The robot is an electromechanical system that combines cutting-edge technologies from various interdisciplinary fields and has a wide range of applications in the real world. With the advent of "Industry 4.0" and the era of artificial intelligence, a new generation of robots is developing in the direction of lightweight and modularization, and its functions are becoming more and more abundant. Automatic trajectory generation, automatic parameter identification, force control, human-machine cooperation and other functions have gradually become the standard for a new generation of robots. However, the architecture of the robot control system has not changed much, it is essentially a two-level architecture of the upper controller-motor driver. Since the intermediate points of robotic joint motion are generated by the upper controller, there are two potential limitations to this widely adopted architecture: 1.Only increasing the number of samples at the intermediate point can increase the trajectory accuracy, which can result in huge data traffic, and a large amount of data may exceed the upper limit of the fieldbus system bandwidth. 2. The joint driver is a passive actuator that can only accept the guidance of the upper controller and cannot generate complex motion trajectories. A traditional robot control system with a two-level architecture is equivalent to humans using the brain to directly guide the movement of each muscle. However, when humans exercise, the brain does not think too much about the details of the movement of muscles and joints. In fact, the nervous system of higher animals uses a three-level architecture (brain-spinal center-muscle) for motor control, and its motion is generated by a combination of basic motion modules. This motor control architecture offers the possibility to overcome the limitations of robot control system. The main contributions are: 1.The characteristics of the existing robot control system are analyzed, and the relevant results of neuroscience are referenced. We imitate the structure of the neural motor control system, introduce the motion modularity concept into the robot system, and propose a robot control system architecture based on the motion modularity technology, which can reduce the real-time data transmission volume of the bus without losing the accuracy of the trajectory， improve joint autonomy. 2. The theory of Motion Description Language is studied. Combining this theory with the theory of Hilbert's space function decomposition, the MDLg method for robot arm trajectory generation is proposed for the characteristics of manipulator motion, and the motion modularity control system is preliminarily established. 3. The Dynamical Movement Primitives theory is introduced to solve the problem that the dynamic performance of the MDLg method is not good and the number of primitives is indeterminate. The feedback signal is introduced into the dynamical primitive, and a DMP time-coupled controller is designed for the joint, which improves the joint antijamming capability. A more complete motion modularity robot control system is established. The study of the thesis presents a feasible motion modularity solution for robot control system. Based on the research results presented in this paper, a robot control system with motion modularity function is designed, which includes three subsystems: upper controller, fieldbus protocol stack and motor driver. Each part develops special components for the realization of motion modularity technology. The scheme can be applied to an actual robot control system to improve the motion performance of the robot without increasing the fieldbus load. This thesis also contributes to the research and development of key technologies for EtherCAT system and motor driver.