A class of active learning control

In the last few years more attention has been given to learning and self-learning in the world of control engineering. These algorithms are restricted to a small class of dynamical systems and some of the conditions for the applicability of these algorithms are very hard, if not impossible, to verify in practice, especially when the a priori knowledge about the system under test is very small or nonexistent. These points are investigated. First, the authors briefly review Betterment theory: both the C⁰ and the C¹-type Betterment are discussed along with the mathematical background. It is noted that from the development of the mixed type Betterment it is possible to generate the proportional-integral-derivative (PID)-type Betterment in a natural way; also some of the strong conditions can be relaxed by the introduction of the C² and C¹-types of Betterment. The physical interpretation of these algorithms is then investigated. An example of the use of the C²-type Betterment is given