At this event, I will present and defend my PhD thesis titled "An Expert Knowledge Representation Based on Real World Experiences in Proof-of-Concept Robot System Design".
A summary of the topic is given below.
The development of proof-of-concept robotic systems requires a high level of domainoriented and expert knowledge in an interdisciplinary field to provide individual solutions. The direct interdisciplinary use of existing solutions for such systems can be limited due to domain, environment and task differences. Therefore, it may be necessary to build an appropriate solution from scratch. However, this is time consuming and potentially expensive. In addition, the solution parts require knowledge of the system and precise prior knowledge of the robot’s task. This means that the structure and functions of the robot and its subsystems are not only important for the development of the robot, it also requires knowledge that can be used in an interdisciplinary and interactive way in the execution of the robot’s own task, like self-maintenance or reassembly, or in the disposal of the robot. In practice, applications that can be used as automated development tools (such as Q-Rock) support the user and help to take an important step towards customer centric design and manufacturing. It can move closer to the reality of outcome-based product definition, while reducing development costs and increasing product variety. The key challenge of this thesis was to identify existing robot design approach and provide a widely applicable knowledge base that allowed to simplify one streamline robot design. This covers prior knowledge about the subsystem, facts and relationships, as well as information about hardware design, components and dependencies.
Knowledge representation based on ontology for robotic system design can be used as a unified and standardized method for knowledge generation, using and sharing. The problem is that there is an inverse relationship between the quantity of information and its quality, relevance and usability, named as information paradox. Concepts defined in the ontology and the extraction of valuable additional information from explicit knowledge using reasoning would counteract this phenomenon. Considering the life cycles of robotic systems, the ontology would be a suitable way to collect and represent the required knowledge about all the phases that this system goes through.
My thesis focuses on the idea of solving the challenge of developing proof-of-concept system solutions for terrestrial and extraterrestrial robotics as an ontology-based knowledge representation of prior knowledge from different domains. This empirical approach is based on the experience gained in various projects. To organise the interdisciplinary knowledge which is required for designing a robot, this thesis focuses on solution pproaching the problem from two perspectives. The first is the formalism approach. Formalism is the definition of the concepts in a predetermined way, i. e. as a standardised body of knowledge. In this context, the use of manually created ontologies to be used as Knowledge Base in the field of robotic system design is envisaged and methodically tested for examples in different robot development domains. Knowledge in this form is appropriate for a sustainable development and evaluation. The second perspective is modularity for model-based design, where the hardware modules, into which methods such as Rapid Control Prototyping can be incorporated, are defined by the ontology. This prior knowledge is used in the system framework for complex robot control using auxiliary inference. These points have been tested for implementation in several publications that form the basis of this work. The results have been evaluated in laboratory conditions for specific use cases for different applications, such as robot development, robot control and in-orbit factory tasks. The study shows that it is possible to model robot components or tasks with different characteristics using an ontology and incorporate them into the robot control application. As conditions change, the required capabilities can be adaptively generated manually or automatically and information about them can be retrieved using reasoning. For this purpose, eight different ontologies and a python-based library supporting their use were developed. In addition, more than 60 robotic systems and subsystems well known in the robotics community were collected as hardware models, edited and further modeled for simulation purposes. Based on the knowledge obtained from the ontology, the properties of these systems were transferred to the models. These upgraded models have been made available to the community. Moreover, an ontology has been used for the internal modelling of the standard interface, which can be used as a support for the Standard Interconnect (SI) definition standard.
