Technology Review

Shared virtual environments are often fantasy spaces where people interact using avatars endowed with supernatural powers. In contrast, simulations of large, real places are usually dedicated to traffic or disasters and involve independent software agents rather than avatars controlled by humans.

A researcher at Kyoto University in Japan has combined the two to make a virtual environment that allows large numbers of avatars and software agents to interact in a space constructed from photographs mapped onto surfaces of a three-dimensional model of the photographed space.

FreeWalk taps social psychological methods to make the virtual environment believable enough that people behave realistically. Agents and avatars appear as humans, and human users and agents perceive and control social cues like gestures to inform their interactions with each other.

Speech recognition software converts a user’s words to text, speech synthesis software produces vocalizations for agents and avatars, and keyword matching provides agents with a rudimentary form of language understanding. The speech recognition and speech synthesis engines run on the user’s computer to minimize the amount of data transmitted between users.

The researcher modeled a shopping district in Kyoto, complete with street and subway signs and store window displays. Models that mimic disaster situations that include real social interactions could be used for drills that eliminate the cost and disruption of physically moving large numbers of people.

The next step is to add details like fire and smoke. The work appeared in the April, 2004 issue of the International Journal of Human-Computer Studies.