The inverse dynamics control of robot manipulators is presented. The main idea of this control method is to cancel the nonlinearities and coupling effects, that describe the dynamic behavior of manipulators, using a dynamic model of the system (primary controller). Since the resulting system is linear and uncoupled, it can be controlled by linear control techniques (secondary controller). The method is initially derived considering the ideal case of the primary controller (where the dynamic model is perfect) and a PD for the secondary controller. The implications of inexact cancelling of the system nonlinearities and coupling effects by the primary controller are shown. The two existing primary controller formulations – computed-torque and feedforward – are described. A hybrid formulations is suggested to overcome the implementation problems of the two formulations. Special attention is given to the simplified computed-torque schemes, which are subject of controversy in the literature. Simulations are performed to better illustrate this matter. A robust linear compensator design, based on the stable factorization approach, is described analyses of some questions and with corrections of the detected mistakes, regarding the existing design. Comparative analyses with the PD are done. The effects of sampling rates on the tracking performances and PD gains are explained.