Towards a Standard for Grounding Symbols for Robotic Task Ontologies

We discuss the “symbol grounding problem” in the development of robotic tasks and how different types of grounding can affect the generality of automatic planning and recognition algorithms. Grounding is the mechanism through which task symbols get associated with concrete meanings and effects on the world. We discuss the implications of different strategies when grounding symbols in robotic task models. In practice, the extent to which a symbol is grounded lies on a continuum between “implict” and “explicit” grounding. Implicitly grounded symbols are those which have requirements and effects that are not represented in the symbol system. This can occur when the effects of a symbol are only determined by the result of executing an opaque action or because the task model is constructed from a set of simple rules that lack the expressiveness to wholly capture the task requirements. Alternatively, symbols can be explicitly grounded by describing their requirements and effects on the world in terms of real physical quantities. Such grounding enables a system to better reason about physical entities in the world, their properties, and their behavior. This approach comes with its own limitations, due to increased complexity and difficulty creating accurate models. We discuss a handful of different grounding approaches in terms of a case study with a simulated multi-arm manipulation task. These considerations will help develop standards of grounding for task ontologies and describe how robust they make a given task model.