Application of Change Detection to Dynamic Contact Sensing

Abstract

The forces of contact during manipulation convey substan tial information about the state of the manipulation. Textures, slip, impacts, grasping, and other contact conditions produce force and position signatures that can be used for identifying the state of contact. This article addresses the fundamental problems of interpreting the force signals without any addi tional context on the state of manipulation. Techniques based on forms of the generalized sequential likelihood ratio test are used to segment individual strain signals into statistically equivalent pieces. The results of the segmentation are designed to be used in a higher level procedure that will interpret the results within a manipulation context. We report on our ex perimental development of the segmentation algorithm and on its results for detecting and labeling impacts, slip, changes in texture, and condition. The sequential likelihood ratio test is reviewed, and some of its special cases and optimal properties are discussed. Finally, we conclude by discussing extensions to the techniques and lessons for sensor design.