2007 年 12 巻 4 号 p. 509-517
It is very costly to conduct experiments to test smart environments installed in large-scale public spaces (e.g. airports and railway stations). As a tool to examine the design of such experiments beforehand, we developed a multiagent simulator integrating a large number of software agents and humans into a crowd that can produce complex group behaviors. In the testing of our emergency guidance system installed in a central railway station, human-like agents generated virtual emergency evacuations, which the subjects experienced via virtual reality or augmented reality. Consequently, we could hypothesize that our system is superior in trustworthiness, usefulness, and so on.