2001 年 19 巻 5 号 p. 667-675
In human robot behavioral interaction, characteristic of the robot's reaction-decision algorithm produces impres-sions that the human user has about the robot. The purpose of this paper is to analyze the relationship between robot's interactive characteristic and produced impression about the robot in a human-robot-interaction experiment. The analysis was done quantitatively by using methods of information theory. The robot's interactive characteristic is quantified as information transfer efficiency of Shannon information theory; the algorithm that selects a fixed certain reaction in respect to certain input (1-to-1 reaction) has very high informational transfer efficiency. In con-trast, very low informational transfer efficiency of the interaction means the robot makes reaction almost randomly. In the experiment, 3 reaction algorithms of the robot were prepared; the informational transfer efficiencies of them are 1.0, 0.5, and 0.0. The result shows that high transfer efficiency produces impressions that the robot has strong intention. Also it is shown that maximum of degree of interest on interaction was got at moderate transfer efficiency (0.5) . Those results will be utilized in interaction algorithm of robots that adjusts robots' behavioral characteristic in order to produce proper impression toward humans.