This paper proposes a novel architecture and interactive learning
method for an artificial creature implemented in the 3D virtual
world. The artificial creature can decide its behavior by itself
based on its own motivation, homeostasis, emotion, and external
sensor information.
A behavior is selected by both probabilistic method and
deterministic method. The probabilistic method uses mathematically
modeled internal states and external sensor information, and the
deterministic method which imitates animal`s instinct, uses only
external sensor information. Both two methods are complementary to
each other.
A user can teach the creature to do a desired behavior by an
interactive training method. To select a desired behavior among
many behaviors, behaviors are grouped into analogous behavior
sets. A behavior learning is carried out with a rewarding and a
penalty signal. The learning algorithm includes the emotional
parameter by which the training efficiency is affected.
The proposed creature frame uses a data driven architecture, which
is separated into two parts, algorithm part and data part. A user
can modify the creature`s personality easily without correcting
main algorithm part.
The performance of the Artificial Creature, `RITY` with the
proposed a creature frame is demonstrated in the 3D virtual world.