While the knowledge about the structures of a system’s underlying causal relationships is crucial within many real-world scenarios, the omnipresence of heterogeneous data characteristics impedes applying methods for causal structure learning (CSL). In this dissertation project, we reduce the barriers for the transfer of CSL into practice with threefold contributions: (1) We derive an information-theoretic conditional independence test that, incorporated into methods for CSL, improves the accuracy for non-linear and mixed discrete-continuous causal relationships; (2) We develop a modular pipeline that covers the essential components required for a comprehensive benchmarking to support the transferability into practice; (3) We evaluate opportunities and challenges of CSL within different real-world scenarios from genetics and discrete manufacturing to demonstrate the accuracy of our approach in practice.

The examination of causal structures is crucial for data scientists in a variety of machine learning application scenarios. In recent years, the corresponding interest in methods of causal structure learning has led to a wide spectrum of independent implementations, each having specific accuracy characteristics and introducing implementation-specific overhead in the runtime. Hence, considering a selection of algorithms or different implementations in different programming languages utilizing different hardware setups becomes a tedious manual task with high setup costs. Consequently, a tool that enables to plug in existing methods from different libraries into a single system to compare and evaluate the results is substantial support for data scientists in their research efforts.

Learning the causal structures from high-dimensional observational data is an omnipresent challenge in data science.State-of-the-art methods for constraint-based Causal Structure Learning (CSL) apply conditional independence (CI) tests to determine the underlying causal structures.In the context of discrete data, each CI test requires calculating the marginals over contingency tables based on the respective observations.This calculation leads to long overall execution times.In our work, we propose a parallel execution strategy tailored for constraint-based CSL on a Graphics Processing Unit (GPU) to accelerate the execution for discrete data.Hence, we introduce the gpuPC algorithm that performs all CI tests on a GPU and extends the existing parallel execution strategy for constraint-based CSL by calculating the marginals over contingency tables within units of threads.Further, gpuPC implements explicit memory management to handle the corresponding auxiliary data structures in GPU memory.An experimental evaluation shows that gpuPC scales well even for higher-dimensional settings, with auxiliary data structures exceeding on-chip memory.In particular, running on NVIDIA Tesla V100 hardware gpuPC outperforms a GPU baseline by factors of up to 45.6 and further outperforms existing parallel CPU-based implementations running on 40 cores by a factor of 62.1.

The efficiency of modern automotive body shop assembly lines is highly related to the reduction of downtimes due to failures and quality deviations within the manufacturing process. Consequently, the need for implementing tools into the assembly lines for on-line monitoring, and failure diagnosis, also under the prism of improving the troubleshooting, is of great importance. While the identification of root causes and elimination of failures is usually built upon individual on-site expert knowledge, causal graphical models (CGMs) have opened the possibility to make a purely data-driven assessment. In this demo, we showcase how a CGM of the production process is incorporated into a monitoring tool to function as a decision-support system for an operator of a modern automotive body shop assembly line and enables fast and effective handling of failures and quality deviations.

Classic off- and online programming approaches offer different advantages but cannot provide intuitive programming as it would be needed for frequent reconfiguration of industrial robotic cells and their programs. While a simulation and therewith a collision control can be used offline, highest precision can be achieved by using online teach-in. We propose a system that combines the advantages of both offline and online programming. By creating a programming framework that is independent of in- and output devices we allow the use of augmented as well as virtual reality. Thus, enabling the user to choose for a programming technique that best suits his needs during different stages of programming. Bringing together both methods allows us to not only drastically reduce programming time but simultaneously increase intuitiveness of human robot interaction.

In this paper we present an automated and precise calibration approach to enable a vision-based robot control with a multi-camera setup. In our case we use a setup consisting of four fixed cameras above the workplace (eye-to-hand configuration) and a collaborative robot. The robot will be used to automate the intrinsic camera calibration and the necessary process of capturing images of a calibration pattern from a maximum number of different viewpoints. For the extrinsic calibration of the cameras and the calibration to the robot we use only two small markers, which are permanently attached to the base of the robot. This allows a fast online calibration of the setup. With the help of first experiments, we can show that the calibration method works well under laboratory conditions. Furthermore, we will discuss the achieved accuracy.

With the demand to process ever-growing data volumes, a variety of new data stream processing frameworks have been developed. Moving an implementation from one such system to another, e.g., for performance reasons, requires adapting existing applications to new interfaces. Apache Beam addresses these high substitution costs by providing an abstraction layer that enables executing programs on any of the supported streaming frameworks. In this paper, we present a novel benchmark architecture for comparing the performance impact of using Apache Beam on three streaming frameworks: Apache Spark Streaming, Apache Flink, and Apache Apex. We find significant performance penalties when using Apache Beam for application development in the surveyed systems. Overall, usage of Apache Beam for the examined streaming applications caused a high variance of query execution times with a slowdown of up to a factor of 58 compared to queries developed without the abstraction layer. All developed benchmark artifacts are publicly available to ensure reproducible results.

Learning the causal structures in high-dimensional datasets allows deriving advanced insights from observational data, thus creating the potential for new applications. One crucial limitation of state-of-the-art methods for learning causal relationships, such as the PC algorithm, is their long execution time. While, in the worst case, the execution time is exponential to the dimension of a given dataset, it is polynomial if the underlying causal structures are sparse. To address the long execution time, parallelized extensions of the algorithm have been developed addressing the Central Processing Unit (CPU) as the primary execution device. While modern multicore CPUs expose a decent level of parallelism, coprocessors, such as Graphics Processing Units (GPUs), are specifically designed to process thousands of data points in parallel, providing superior parallel processing capabilities compared to CPUs.

This paper is a work in progress report on a novel system for intuitive gesture based robot programming. The major contribution is the addition of an expert system into the robot programming process. The expert system uses static knowledge, such as seam types, and dynamic knowledge to reason about the intended process. The dynamic knowledge holds information about the environment and is derived from sensor data. The paper outlines the required components and the current state of development of a prototype system. An example of the inference process is given for a robot laser welding application.

We present a hybrid programming method for industrial robots combining advantages of manual haptic guidance of the end-effector and programming approaches using non-haptic pointing gestures for the spatial definition of poses and trajectories. Whereas the bare-hand spatial interaction can be implemented and performed cost- and time-efficiently but lacks accuracy, haptic-interaction is more time-consuming but it is used in a reduced manner in order to enable a highly-accurate refinement of target working poses. Additionally, the user is supported by a mobile Augmented Reality simulation providing spatial validation of the robot program, program management and transmission towards the robot controller. The implementation is realized by a compliance control based on a sensor mounted between flange and end-effector combined with our former introduced approach for spatial programming. We conducted a user study comparing Teach-In and Offline programming. The analysis shows a significant reduction of programming duration as well as a reduction of programming errors compared with Teach-In. Most participants favor the hybrid programming system. No significant differences for the programming duration could be determined between experts and non-experts. In comparison between haptic and non-haptic interaction, non-experts favor non-haptic interaction due to the higher intuitiveness of pointing gestures compared to direct physical interaction.