Flexible Self-Calibrated Visual Servoing for a Humanoid Robot

摘要

This paper develops a flexible position-based visual servo framework to enable a humanoid robot to perform a variety of visually controlled manipulation tasks. The system overcomes classical drawbacks of position-based visual servoing, including hand-eye calibration and handling large pose errors, while catering for complex motion planning techniques. Active vision is used to track the gripper during servoing, which reduces the need for accurate camera calibration and allows the system to handle very large pose errors without losing vital visual information. Robust, continuous hand-eye calibration is achieved using a Kalman filter to estimate the pose of the gripper. Experimental work demonstrates the robustness and flexibility of the system in performing a complex task requiring grasping and assembling objects.