The Marionette is a simple performance animation system that
allows the user to explore a variety of ways to map the output of
motion capture devices (such as the Polhemus FastTrak or
InsideTrak) to the pose of a graphical character.
My goal with the Marionette is to explore these issues:
- How should the mapping from the sensor data to the character be
specified? This gets interesting when the mapping is
nonliteral, as in for example, when you want to
animate a Luxo lamp, or in the case of the Marionette, the
sensors have fewer degrees of freedom than the character.
- How can very simple pattern recognition techniques be used to
overcome the problem (opportunity?) of having very few input
degrees of freedom?
- What kind of funky, nonliteral maps can you design with such a
system? Is it possible to get interesting cartoony effects, for
example, with a small set of maps that the user trains up