Grasping Stereo and Texture Cue Combinations

Abstract

Studies of grasp indicate that the reach-to-grasp movement is essentially preprogrammed and relies heavily on estimates of object geometry provided by vision. Moreover, computational analyses of two-point grasps suggest that the identification of pairs of grasp points requires extensive computation of surface geometry. Thus, one can view subjects' decisions of how to obtain a stable two-point grasp as a good indicator of the perceptuo-motor system's appreciation of surface structure. We probe this perceptuo-motor surface representation by examining grasps to ellipsoidal surfaces defined by the cues of texture and horizontal disparity under independent levels of noise added to both texture and stereo. Subjects viewed the sequential presentation of left and right images on a computer display through liquid-crystal shutter glasses and were instructed to perform either an overhand or underhand grasp to the presented virtual ellipsoid. The virtual ellipsoid was oriented with its long axis in the horizontal plane. Rotation in depth about a vertical axis was varied between trials by a staircase procedure, which obtained the orientation at which overhand and underhand grasps were equally likely as well as the differential threshold. Results showed that the differential thresholds were a few degrees for the zero noise condition and increased with the addition of noise. These results are discussed in terms of theories of cue combination and perception for action.