Direct adaptive neurocontrol of flexible joint robots using localized polynomial networks

Presents a novel direct adaptive neurocontrol scheme for flexible joint robots with structural uncertainty and arbitrary nonlinear joint flexibility. The localized polynomial networks are used to represent unknown system dynamics. It is proved that all the signals in the closed-loop direct adaptive neurocontrol systems can be made uniformly bounded and the output tracking errors can be guaranteed to converge globally to a specified neighborhood of zero. Therefore, global stability of the neurocontrol systems is guaranteed. The learning process is fast convergent and needs less computation, due to the usage of the localized polynomial networks. Compared with conventional schemes, this direct adaptive neurocontrol scheme allows robotic dynamics to have more general structure and is more robust to robotic system modeling errors. No link acceleration and jerk measurements are needed, and the control actions are chatter-free