Christian Kuka

Quality matters: supporting quality-aware pervasive applications by probabilistic data stream management

Many pervasive computing applications need sensor data streams, which can vary significantly in accuracy. Depending on the application, deriving information (e.g., higher-level context) from low-quality sensor data might lead to wrong decisions or even critical situations. Thus, it is important to control the quality throughout the whole data stream processing, from the raw sensor data up to the derived information, e.g., a complex event. In this paper, we present a uniform meta data model to represent sensor data and information quality at all levels of processing; we show how this meta data model can be integrated in a data stream processing engine to ease the development of quality-aware applications; and we present an approach to learn probability distributions of incoming sensor data which needs no prior knowledge. We demonstrate and evaluate our approach in a real-world scenario.

SaLsA Streams: Dynamic Context Models for Autonomous Transport Vehicles Based on Multi-sensor Fusion

Due to the fact that currently operating autonomous vehicles can observe only a limited area with their onboard sensors, safety regulations often dictate a very slow speed. However, as more and more sensors in the environment are available, we can fuse their information and provide extended information as a shared context model to support the autonomous vehicles. In this paper, we consider a scenario with a publicly accessible area that is populated with autonomous transport vehicles, human guided vehicles like trucks or bicycles, and pedestrians. We analyze requirements and challenges for highly dynamic context models in this scenario. Furthermore, we propose a comprehensive system architecture that can cope with these challenges, namely deterministic processing of multiple sensor updates with high throughput rates, prediction of moving objects, and on-line quality assessments, and demonstrate the feasibility of this approach by implementing the generic system architecture with laser scanners for object detection.

Context-model generation for safe autonomous transport vehicles

Autonomously operating vehicles highly depend on the quality of its sensors as they have to be aware of its surroundings to react appropriately. Currently operating automated guided vehicles cover only a limited area with their sensors and therefore can only drive at low speeds. However, as more and more sensors are available, it is essential to build a context-model, which fuses information of different sensors to cover a larger area and allow for an increased level of autonomy. In this paper, we present a context-model based on a Bayesian occupancy filter which can be queried via a data stream management system in order to provide the necessary information at any point in time. Additionally, the Bayesian filter is pessimistically as it is constructed such that probability of occupancy is always upper bounded, to ensure a sufficient level of safety.

Enriching sensor data processing with quality semantics

Sensors and their observations are used in almost all pervasive applications to configure and adjust the behavior of applications to the users need. However, the quality of those sensor observations are influenced by different factors including the physical or chemical principle of measurement, the internal processing of the sensing device, and the prevailing environmental conditions at the time of measurement. Thus, it is of great interest to not just measure and process the sensor observation but to also handle the quality of the sensor observation correctly and propagate the quality of the observation along the processing path to the user. To do so, we use the Semantic Sensor Network Ontology (SSN) to combine necessary sensor observations from multiple sources in a Probabilistic Data Stream Management System (PDSMS) to estimate prevailing conditions and propagate current quality information.

Supporting quality-aware pervasive applications by probabilistic data stream management

Many pervasive computing applications need sensor data streams, which can vary significantly in accuracy. Depending on the application, deriving information (e.g., higher-level context) from low-quality sensor data might lead to wrong decisions or even critical situations. Thus, it is important to control the quality throughout the whole data stream processing, from the raw sensor data up to the derived information, e.g., a complex event. In this paper, we describe the demonstration of the integration of a uniform metadata model to represent sensor data and information quality at all levels of processing in a data stream processing engine to ease the development of quality-aware applications.

Approximating complex sensor quality using failure probability intervals

Many pervasive applications depend on data from sensors that are placed in the applications physical environment. In these applications, the quality of the sensor data--e.g., its accuracy or a failed object detection--is of crucial importance for the application knowledge base and processing results. However, through the increasing complexity and the proprietary of sensors, applications cannot directly request information about the quality of the sensor measurements. However, an indirect quality assessment is possible by using additional simple sensors. Our approach uses information from these additional sensors to construct upper and lower bounds of the probability of failed measurements, which in turn can be used by the applications to adapt their decisions. Within this framework it is possible to fuse multiple heterogeneous indirect sensors through the aggregation of multiple quality evidences. This approach is evaluated using sensor data to detect the quality of template matching sensors.

Processing the uncertainty: Quality-aware data stream processing for dynamic context models

Current autonomic vehicles in outdoor scenarios perform mobility operations with walking speed to ensure safety. For a faster mobility of autonomous vehicles, concrete knowledge of the environment is needed. This will be achieved through a dynamic context model based on sensor data with uncertainties from the environment. These uncertainties arise through existential uncertainty, consistency, and co/variance of and between the sensor data. To allow a flexible processing and to allow different approaches for object detection, object classification, and object tracking, data stream management technology is used. Therefore, a new algebra and operators based on the relational algebra are defined to preserve and process the uncertainties about the sensor data.

Safe Autonomous Transport Vehicles in Heterogeneous Outdoor Environments

Autonomous transport vehicles (AGVs) steadily gain importance in logistics and factory automation. Currently, the systems are mainly operating in indoor scenarios at limited speeds, but with the evolution of navigation capabilities and obstacle avoidance techniques, AGVs have reached a degree of autonomy that, from a technical perspective, allows their operation beyond closed work environments. The major hurdle to overcome is to be able to guarantee the required safety level for industrial applications. In this paper, we propose a general architecture for AGVs that formalizes the current safety concept and extends it to vehicles driving at higher speeds in outdoor environments. Technically, the additional safety level is achieved by integrating information from stationary sensors in order to increase the perception of the vehicles.