Nonlinear feedback control for force-controlled robots with constrained end-effector motion is considered. A general method is presented that assures an exact feedback linearization for both rigid and flexible-joint robots, as the joint flexibility can cause instability of robot control. The feedback control linearizes and decouples the original nonlinear system into a number of decoupled linear subsystems. The effect of stiction on the end-effector contact with the environment is inherently incorporated in the formulation, using the same constrained system formalism. A version of the controller with improved robustness characteristics, based on the robust servomechanism theory, is proposed. The derivation of the control algorithm for a two-link planar robot interacting with a rough plane surface is presented as an example. Numerical simulation results confirm the effectiveness of the method. The issues associated with real-time robot control, such as the choice of sampling frequency and the influence of modeling errors, are discussed. >