ADynamicQuasiNewtonMethod forUncalibrated VisualServoing.pdfVIP

A Dynamic Quasi-Newton Method for Uncalibrated Visual Servoing Jenelle Armstrong Piepmeier Gary V. McMurray Harvey Lipkin Mechanical Engineering Georgia Tech Research Institute Mechanical Engineering Georgia Institute of Technology Atlanta, GA 30332-0823 Georgia Institute of Technology Atlanta, GA 30332-0405 Atlanta, GA 30332-0405 Abstract In [12], the authors demonstrated the limited ca- pability of a quasi-Newton method for tracking mov- Tracking of a moving target by uncalibrated model ing targets. Target state prediction schemes improved independent visual servo control is achieved by de- tracking accuracy, but did not improve precision. It is veloping a new \dynamic quasi-Newton approach. shown here that Newtons method as applied in [10] Model independent visual servo control is de¯ned as and [12] is designed to converge on a static target, using visual f eedback to control a robot without pre- and is not guaranteed to converge on a moving target. cisely calibrated kinematic and camera models. The This paper addresses a moving target scenario giving control problem is formulated as a nonlinear least an objective function in both joint and time variables. squares optimization. For the moving target case, this The result is a dynamic quasi-Newton control law im- results in a time-varying obj ective function which is plementing a dynamic Broyden update that is applica- minimized using a new dynamic Newtons method. A ble to the moving target problem without target state second-order convergence rate is established, and it is prediction. shown that the sta