Geometric Specification of Dynamics: Learning to Visually Perceive Kinetic Quantities from Static Images

Abstract

In a single experiment, perceivers viewed computer-generated hammer-like objects and estimated a kinetic property. Each hammer comprised a cylinder of some diameter and height, attached to a handle-a second cylinder of fixed diameter and variable length. On 5 blocks of trials, three groups of perceivers reported a property reflecting mass, torque, or moment of inertia. Blocks 2-4 were training blocks on which numerical feedback was given. On blocks 6 and 7, perceivers were asked to report the kinetic properties not reported on blocks 1-5. We were interested in how geometrical object properties affected the perception of kinetic properties, and how that perception changed with experience. To determine how geometrical object properties informed reports of kinetics, a continuous information space of head diameter and height, and handle length was created, and perceivers' loci within that space were determined. Perceivers began at various distances from the optimum for the to-be-reported property and were closer to the optimum after training. On blocks 6 and 7, performance did not reflect transfer from training on another property. The results were interpreted in the context of the theory of direct learning.