Path planning in image space for uncalibrated image-based visual servoing

Abstract

This dissertation lays a path planning method in the image space for image-based visual servoing. Image-based visual servoing is a control scheme where two-dimensional image features are directly used as control inputs to a robot system. Because of the direct visual feedback, three-dimensional pose estimation is not necessary and the positioning accuracy is insensitive to calibration parameters compared to position-based visual servoing. Although image-based visual servoing exhibits superior performance, however, it shows fatal limitations when initial pose discrepancy is large. The feature points may leave the camera`s field of view and, also, the robot may not converge to the goal configuration. One possible approach to resolve these limitations is to construct a number of intermediate subgoal images of the gripper such that they can define a straight path so as to keep the feature points always within the field of view during the task. Under the assumption that only the initial and goal gripper images are given without any metric information, a novel method within a projective framework to generate such a straight path is proposed. The proposed method synthesizes physically valid intermediate subgoal images based on screw-motion constructions, which are represented in the projective space with the help of the fundamental properties of projective geometry. An important benefit in using a projective framework is that we always obtain the same images no matter how the projective frame is selected. To exclude erroneous estimations due to the image noise in the projective displacement, computational issues are also considered by adding cost functions for a nonlinear optimization. The cost functions are derived from the fact that the undergoing projective displacement is conjugated to an unknown Euclidean displacement. The main idea is extended to take the mechanical constraints into account to make the resultant image trajectories are acceptable to real manipula...