Real-time control of an inverted pendulum system using complementary neural network and optimal techniques

Inverted pendulum is a simple, inherently unstable system which exhibits the fundamental characteristics of balance. This paper explores the control of a robotic pole-balancing system using complementary optimal and neural network techniques. A full state feedback linear regulator type optimal controller was developed from the approximate system model. When applied to the real physical system, this controller produced a relatively large limit cycle, due primarily to unmodelled system nonlinearities. The CMAC neural network was then introduced into the controller to learn any nonlinearities, reject residual noise, and, as a result, shrink the system limit cycle.