Learning in ICARUS
Learning is the acquisition of knowledge through experience, in which ICARUS'
concept hierarchy is altered to reflect perceived object and places, generated
plans, and executed action sequences. The altering can consist of modifying
existing nodes to generalize or refine a concept, or by creating new nodes to
remember a novel experience. This aspect of learning is done mainly in
LABYRINTH. In this learning is also the updating of probabilities of the
success of the problem solving method used for future analogous problems,
the possibility of abandonment of a plan, and probability that a specific
state will be reached if the plan is executed. This is closed-loop processing,
but in the case of low probabilities of violated expectations, open-loop
processing is used.
Return HOME. Goto NEXT.