Intelligence in control of complex robotic systems

Abstract

In order to give complex mechanical systems more capabilities in their dynamic behavior, it is recognized that they belong to a class of systems for which the classical approach has to be amended to comply with their very structure. This means that, as they exhibit a complicated and mixed pattern of displacements and deformations, their trajectories become exceedingly complicated and no structured information can be obtained from them. This leads to the concept of functional control, which is precisely aimed at globally controlling the system from its accessible previous rigid variables without entering the (unusable) detail of exact trajectories. Milder asymptotic convergence of the solution of the error equation is obtained than with (inapplicable) classical trajectory control (typically, polynomial vs. exponential decay). A possible improvement is obtained by adding a learning-type control, allowing past trajectory information to be accounted for by setting its parameters so that a converging fixed point property can be obtained.