Neural-based control and stability analysis of a class of nonlinear systems: Base-excited inverted pendulums

This paper presents a novel application of multilayer neural networks for online control of a class of base-excited inverted pendulums. The pendulum has two degrees of rotational freedom and its base-point moves freely in three-dimensional space. The goal is to apply control torques to keep the pendulum in a desired orientation, in spite of disturbing base-point movement. Four three-layered neural networks are trained online to represent the inverse dynamics of the plant within a controller. The conditions of training accuracy, to guarantee the stability of such a non-autonomous closed-loop system, are established using Lyapunov stability theory. The proposed neural controller is examined through simulations. Its performance is also compared with the performance of a Lyapunov controller from the most recent published work. It is shown that the proposed control scheme is simple in implementation in the sense that it does not require a mathematical model of the target pendulum or the measurement of the base-point movement. At the same time, it produces fast, yet well-damped responses with smooth control torques. The work presented here can benefit practical problems such as the study of stable locomotion of the human upper-body and bipedal robots.