The Intelligent Hand: An Experimental Approach to Human Object Recognition and Implications for Robotics and AI

摘要

The scientific study of biological systems offers an approach to the development of sensorbased robots that is complementary to the more formal analytic methods currently favoured by roboticists. I initially propose several general lessons from the biological field. Next, I consider a specific example selected from the work of Lederman & Klatzky, which focuses on human haptic object processing. An empirical base and recent theoretical developments from our research program on this topic are described. The human haptic system is an information-processing system that combines inputs from sensors in skin, muscles, tendons, and joints with motor capabilities to extract different object properties. A general model of human haptic object identification, which emphasises how object exploration is controlled, is presented. The model describes major architectural elements, including representations of haptically accessible object properties and exploratory procedures (EPs), which are dedicated movement patterns specialized to extract particular properties. These architectural units are related in processing-specific ways. The resulting architecture is treated as a system of constraints, which guide the exploration of an object during the course of identification. Empirical support for the model is also examined. To conclude, I show how this 774 Invited Speakers scientifically-based approach might be applied to developing strategies for active manual robotic exploration of unstructured environments. 1