Controle neuro adaptativo modular de manipuladores robóticos.

Abstract

In this work, robot manipulator control techniques based on artificial neural networks are developed. A theoretical review of Neural Networks and neurocontrollers is presented. A Block Diagram representation of the robot dynamics equation is developed. Based on this representation, two recursive algorithms for efficient computation of the inertia matrix and dynamics equation are proposed: the Inverse Path Method and the Central Path Method. A factored form of the inertia matrix is derived. A Linear-ln-Parameter Robot Model is developed and a recursive method for the computation of regressor matrix is proposed. The concept of modular adaptive control of robotic manipulators is presented. In the modular scheme, control and adaptation modules are assigned to each link, allowing the control and adaptation of each link dynamic parameters independent of the other links. A Linear-In-Kinematic- Terms robot model is developed. Based on this model, the concept of Modular Ncuroconlrollcr for robotic manipulators is presented. In the proposed scheme, Neural Modules (Standardized Neural Networks) are assigned to each link. The modules are joined in a global control system that can be applied to any serial manipulator. A direct inverse training method for the modular ncurocontroller is proposed and, based on adaptive control techniques, a direct adaptive training scheme is developed. In order to test the proposed control schemes, the ROBOTLAB software tool was developed. This tool allows the robot dynamics simulation, including 3D graphics output.