关键词: L-M方法；无标定视觉伺服；雅可比矩阵估计, 机器人

Abstract:

Gauss-Newton method is most commonly used in uncalibrated visual servo control, but it may not be convergent or unable to converge to the expected value in large residual cases. So, a new uncalibrated visual servo control method using Levenberg-Marquardt method was proposed. The trust region strategy was used to insure that the model was a good simulation of the nonlinear function and the step was adjusted adaptively by the ratio between actual decrease of the function and the predicted decrease in this method. Besides, the Broyden’s method was used to estimate the image Jacobian online. The disadvantages of traditional method are avoided and it does not need to calibrate the camera and the robot by using this method. Simulation and experimental results demonstrate the success of the method.

Key words: L-M method, uncalibrated visual servoing, Jacobian estimation, robot