101

Ringvorlesung: Rigid Motion Estimation using Mixtures of Projected Gaussians

  • Date:

    31.01.2014, 11:30 Uhr

  • Speaker:

    Dr. Wendelin Feiten

  • Abstract:
    Modeling the position and orientation in three-dimensional space is important in many applications. In robotics, the position and orientation of objects as well as the rigid motions of robots are derived from sensor data that are uncertain. The uncertainties of these sensor data result in position and orientation uncertainties that can be very widely spread or have several peaks.
    In this talk, we describe a class of probability density functions (pdf) on the group of rigid motions that allows for modeling wide-spread and multi-modal pdf and offers most of the operations that are available for the mixtures of Gaussians on Euclidean space. The use of this class of pdf is illustrated with an example from robotic perception
  • Place:

    Geb. 50.34, R -120

Ringvorlesung: Large-Scale Data Collection for Human Contact Network Research

  • Date:

    24.01.2014, 11:30 Uhr

  • Speaker:

    Prof. Thomas Schmid

  • Abstract:
    Advances in electronic components have driven the trend in development of adaptive and highly configurable sensor systems, such as the electronically steered array antenna. The electronically steered array antenna enables an agile radar beam, allowing dynamic allocation of a radar system’s time/energy budget between multiple radar tasks. This creates the radar resource management problem, which aims to optimally allocate the finite resource between radar tasks whilst also optimizing the operation of each individual radar task. In this talk, an overview of solutions to radar resource management will be given, including conventional rule based methods and the Quality of Service Resource Allocation Method (Q-RAM). The Continuous Double Auction Parameter Selection (CDAPS) algorithm, which utilizes a synthesized market mechanism to solve the radar resource management problem, will be shown to improve upon the rule based and Q-RAM methods. Simulated results will be presented, indicating that both Q-RAM and CDAPS enable substantial performance improvements for a multi-function radar system in contrast to the conventional rule based methods.

Ringvorlesung: Radar Resources Management Using Market Mechanisms in Agent Systems

  • Date:

    17.01.2014, 11:30 Uhr

  • Speaker:
    Dr. Alexander Charlish, Fraunhofer Institute for Communication, Information Processing and Ergonomics (FKIE)
  • Abstract: Advances in electronic components have driven the trend in development of adaptive and highly configurable sensor systems, such as the electronically steered array antenna. The electronically steered array antenna enables an agile radar beam, allowing dynamic allocation of a radar system’s time/energy budget between multiple radar tasks. This creates the radar resource management problem, which aims to optimally allocate the finite resource between radar tasks whilst also optimizing the operation of each individual radar task. In this talk, an overview of solutions to radar resource management will be given, including conventional rule based methods and the Quality of Service Resource Allocation Method (Q-RAM). The Continuous Double Auction Parameter Selection (CDAPS) algorithm, which utilizes a synthesized market mechanism to solve the radar resource management problem, will be shown to improve upon the rule based and Q-RAM methods. Simulated results will be presented, indicating that both Q-RAM and CDAPS enable substantial performance improvements for a multi-function radar system in contrast to the conventional rule based methods.
  • Place:
    Geb. 50.34, R -120

Ringvorlesung: Suboptimal Kalman Filter for Parameter Estimation under Dynamical Uncertainties

  • Abstract:
    Kalman filter (KF) is an online optimal estimation method that recovers the true states of the system of interest from a series of the noise-corrupted observation data. However, due to significant disparity between real-world and filter-world (models), for example, a case that system parameter dynamics includes uncertainties such as abrupt changes in time, KF often loses adaptability and tends to output the state estimates that are considerably deviated from the true values. For this kind of problems, to suppress the ill behavior of the filter, several covariance inflation methods have been proposed to enlarge the state estimation error covariance matrix, e.g. fictitious system noise, fading-memory filter, H-infinity filter, and the like. However, we can derive a rather new filter retaining almost the similar effect but with other advantages by making the KF deliberately suboptimal in the frame of square errors. We call this new filter as the suboptimal Kalman filter (SKF). Since the SKF is much easier in setting filter parameters than the existing covariance inflation methods, it will provide more capable filtering method for the aforementioned complex and uncertain problems. In the SKF the nonlinearity of system models is handled by using an unscented transforming statistical linearization (UTSL) without losing the Gaussianity of filter’s states, which is the basic prerequisite behind the SKF design. We show the effectiveness of the proposed filter by numerical simulations using artificially generated data.

Ringvorlesung: Using Realistically Simulated Undersea Channels to Evaluate the Performance of Authenticating Sonar

  • Venue:

    Fraunhofer IOSB, Max-Syrbe-Saal (2. OG), Fraunhoferstraße 1, 76131 Karlsruhe

  • Date:

    30.10.2013, 10:00 Uhr

  • Speaker:

    Dr. Robert S. Lynch, Analytic Information Fusion Systems, USA

  • Abstract:
    Undersea communication channels are filled with acoustic emissions of various kinds. From sonar to the signals used in acoustic communications, man-made noise, and biological signals generated by marine life, the ocean is a complex conduit for diverse emissions. In this talk, we discuss an algorithm in which an acoustic emission such as a sonar signal is transparently and securely embedded with signatures known as a digital watermark. Extracting the watermark helps to distinguish, for example, a friendly sonar from other acoustic emissions that may exist as part of the natural undersea environment, or from pings that may have originated from hostile forces or echoes fabricated by an adversary. We have adopted spread spectrum as an embedding technique. Spread spectrum allows for matching the watermark to propagation, multipath, and noise profiles of the channel. The sonar is first characterized by its spectrogram and divided up into non-overlapping blocks in time. Each block is individually embedded with a single bit drawn from the watermark payload. The seeds used to generate the spreading codes are the keys used by authorized receivers to recover the watermark. The detector is a maximum likelihood detector using test statistics obtained by integrating a correlation detector output over the entire sonar pulse width. Performance of the detector is controlled by signal-to-watermark ratio, specific frequency bands selected for watermarking, watermark payload, and processing gain. For validation, we use Sonar Simulation Toolset (SST). SST is a software tool that is custom-made for the simulation of undersea channels using realistic propagation properties in oceans. Probabilities of detection and false alarm rates, as well as other performance boundaries, are produced for a shallow water channel subject to multipath and additive noise.

