PUB - Publikationen an der Universität Bielefeld

Understanding the prerequisites of an effective and intuitive human-robot teaching situation requires an experimental approach to human-robot interaction that includes cross-platform variation as well as fast inter- and intra-component change. Although re-use has been significantly increased over the last years, the complexity faced in robotic systems is still leading to frequent lock-in rather than change. Today, most approaches address the complexity in a robot’s software system by decomposition into components, coupled through middleware. However, the integration of core algorithms, utility libraries and middleware into a component is itself becoming more complex and, due to the embedded external references, can create lock-in. To reduce both lock-in and complexity, we have developed a component construction framework based on modelling the various building blocks as directed, typed graphs. These graphs may be (re-)assembled efficiently into components. The approach has been tested on diverse applications, from device drivers to robot behavior generation and we report the experience gained. Furthermore, we propose a decomposition strategy for converting legacy components into such graphs on a coarse level. The strategy is based on principles from event-based systems to optimize the re-usability of the resulting sub-graphs.