As a consequence of our commitment to interactionist models of development, we
plan to build integrated robot systems to test our hypotheses. Our objective
is not to invest too heavily in the details of the techniques presented in
previous sections, but instead to prove that interaction, through sensorimotor
primitives and policies, leads to conceptual structure.
In this section, we will present a tentative experimental agenda for
demonstrating our objectives. We will embed a set of robot resources (see
Section 
) in a common work area. For example, a stereo
pair of cameras and our integrated hand/arm system will interact with objects
on a table top. In all such experimental designs, our goal is to document a
developmental sequence from physical primitives, through physical and
figurative schema, and toward a conceptual structure with predictive semantics.
Consider the following sequence for acquiring concepts in the blocks world domain. Our robot configuration consists of a dextrous hand, a robot arm, an articulated stereo vision system, and a binaural auditory system. The robot will interact with the blocks world environment in order to build domain rules for this class of tasks. This expertise will necessarily proceed from coordinated sensorimotor policies (spatial sense), to models describing the relationship between the robot and environmental features (categorization), and finally to declarative models of the interactions between extrinsic features (stability).