A practical, non-linear controller design methodology is presented for robot manipulators guaranteeing that the robot end-point follows an input command vector ‘closelj’ when the robot is not constrained by the environment, and that the contact force is a function of the same input command vector (used in the unconstrained environment) when the robot is constrained by the environment. The controller is capable of ‘handling’ both types of constrained and uncon5trained manoeuvres, and is robust to bounded uncertainties in the robot dynamics. The controller does not need any hardware or software switch for the transition between unconstrained and constrained manoeuvring. Stability of the environment and the manipulator taken as a whole has been investigated, and a bound for stable manipulation has been derived. For stability, there must be some initial compliancy either in the robot or in the environment. A unified approach to modelling robot dynamics is expressed in terms of sensirivity functions, as opposed to the lagrangian approach. allowing the incorporation of the dynamic behaviour of all the robot manipulator elements.