Representing knowledge with Semantic Web technologies is becoming more and more common in robotics, leading to a proliferation of ontologies that represent several aspects of robotics. Thus, it is necessary to define strategies and approaches to evaluate the performance and suitability of such ontologies. Most existing strategies consist on formal evaluations, centered in the quality and correctness of the ontologies and their ability of modeling the domain, considering features at the lexical and structural level. However, these strategies neglect the empirical evaluation in terms of the ontologies performance in robotics experiments. In this sense, we propose a novel strategy, based on a combination of different approaches, to evaluate ontologies through experimentation. To demonstrate the suitability of the proposed strategy, we evaluate an ontology in the SLAM (Simultaneous Location and Mapping) domain and perform experiments on different maps with two social robots with different characteristics. Results validate that the use of the ontology allows obtaining benefits of the Semantic Web, like the capability of inference from organized knowledge representation, without compromising the information for the application in robotics and demonstrate the suitability and effectivity of our proposal.