Empirical Analysis of Visual Servoing Complexity Across Different DOF Robots via the Global LLS Method

Olivia Luping Cai, Martin Jagersand

Keywords: visual servoing, robot, robot arm, LLS, least local squares, DOF, degrees of freedom, kinova, visual servo, visual servoing complexity, newton method, global model

TL;DR: We compare the visual servoing complexity between different DOF robots: higher DOF robots excel in visual servoing despite poor Jacobian approximation, highlighting joint redundancy and the feasibility of small global models.

Abstract: Despite the long-standing development of uncalibrated visual servoing (VS), there is little research exploring how the complexity of the VS task varies across robots with different degrees of freedom (DOF). To analyse this, we generate a globally valid visual-motor model of the VS function using the Local Least Squares (LLS) method, where the Jacobian is estimated by fitting a hyperplane to the k-nearest datapoints from previous robot trajectories. We examine the number of data points, N, required to construct a global model for robots with 2, 3, 6 and 7 DOF to compare complexity of the VS task across different systems. Results show that the higher DOF robots perform very well for visual servoing despite poor Jacobian estimates calculated from the LLS model; highlighting the benefits of joint redundancy and the reduced necessity of an accurate Jacobian for nearby tasks and higher DOF robots, and showing the potential for further development of data-efficient global models of the visual servoing function.

Submission Number: 25