Signature search method for 3-D pose refinement with range data

In many applications in robotics, the geometry of the task environment is uncertain and so the pose of the target object may be known only approximately. For the object to be grasped successfully its actual pose must be determined using some form of vision sensing. This paper presents a novel method of processing 3-D range data for pose refinement. Given the model of the object and its approximate pose, the algorithm adjusts the pose of the object model for a best fit to the measured 3-D data. The main attributes of this algorithm are that its performance is largely unaffected by background clutter including some tolerance to occlusion and that it imposes no restrictions on the surface shape or representation scheme used for the model