Environmental Representation for Human Robot Interaction

Abstract

This study proposes an interaction in which the goal state of human verbal instructions is realized through mutual understanding between humans and robots. To achieve this, a system is proposed that realizes the interpretation of the goal state according to human verbal instructions within the scope of robot action possibilities. In this approach, the feasible environment recognized by verbal instructions and action graphs is evaluated using scene graphs and word embedding techniques. Simulation and subjective evaluation confirm the effectiveness of the proposed interaction and system. As a result, the proposed interaction considering robot action possibilities improves human satisfaction.