Tensor-based Format for Exchanging Hypergraphs between Cognitive Entities

Robotic and other intelligent systems have recently gained prominence, integrating a growing number of new functionalities. Since the beginning, the flexible semantic self-recognition of intelligent devices has been a challenge. The traditional solution is to provide a centralized source of semantic descriptions (in parameter servers or external sources). This paper introduces a graph-like binary format for machine-to-machine information exchange to efficiently index and process graph-based data at each endpoint. The exchanged graphs, which are represented as multidimensional tensors, are based on the formal theory of hypergraphs, allowing for the description of associations between not just two, but multiple modalities. We show in addition that hierarchical information is also compatible with this hypergraph-based formalism. Thus, structured information can be shared between intelligent entities, using the proposed format, to share information for reasoning, planning (e.g., motion planning in robotics), as well as environment representation. A further motivation includes the description of vision data by providing semantic information that can be rewritten in interpretable form for the target visualization engine.