University thesis:
Dissertation, Universität Bremen, 2022
Footnote:
Description:
It has been demonstrated many times that modern robotic platforms can generate competent bodily behavior comparable to the level of humans. However, the implementation of such behavior requires a lot of programming effort, and is often not feasible for the general case, i.e., regardless of the situational context in which the activity is performed. Furthermore, research and industry have an enormous need for intuitive robot programming. This is due to the high complexity of realizing an integrated robot control system, and adapting it to other robots, tasks and environments. The challenge is how a robot control program can be realized that can generate competent behavior depending on characteristics of the robot, the task it executes, and the environment where it operates. One way to approach this problem is to specialize the control program through the context-specific application of abstract knowledge. In this work, it will be investigated how abstract knowledge, required for flexible and competent robot task execution, can be represented using a formal ontology. To this end, a domain ontology of robot activity context will be proposed. Using this ontology, robots can infer how tasks can be accomplished through movements and interactions with the environment, and how they can improvise to a certain extent to take advantage of action possibilities that objects provide in their environment. Accordingly, it will be shown that parts of the context-specific information required for flexible task execution can be derived from broadly applicable knowledge represented in an ontology. Furthermore, it will be shown that the domain vocabulary yields additional benefits for the representation of knowledge gained through experimentation and simulation. Such knowledge can be leveraged for learning, or be used to inspect the robot's behavior. The latter of which will be demonstrated in this work by means of a case study.