A robot manipulator is a force and motion server for a robot. The robot, interpreting sensor information in terms of a world model and a task plan, issues instructions to the manipulator to carry out tasks. The control of a manipulator first involves motion trajectory generation needed when the manipulator is instructed to move to desired positions. The procedure of generating the trajectory must be flexible and efficient. When the manipulator comes into contact with the environment such as during assembly, it must be able to comply with the geometric constraints presented by the contact in order to perform tasks successfully. The control strategies for motion and compliance are executed in real time by the control computer, which must be powerful enough to carry out the necessary computations. This thesis first presents an efficient method for manipulator motion planning. Two fundamental modes of motion, Cartesian and joint, are considered and transition between motion segments is uniformly treated to obtain an efficient and simple system. A modified hybrid control method for manipulator compliance is then proposed and implemented. The method overcomes the problems existing in previous approaches such as stiffness control and hybrid control. Finally, a controller architecture is studied to distribute computations into a number of processors to satisfy the computational requirement in a cost-effective manner. The implementation using Intel's single board computers is also discussed. Finally, to demonstrate the system, the motion trajectory. and the modified forced/motion control scheme are implemented on the controller and a PUMA 260 manipulator controlled from a multi-user VAX/Unix system through an Ethernet interface.