A robust line search for learning control

In this paper a new line search for a Newton-Raphson learning control algorithm is presented. Theorems and rigorous proofs of its increased robustness over existing line searches are provided, and numerical examples are used to further validate the theorems. Also, the previously posed open question of whether robust optimal trajectory learning is possible is also addressed. It is shown that the answer is generally no, at least for gradient-based learning control algorithms.

[1]  Andrew A. Goldenberg,et al.  Learning approximation of feedforward dependence on the task parameters: Experiments in direct-drive manipulator tracking , 1995, Proceedings of 1995 American Control Conference - ACC'95.

[2]  Suguru Arimoto,et al.  Bettering operation of Robots by learning , 1984, J. Field Robotics.

[3]  Dimitry Gorinevsky.  An application of on-line parametric optimization to task-level learning control , 1995, Proceedings of 1995 American Control Conference - ACC'95.

[4]  Roberto Horowitz,et al.  Learning Control of Robot Manipulators , 1993 .

[5]  N. Sadegh,et al.  Minimum time trajectory learning , 1995, Proceedings of 1995 American Control Conference - ACC'95.

[6]  Dimitry Gorinevsky,et al.  An algorithm for on-line parametric nonlinear least square optimization , 1994, Proceedings of 1994 33rd IEEE Conference on Decision and Control.

[7]  Danwei Wang,et al.  Learning control for a class of nonlinear differential-algebraic systems with application to constrained robots , 1996 .