Case-Based Reasoning for Action Planning by Representing Situations at the Abstract Layers

A robot which can act according to natural language instructions must complement information from experiences and common knowledge to adapt to various situations. In this paper, we propose a method to evaluate the similarities of concepts by representing situations in abstract layers which are hierarchically arranged. We also present a method for case-based reasoning for planning which uses the layers as viewpoints for case retrieval. The technique is applied to the action planning for the house-work robot.