In the growing interest about robot ontologies, the role of description logics (DL) used to query the ontology is often neglected. After discussing the representation problems of using ontologies for robotics problems, we develop two case studies relevant for dexterous manipulation, grasping unknown objects and understanding objects as the sum of their parts. We develop knowledge bases for solving those problems, and examine the possible queries that could be answered. We observe that the creation of such relationships in ontologies can follow an embodied approach, letting the robot to create new relationships from experiments, so making knowledge acquisition an embodied and iterative process. From those two cases we conclude that the creation of robot ontologies able to improve reasoning in manipulation is still a challenge itself. However ontologies are expected to be a necessary tool to share knowledge between humans and robots, and their role in integrating available symbolic knowledge and experimental data is still to be exploited.
展开▼