Volume 57, Issue 4 - Special Issue: Autonomous Driving / J. Marius Zöllner, Thomas Schamm

The term safe state is not yet defined accurately in the context of automated road vehicles. In this paper we derive several conditions, which are in our opinion important for the discussion of a safe state for automated road vehicles. From an engineering view, several concepts on the classification of automation levels are analyzed and from their results and current research and development approaches a definition of a safe state is proposed. Especially the relation between risk and safety for passengers of automated vehicles and other traffic participants is considered. Several use cases for automated vehicles and events with impact to safety are discussed to identify conditions for a safe state in public traffic, which have to be fulfilled to keep the risk of operating an automated vehicle below a reasonable level.

Automated driving on public roads is affected by many foreseeable and unforeseeable driving situations. Depending on the driving task, the environmental and road conditions, and the behavior of other drivers, different actions have to be taken. This paper provides a high-level overview of the development of highly automated driving systems and illustrates challenging situations and use cases. We outlined the impact of these use cases on system design, key technologies, and their technical realization for a highly automated driving system. Furthermore, the paper demonstrates how certain aspects of the system design as well as their implementation are country specific and how continuous testing is required for robust implementation of the functionalities.

In this work an overview of the local motion planning and dynamic perception framework within the V-Charge project is presented. This framework enables the V-Charge car to autonomously navigate in dynamic mixed-traffic scenarios. Other traffic participants are detected, classified and tracked from a combination of stereo and wide-angle monocular cameras. Predictions of their future movements are generated utilizing infrastructure information. Safe motion plans are acquired with a system-compliant sampling-based local motion planner. We show the navigation performance of this vision-only autonomous vehicle in both simulation and real-world experiments.

The automotive innovations in the last decades are dominated by new software-based functions. A validation of these functions is always needed for ensuring the safety of the vehicles. XiL-methods are widely used for this purpose. Along with the development of automated driving vehicles, the demand for simulation based validation of such vehicles is rising, as shown by examples. For the simulations, the modeling of the perception sensors plays an essential role. A state-of-the-art survey clarifies, that a dilemma between simulation efficiency and fidelity of the current methods exists, especially when modeling active sensors. A novel concept of grey box modeling promises to solve the dilemma, or at least to defuse it. With a generalized sensor structure and a simplified propagation modeling, the grey box model considers the most important physically based performances of the sensors and their ranges. The basic structure of the modeling is sketched here, and an outlook for future requirements on it is presented.

Better understanding and controlling complex systems has become a grand challenge not only for computer science, but also for the natural and social sciences. Many of these systems have in common that they can be studied from a network perspective. Consequently methods from network science have proven instrumental in their analysis. In this article, I introduce the macroscopic perspective that is at the heart of network science. Summarizing my recent research activities, I discuss how a combination of this perspective with Big Data methods can improve our understanding of complex systems.

This thesis addresses the timely concern of protecting privacy in the age of big data. We identify the two following problems as the fundamental problems in computational privacy: (i) consistently quantifying privacy in different systems and (ii) optimally protecting privacy using obfuscation mechanisms. We cast the problem of quantifying privacy as computing the estimation error in a statistical (Bayesian) inference problem, where an adversary combines his observation, background knowledge and side channel information to estimate the user's sensitive information. This enables us to evaluate privacy of users in different systems, and consistently compare the effectiveness of different privacy protection mechanisms. We also formulate the problem of optimizing user privacy while respecting data utility as an interactive optimization problem (Bayesian Stackelberg game), where both user and adversary want to maximize their own objectives which are in conflict with each other. We apply our methodologies to quantifying and protecting location privacy in location-based services. We also provide an open-source tool, named Location-Privacy and Mobility Meter (LPM), that enables researchers to learn and analyze human mobility models as well as evaluating and comparing different location-privacy preserving mechanisms.