Visual Error Constraint Free Visual Servoing Using Novel Switched Part Jacobian Control

This research advances the state of the image-based visual servoing (IBVS) of robotic arms to handle very large visual errors without the camera advance/retreat problem. Conventional visual servoing schemes either consist of a partitioned or a switched system that relies on the feature Jacobian to find a unique feature for partitioning control along the specific DoF. We suggest a new IBVS scheme based on part-manipulator Jacobian approach for building a hybrid switched-partitioned task jacobian without the need to define new features. Utilizing this computationally efficient, directly defined Part-manipulator Jacobian an efficient second order minimization(ESM) based adaptive switching controller was constructed. The proposed scheme was tested in the eye-in-hand configuration on a 6-DoF simulated robotic arm and a 7-DoF real robotic arm for a set of large visual errors between the initial and the desired frames, including a rotational error of 180 degrees around the camera optical axis. Compared to other IBVS schemes under various simulation conditions, the performance of the proposed scheme remained superior to that of the Jacobian-pseudo-inverse and other ESM-based IBVS schemes. The experimental results showed a notable expansion of the convergence zone up to 180 degrees rotational errors with a 40% improvement in the convergence rates with a significant 90% reduction in the joint velocities and joint energies required to complete the task. The proposed controller possesses no camera advance/retreat motion, has a task Jacobian matrix that is well-conditioned, and it is computationally efficient. Moreover, the method is independent of the robot's DoF and is extendable to other visual servoing schemes.