In this thesis, the robust control of flexible joint robots with the emphasis on the performance and control effort limitation is analyzed in detail. First, a special control algorithm, named composite PID, is elaborated, and the main drawback of it to provide solutions for limited control effort is described. In order to remedy this drawback, with a new approach to control synthesis, an H. framework is proposed. In this framework, linear identification techniques are used to represent the open-loop system nonlinear dynamics as a linear model with multiplicative uncertainty. Then, an  controller is designed for the system with the emphasis on robust performance and especially limited control effort. By this means, relatively suitable tracking performance is obtained with much smaller control effort. These desirable performances cannot be obtained if in the uncertainty representation, sharp peaks were not neglected. Despite the practical importance of this method, above assumption causes robust stability of the closed-loop system cannot be claimed rigorously. In order to remedy this theoretical draw back and have the benefits of composite control in addition to  controller, it is proposed to combine these methods, in which the PID controller of the composite control algorithm is replaced with an  controller, designed for performance. It is observed that on the contrary to composite PID control, the composite  control is robustly stable, despite control effort limitations. Moreover, the proposed composite  control law can provide similar tracking performance to composite PID, with much smaller control effort.