A learning model for the selection of problem solving strategies in continuous physical systems

摘要

The authors present a learning model for automatic selection of problem-solving strategies in a physical system working in a continuous, dynamic environment. The model is intended to provide (1) a means to classify under what circumstances one problem-solving strategy (goal-directed or data-driven) is better than another; and (2) a simple online learning mechanism for obtaining a better classification when the strategy chosen based on the initial one does not yield the desired performance. The ideas and applicability of the model are illustrated by an example system for robotic assembly which has been under implementation.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>