Actively adaptive control for nonlinear stochastic systems

An adaptive control method can be classified into two categories, actively adaptive or passively adaptive, according to how the available information is being utilized in the on-line calculation of the control. An actively adaptive controller utilizes, in addition to the , available real-time information, the knowledge that future observations will be made, and regulates its adaptation (learning). This is done by anticipating how future estimation will be beneficial to the control objective. On the other hand, a passively adaptive controller, while utilizing the available real-time measurements, does not account for the fact that future observations will be made. Thus any learning in such a case will occur in an accidental manner. This paper summarizes a recent research effort in the development of an actively adaptive control method for nonlinear stochastic systems. A new and simpler set of equations for the original algorithm is given that provides further insight into the concepts of probing and caution in adaptive control. Several examples are chosen to illustrate the actively adaptive control method.