Abstract: This paper presents a task based global parallel algorithm for robot control. Considering the centralized MIMD (Multiple Instruction Multiple Data) parallel platform, the basic computation tasks such as kinematics, dynamics and control theory are respectively partitioned into several subtasks. After that, all the subtasks are assigned to processors using the centralized dynamic load balancing strategy. A simple homogeneous loosely coupled MIMD parallel process system is built with five DSPs. With pipeline and centralized dynamic load balancing strategy, real-time simulations on a planar robot manipulator gain satisfactory performance.