This three-part paper presents a unified approach to the control of a manipulator applicable to free motions, kinematically constrained motions, and dynamic interaction between the manipulator and its environment. In Part I the approach was developed from a consideration of the fundamental mechanics of manipulation. Part II presented techniques for implementing a desired manipulator impedance. In Part III a technique for choosing the impedance appropriate to a given application using optimization theory is presented. Based on a simplified analysis it is shown that if the task objective is to tradeoff interface forces and motion errors, the manipulator impedance should be proportional to the environmental admittance. An application of impedance control to unconstrained motion is presented. The superposition properties of nonlinear impedances are used to develop a real-time feedback control algorithm which permits a manipulator to avoid unpredictably moving objects without explicit path planning.