| Abstract: | We present a system that enables a robot to learn to plan through
demonstration and imitation. An imitator acquires planning operators
by observing a demonstrator, segmenting the demonstrators actions into
planning steps, and learning the preconditions and effects of the
operators. When the imitator tries to execute its own plans, it
learns to perform the operations through reinforcement learning, and
corrects errors in the previously learned operator effects. Note:
This document was originally written in August of 2003, but submitted
as a tech report on the above date. |