Frequency Sensitive Motion Control for a Single Joint Arm Model

Abstract

The proposed feedforward controller is composed of three parts: a lookup table controller, an artificial neural network controller and a decision block. The lookup table controller has the information about the relation between the activation levels of muscles and the output at steady state. To compensate for the delayed tracking ability of the lookup table controller for rapid movement, an artificial neural network (ANN) controller is used in parallel with the lookup table control. The ANN controller is trained to learn the inverse dynamics of the musculoskeletal model. The decision block determines the contribution ratio of each controller based on the frequency analysis of the reference trajectory. The control algorithm was tested on single joint elbow model with a flexor and an extensor. Results show that the combined controller reduces the overall output errors.