The subsequently listed PhD theses are currently being performed within the Automotive IUI group or have been completed as part of this group.
Robert Neßelreath: „An open platform for the model-based development of dialogue applications with distributed input and output in cyber-physical systems“
Cyber-Physical systems integrate computational elements into the physical world. An increasing number of interconnected human-computer interfaces, sensors and actuators in the intelligent environment enable new HCI concepts supporting multimodal distributed input and output.
The major contribution of the dissertation is the SiAM dialogue platform (SiAM-dp) that contains a fully tool-supported development environment for multimodal dialogue applications with high flexibility in the mode and number of connected devices. A model-based approach supports the rapid creation of new multimodal user interfaces independently from the domain and applied devices.
Introducing smart and modern interaction concepts requires a system that supports both the information from various input modalities as well as the description of physical acts. Additionally the actual context of the dialogue, participants, and the environment can provide valuable information for the resolution of uncertain, missing, or implicitly given content or can influence the course of the further interaction. The thesis analyzes how the semantic representation of knowledge, user intentions, and events can be exploited for the fusion of multimodal user input, physical acts, and context knowledge.
A further contribution of the thesis is the development of strategies for the situation dependent selection of adequate output devices and distribution of output acts.
M. Mehdi Moniri: „Real-Time Multimodal Reference Resolution in Indoor and Outdoor Environments“
Within the scope of the dissertation of Mohammad Mehdi Moniri different algorithms and methods for driver monitoring, vehicle positioning and fusion of various sensor sources are developed. This system will make it possible to determine exactly which objects in the environment (at the given time) were visible to the driver and for how long. In addition, this system conducts a scene analysis for different driving situations by fusion of different information sources such as driver’s focus-of-attention and road context. This research prototype aims to achieve a significant increase in the safety of road and to improve the usability of infotainment systems.
Catalin Barbu: „Facilitation of Learning-While-Choosing as a Form of Choice Support“
Within computer science, many approaches to supporting choice and decision making have been developed, but one strategy has been almost entirely neglected: Given that many choices that people make are faced repeatedly, it makes sense not only to help a person make the choice that they are currently facing but also to learn how to make such choices better over time. Pursuing this strategy requires the development of forms of interaction and algorithms that are complementary to those that are already widespread. In his thesis, Catalin-Mihai Barbu is formulating a theoretical basis for this approach, building on the conceptual framework of “choosability engineering” being developed at DFKI by Dr. Anthony Jameson. He is instantiating this framework with choice support systems that represent various ways of supporting “learning while choosing”.
Sandro Castronovo (2013): „The Pull Paradigm : foundations of user-centric advanced driver assistant systems based on bidirectional car2X communication“
This thesis develops applications for vehicular ad-hoc networks that go far beyond the currently established areas of driving safety and traffic efficiency. The ad-hoc network is regarded as a dynamic information resource which is available to any vehicle at any time. In contrast to current state-of-the-art research, the proposed Pull Paradigm starts at the user’s vehicle rather than at an information source somewhere in the network, e.g. a braking car. To access information from highly dynamic ad-hoc networks, bidirectional communication and information discovery and retrieval play a vital role. Therefore, in the course of the work, the applicability of the Pull Paradigm to established vehicular ad-hoc networks is thoroughly examined and missing aspects are identified. It turns out that a number of enhancements to almost all layers of the network stack are necessary in order to apply the Pull Paradigm using existing technology. The central elements here are two novel algorithms for managing information flow and dissemination in ad-hoc networks, which are at first formulated from the abstract perspective of graph theory. Using the knowledge gained leads to the development of PADE, a platform that supports development of vehicular ad-hoc network applications. The designed algorithms are then implemented as a routing scheme, integrated and evaluated in large, simulated city scenarios. Furthermore, PADE combines “real” and simulated communication technologies and abstracts from them, so that applications can be transferred from the lab into a test vehicle with minimal effort. In order to achieve this ambitious goal, PADE builds on a number of existing simulation and communication technologies. The practical applicability of the Pull approach is shown in two demonstrators that are integrated into a BMW 5 series test vehicle. The presentation module of the PADE platform was tested in the currently largest field operational test for vehicular ad-hoc communication. Over 400 drivers in 120 vehicles experienced the system on a daily basis.
Christoph Endres (2012): „PRESTK : situation-aware presentation of messages and infotainment content for drivers“
The amount of in-car information systems has dramatically increased over the last few years. These potentially mutually independent information systems presenting information to the driver increase the risk of driver distraction. In a first step, orchestrating these information systems using techniques from scheduling and presentation planning avoid conflicts when competing for scarce resources such as screen space. In a second step, the cognitive capacity of the driver as another scarce resource has to be considered.
For the first step, an algorithm fulfilling the requirements of this situation is presented and evaluated. For the second step, I define the concept of System Situation Awareness (SSA) as an extension of Endsley’s Situation Awareness (SA) model. I claim that not only the driver needs to know what is happening in his environment, but also the system, e.g., the car. In order to achieve SSA, two paths of research have to be followed:
- Assessment of cognitive load of the driver in an unobtrusive way. I propose to estimate this value using a model based on environmental data.
- Developing model of cognitive complexity induced by messages presented by the system.
Three experiments support the claims I make in my conceptual contribution to this field. A prototypical implementation of the situation-aware presentation management toolkit PRESTK is presented and shown in two demonstrators.
Michael Feld (2011): „A Speaker Classification Framework for Non-intrusive User Modeling: Speech-based Personalization of In-Car Services“
1st Reviewer: Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster
2nd Reviewer: Prof. Dr. Bernd Möbius
Speaker Classification, i.e. the automatic detection of certain characteristics of a person based on his or her voice, has a variety of applications in modern computer technology and artificial intelligence: As a non-intrusive source for user modeling, it can be employed for personalization of human-machine interfaces in numerous domains. This dissertation presents a principled approach to the design of a novel Speaker Classification system for automatic age and gender recognition which meets these demands. Based on literature studies, methods and concepts dealing with the underlying pattern recognition task are developed. The final system consists of an incremental GMM-SVM supervector architecture with several optimizations. An extensive data-driven experiment series explores the parameter space and serves as evaluation of the component. Further experiments investigate the language-independence of the approach. As an essential part of this thesis, a framework is developed that implements all tasks associated with the design and evaluation of Speaker Classification in an integrated development environment that is able to generate efficient runtime modules for multiple platforms. Applications from the automotive field and other domains demonstrate the practical benefit of the technology for personalization, e.g. by increasing local danger warning lead time for elderly drivers
Christian Müller (2005): „Zweistufige kontextsensitive Sprecherklassifikation am Beispiel von Alter und Geschlecht“
This dissertation describes a two-layered speaker classification approach on the example of age and gender. First of all, the results of comprehensive corpus analyses are presented that are suitable to serve as a reference basis for further studies in human sciences. It is showed, that the models which are trained using these data are able to recognize the above mentioned characteristics with an accuracy that is up to five times better than the respective chance level. In addition, the presented approach distinguishes itself by the so called Second Layer , on which a context sensitive fusion of multiple classification results is accomplished using Dynamic Bayesian Networks. The dissertation also describes a concrete speaker classication system which was developed for the application scenario of mobile spoken dialog systems.