NEURAL NETWORK BASED ROBUST CONTROL STABILIZATION OF SPHERICAL INVERTED PENDULUM WITH DUAL-AXIS REACTION WHEELS

Abstract

As the new paradigm shift happened in the industrialization, introducing "Industry 4.0", many new applications in robotics are emerged. One of them safe human robot interaction field. Such an approach can be executed by utilizing variable impedance actuators, which imitates human and animal bodies, guaranteeing safety during the interaction. But on the other hand, these actuators have some limitations due to the damping elements in their construction. These elements behave as filters and may decrease the accuracy and positioning of the end effector, and may cause vibrations. Therefore, to avoid such unfavorable conditions, some stabilization elements should be integrated with the actuators. The thesis work will introduce a two-axis reaction wheel based inverted pendulum, which can deal with the nuisance by liquidating vibrations and adjust accuracy. In this work the utilization of Neural Networks is implemented to stabilize the pendulum. Moreover, it will consider the robust control Neural Network for the case of uncertainties and external disturbances. By generating the simulations and conducting real world experiments, the work will demonstrate the advantages of Neural Network employment.