As part of our research, the Virtual Theater project has implemented
several systems for exploring how our agents interact, and how users can
guide them to craft stories or simply to experience improvisational,
Here we outline the projects we have worked on and are working on
Animated puppets is a graphical world inhabited by creatures called
woggles who move about and talk and play with one another. In
this world, the user can control one woggle, while the other woggle is
directed by the computer. The woggles use intelligent improvisational
ability to choose interesting behaviors.
There is more detailed information
available about the woggles and animated puppets. Ruth Huard is using
the woggles system for her own research in childrens' storycrafting;
there is a set of slides outlining her
For more information on the animated puppets, see also the paper Directed
Improvisation by Computer Characters.
The CyberCafé project uses a text-based, multi-agent environment
in which users can interact via avatars with autonomous characters. In
the basic scenario, two users take the forms of customers in the
CyberCafé, while the computer directs the actions of their waiter.
Daniel Rousseau used the CyberCafé to study models of
personality in computer-controlled characters. In his studies, several
different waiter personalities were used in the scenario, to compare how
their personality models generate substantively similar actions, but
actions infused with distinct mannerisms and behaviors. There is a set of slides
overviewing the Cybercafé and Daniel Rousseau's research in it.
The summary papers on this project are A
Social-Psychological Model for Synthetic Actors, and Interacting
with Personality-Rich Characters, and
Improvisational Synthetic Actors with Flexible Personalities.
The Forest Sauvage
In the Forest Sauvage project, we are creating believable characters
with broad, abstract domain knowledges, and placing them in virtual
environments that provide specific domain details in the form of embedded
annotations. Our goal is to study how, in this way, believable agents may
visit a variety of domains and act naturally and entertainingly within
Our current testbeds are text-based MUDs and educational Web pages, in
which we have placed a guide character called Merlyn, named after the
befuddled old magician in T. H. White's The Once and Future King.
Merlyn exists and acts independently of the user. He can scan his
environments for actions and information about them, communicate with the
user, suggest activities, play with the user, and develop a model of the
user across multiple sessions by observing the user's actions and
An introductory overview of this work is Guided
Exploration in Virtual Worlds. Current research is reflected in Agents in
Annotated Worlds. There is also a set of
slides overviewing the project.
Tigrito is intended to contrast the effectiveness of different modes of
interaction with a synthetic character. The character in question is an
animated stuffed tiger. Users may interact with the tiger through another
tiger avatar or as disembodied entities; they may also observe the
computer-controlled actions of two tigers, Tigrito and Hobbes. This work
is now being extended by Heidy Maldonado as a potential mechanism for
second-language learning of English or Spanish.
A summary of the initial phase of this work is contained in Tigrito:
A Multi-Mode Interactive Improvisational Agent.
In the master/servant scenarios, we studied how two autonomous agents
interact with one another without human intervention. The master and the
servant each have knowledge about the environment and their status within
it. These scenarios test their behavior under computer-controlled stimuli
and emotional variations.
We currently have a series of screen
shots available for viewing. They are part of a video that shows the
master, Otto, reversing roles with his servant, Gregor.
Many thanks go to
Ken Perlin at NYU's Media Research Lab for
the use of their animation system in creating these stories.