Intuitive linguistic joint object reference in human-robot interaction: human spatial reference systems and function-based categorisation for symbol grounding

摘要

The visionary goal of an easy to use service robot implies intuitive styles of interaction between humans and robots. Such natural interaction can only be achieved if means are found to bridge gap between forms of object perception and spatial knowledge maintained by such robots, and forms of language, used by humans, to communicate such knowledge. Part of bridging this gap consists of allowing user and robot to establish joint reference on objects in environment - without forcing user to use unnatural means for object reference. We present an approach to establishing joint object reference which makes use of natural object classification and a computational model of basic intrinsic and relative reference systems. Our object recognition approach assigns natural categories (e.g. desk, chair, table) to new objects based on their functional design. With basic objects within environment classified, we can then make use of a computational reference model, to process natural projective relations (e.g. the briefcase to left of chair), allowing users to refer to objects which cannot be classified reliably by recognition system alone.