Sebastian Zug

An Architecture for a Dependable Distributed Sensor System

In future smart environments, mobile applications will find a dynamically varying number of networked sensors that offer their measurement data. This additional information supports mobile applications in operating faster, with a higher precision and enhanced safety. The potentially increased redundancy obtained in such scenarios, however, is seriously affected by additional uncertainties as well. First, the dependence on wireless communication introduces new latencies and faults, as well as errors, and second, sensors of the environment may be disturbed, of low quality, or even faulty. The quality of the collected data therefore has to be dynamically assessed. Our work aims to provide a generic programming abstraction for fault-tolerant sensors and fusion nodes that cope with varying quality of measurements and communication.

Are laser scanners replaceable by Kinect sensors in robotic applications

Laser scanners are omnipresent in robotic applications. Their measurements are used in many scenarios for robust map building, localization, collision avoidance, etc. But regarding the required precise measurement and mechanical system a laser scanner is quite expensive. Hence the robotic community is looking for alternative sensors. Since 2010 a new 3D sensor system - Microsoft Kinect [1] - developed for computer games is available and applied in robotic applications. With an appropriate filter tool-chain its output can be mapped to a 2D laser scanner measurement. The reduced data set is ready to be processed by the established algorithms and methods developed for laser scanners. But will the Kinect sensor replace laser scanners in robotic applications? This paper compares the technical parameters of the new sensor with established laser scanners. Afterwards we investigate the possibilities and limits of a Kinect for three common robotic applications - map building, localization and obstacle avoidance.

Programming abstractions and middleware for building control systems as networks of smart sensors and actuators

Developing complex sensor/actuator systems, like robot applications, is challenged by a multitude of different hardware platforms, networks, programming languages, data formats, etc. In this paper, we present our architecture that copes with this heterogeneity and allows for a flexible composition of smart sensors and actuators. The main focus lies on a two layered approach combining the communication middleware FAMOUSO and the programming abstraction MOSAIC. FAMOUSO enables the information exchange between networked systems, hides the high degree of heterogeneity on hardware and network level, and is usable from different programming environments. MOSAIC uses FAMOUSO and provides a generic access to the exchanged information. Furthermore, it offers a way to abstract from different sensor and actuator hardware and provides a framework for application development with predefined components, enabling comprehensive fault detection. The paper concludes with a case study that shows how the middleware and programming abstractions are used to build a distributed modular system for a robot manipulator.

Detecting external measurement disturbances based on statistical analysis for smart sensors

The transducer process of a sensor is interference-prone to environmental conditions or external disturbances depending on sensor type, measurement procedure etc. Dependable sensors are characterized by a broad independence of those factors or/and they can both detect situations that make a correct measurement impossible and validate the measurement result. In this paper we describe a statistical approach for the detection of faulty measurements caused by external disturbances. Our fault detection algorithm is based on a comparison of faultless reference measurements with current sensing values. Using this enhancement, a sensor becomes a real smart sensing device and supplies an additional validity estimation of each measurement. The approach was implemented and validated in a demonstration setup that integrates an infrared sensor array disturbed by a strong extraneous light.

An approach towards smart fault-tolerant sensors

Acquisition and processing of sensor data has to cope with measurement uncertainties and complex failure modes. Additionally, multiple sensor types and modalities may be used to improve reliability of environment perception. Our work aims at providing an architecture for fault-tolerant sensors and offering a uniform interface to the application. In the paper, we present our fault-tolerant virtual sensor concept that is based on combining model-based estimation and redundant sensor data. To illustrate and evaluate our concept we simulate a mobile robot in an instrumented environment which integrates several smart position sensors. By using a mathematical model to evaluate sensor data we achieve a more reliable position estimation. The paper presents results of the fusion process and discusses methods for generalization.

Validity-Based Failure Algebra for Distributed Sensor Systems

Distributed applications dealing with data from networked sensors need some indication about the quality of remote information. This paper describes how to derive a dynamic validity value that represents a measure for the confidence in remote sensor data. In contrast to conventional systems which treats a typical processing chain as a whole, this paper describes how individual characteristic of sensing, detection and filter mechanisms can be assessed and how this assessment can be evaluated to a single validity value. In particular the paper defines respective operations combining the run-time validity estimates of every stage in a processing chain. The combination is evaluated by rules as part of a failure algebra. The paper presents the generation of an application specific validity value which is finally demonstrated in a robotic application.

Assessing neural networks for sensor fault detection

The idea of “smart sensing” includes a permanent monitoring and evaluation of sensor data related to possible measurement faults. This concept requires a fault detection chain covering all relevant fault types of a specific sensor. Additionally, the fault detection components have to provide a high precision in order to generate a reliable quality indicator. Due to the large spectrum of sensor faults and their specific characteristics these goals are difficult to meet and error prone. The developer manually determines the specific sensor characteristics, indicates a set of detection methods, adjusts parameters and evaluates the composition. In this paper we exploit neural-network approaches in order to provide a general solution covering typical sensor faults and to replace complex sets of individual detection methods. For this purpose, we identify an appropriate set of fault relevant features in a first step. Secondly, we determine a generic neural-network structure and learning strategy adaptable for detecting multiple fault types. Afterwards the approach is applied on a common used sensor system and evaluated with deterministic fault injections.

Exploiting the FAMOUSO middleware in multi-robot application development with Matlab/Simulink

We describe a framework for the development of distributed systems combining real and virtual components, sensors and actuators. We show the benefits of our approach for the development and validation of multi robot applications. Based on our middleware, which provides a flexible communication for distributed systems, virtual and real components are seamlessly exchangeable during different development steps. This modularity and compatibility allows appropriate adjustments for design, rapid prototyping and examination as soon as an opportunity to reduce the hardware effort for large scenarios.

Visualization of robot's awareness and perception

Today, direct interaction between humans and robots is limited, although the combination of human flexibility and robots power enables a growing productivity. The problem for humans lies in the nearly unpredictable behavior and motion of the robot itself. However, we can enhance humans view with more information to get knowledge about robots perception and awareness. We use Augmented Reality methods for providing the information in an adaptable visualization for different user types. We show that our approach leads to shorter development cycles as well as to safer human-robot interaction.

Sentient Objects for Designing and Controlling Service Robots

Abstract Services related to healthcare and the support for elderly people become more and more important. Autonomous or semi-autonomous robots may play an important role in this area. From a control system point of view these robots are networks of distributed smart components to perceive their environment and react on it in real time. The problem of developing or extending such a robot often is that the designer has to start from scratch struggling with low level issues, where reusability of already designed components would be highly desirable. The paper describes a robot application in the area of a meals distribution service that combines two design worlds. One is the conventional world of modelling the functional properties without any structural considerations, the other is the world of cooperating sentient objects. We explain how the notion of sentient objects will assist the design, simulation and also later extensions and adaptations of the robot.