A conceptual approach to describing and evaluating problem-solving by robotic systems is offered. One particular problem of importance to the field of robotics, disassembly, is considered. A general description is provided of an effector system equipped with sensors that interacts with objects for purposes of disassembly and that learns as a result. The system's approach is "bottom up," in that it has no a priori knowledge about object categories. It does, however, have pre-existing methods and strategies for exploration and manipulation. The sensors assumed to be present are vision, proximity, tactile, position, force, and thermal. The system's capabilities are described with respect to two phases: object exploration and manipulation. Exploration takes the form of executing "exploratory procedures," algorithms for determining the substance, structure, and mechanical properties of objects. Manipulation involves "manipulatory operators," defined by the type of motion, nature of the end-effector configuration, and precise parameterization. The relation of the hypothesized system to existing implementations is described, and a means of evaluating it is also proposed.