Mechanisms of Shared Attention for a Humanoid Robot

摘要

This paper outlines a proposal for constructing mechanisms of shared attention for a humanoid robot through a series of example tasks. Shared attention, the ability to selectively attend to objects that are mutually interesting, is vital for learning from another individual. The platform that we will use in designing this system is the upper-torso humanoid robot called Cog which is currently under construction at the MIT Artificial Intelligence Lab. We present a description of our approach, an outline of the work in progress, and a summary of the work completed so far. Introduction While the past few decades have seen increasingly complex machine learning systems, the systems we have constructed have failed to approach the flexibility, robustness, and versatility that humans display. There have been successful systems for extracting environmental invariants and exploring static environments, but there have been few attempts at building systems that learn by interacting with people usi...