Real-time control and learning using neuro-controller via simultaneous perturbation for flexible arm system

Abstract

This paper describes details of real-time control and real-time learning of neuro-controller for a flexible arm system using the simultaneous perturbation optimization method The simultaneous perturbation optimization method is useful, especially when dimension of the parameters to he adjusted is large. Therefore, it is beneficial to utilize the simultaneous perturbation method for neural networks. On the other hand, when we use the ordinary gradient method as a learning rule of the neuro-controller, Jacobian of the plant is essential. However, the learning rule via the simultaneous perturbation does not require Jacobian of an objective plant so that the neural network uses only outputs of an objective system. Actual real-time control and real-time learning results of a real flexible arm system are described to confirm a feasibility of the proposed method.