A knowledge-based architecture supporting declarative action representation for manipulation of everyday objects

We present a new knowledge-based robot architecture that relies on ontologies and logical programming to represent robot behavior in a declarative fashion. Most state-of-the-art robot architectures, instead, build on very detailed procedural descriptions of robot behavior which are created at design time and typically do not generalize well across new situations. A declarative approach that produces actual robot behavior as a result of reasoning instead can produce many different behaviors that are adapted to the particular context and situation without having to model all influencing factors at design time. Further, this approach allows to add further knowledge or further primitives in a modular fashion. This simplifies the design of complex systems as the behaviour needs to be described at a higher-level of abstraction instead of by very detailed and specific procedures. The actual behaviour is then generated at runtime as a result of interpretation of logical programs. In our architecture we rely on Prolog to represent action knowledge, on (Horn) rules to represent actions and on Prolog resolution to infer execution plans at runtime. In this paper we present a first version of our architecture. As proof-of-concept we demonstrate how to apply it to generate pick-and-place behavior in a simulated environment.