Ringvorlesung: On Moment Problems in Robust Control, Spectral Estimation, Image Processing and System Identification

  • Venue:

    Gebäude 50.20, Raum 148 (1. OG), Adenauerring 2, 76131 Karlsruhe

  • Date:

    29.10.2013, 10:00 Uhr

  • Speaker:

    Prof. Anders Lindquist, Royal Institute of Technology (KTH), Sweden

  • Abstract:
    Moment problems are ubiquitous in both mathematics and engineering. Such problems are typically underdetermined and give rise to families of particular solutions. Therefore finding a solution that also satisfies a natural optimality criterion or design specification is an important general problem.
Many problems in circuit theory, power systems, robust control, signal processing, spectral estimation, statistical modeling, image processing and identification lead to a nonclassical version of the moment problem reflecting the importance of rational functions in engineering applications.
Although this version of the problem is nonlinear, there exists a natural, universal family of strictly convex optimization criteria defined on the convex set of particular solutions. This provides a powerful paradigm for smoothly parameterizing, comparing and shaping the solutions based on various additional design criteria and enables us to establish the smooth dependence of solutions on problem data.

     

    Plakat
     

Ringvorlesung: Application Areas for Classifier Fusion

  • Venue:

    Fraunhofer IOSB, Max-Syrbe-Saal (2. OG), Fraunhoferstraße 1, 76131 Karlsruhe

  • Date:

    28.10.2013, 10:00 Uhr

  • Speaker:

    Dr. Robert S. Lynch, Analytic Information Fusion Systems, USA

  • Abstract:
    In this talk, robust methods of information fusion and classification are discussed, given all relevant information must be learned (supervised, unsupervised) from data. In this work, emphasis is on utilizing both feature
    level (extracted information from data for reducing bandwidth - can result in information loss) and decision level (most basic representation of data - can result in largest degree of information loss) fusion to make target present/absent decisions. Various application areas are illustrated using both real and simulated data, and with an emphasis on modeling information from sensors, waveforms, and images. The modeling of a fusion processor is based on a discrete Bayesian classifier, and its extensions, which leads to interesting quantitative and qualitative results about expected performance with data driven information fusion and classification.

     

    Plakat

Ringvorlesung: Learning the Cooperative Navigation Behavior of Pedestrians

  • Date:

    Freitag, 21.06.2013

  • Speaker:

    Markus Kuderer

  • Time:

    11:30 Uhr

  • Abstract:
    We present an approach to learn a model of the navigation behavior of interacting pedestrians from observed trajectories. Such models can be applied in many areas including robotics, computer graphics, and behavioral science. Many existing methods either neglect the cooperative behavior of pedestrians by modeling them independently of each other or apply reactive controllers, which ignores their predictive abilities. Moreover, many models rely on grid-based representations of the trajectories which do not scale well to real-world applications, especially when taking into account higher order properties of the trajectories such as velocities and accelerations. We propose an approach to learn the cooperative navigation behavior of interacting pedestrians. Therefore, we aim to infer the intent of the agents in terms of relevant features of their composite trajectories that explains the observed behavior. Our approach uses a mixture distribution that comprises a discrete distribution over the homotopy classes, which correspond to the pedestrians' decisions to pass each other on the left or on the right, and continuous distributions over the composite trajectories for each homotopy class. To learn the model parameters of this mixture distribution that match the observed behavior, we utilize a maximum entropy inverse reinforcement learning approach. This requires computing the feature expectations under the continuous distributions over high-dimensional composite trajectories. To this end, our method utilizes Hamiltonian Markov chain Monte Carlo sampling and exploits that the distributions over observed trajectories are highly structured due to physical constraints. An experimental evaluation suggests that our method is able to capture human navigation behavior more accurately than state-of-the-art methods.

3D Traffic Scene Understanding from Movable Platforms

  • Date:

    Freitag, 03.05.2013

  • Speaker:

    Andreas Geiger

  • Time:

    11:30 Uhr

  • Abstract:
    In this work we present a novel probabilistic generative model for multi-object traffic scene understanding from movable platforms which reasons jointly about the 3D scene layout as well as the location and orientation of objects in the scene. In particular, the scene topology, geometry and traffic activities are inferred from short video sequences. Inspired by the impressive driving capabilities of humans, our model does not rely on GPS, lidar or map knowledge. Instead, it takes advantage of a diverse set of visual cues in the form of vehicle tracklets, vanishing points, semantic scene labels, scene flow and occupancy grids. For each of these cues we propose likelihood functions that are integrated into a probabilistic generative model. We learn all model parameters from training data using contrastive divergence.

    Experiments conducted on videos of 113 representative intersections show that our approach successfully infers the correct layout in a variety of very challenging scenarios. To evaluate the importance of each feature cue, experiments using different feature combinations are conducted.

    Furthermore, we show how context from the proposed method is able to improve over the state-of-the-art in terms of object detection and object orientation estimation in challenging and cluttered urban environments.