Serial manipulators, which have large work space with respect to their own volume and occupied floor space, are the most common industrial robots by far. However, in many environments the situation is unstructured and less predictable, such as aboard a space station, a nuclear waste retrieval site, or a lunar base construction site. It is almost impossible to design a single robotic system which can meet all the requirements for every task. In these circumstances, it is important to deploy a modular reconfigurable robotic system, which is suitable to various task requirements. Modular reconfigurable robots have a variety of attributes that are well suited to for these conditions, including: the ability to serve as many different tools at once (saving weight), packing into compressed forms (saving space) and having high levels of redundany(increasing robustness). By easy disassembly and reassembly features, this serial modular robotic system will bring advantages to small and medium enterprise to save costs in the long term. This thesis focuses on developing such a serial reconfigurable modular robotic system with remote control functionality. The robotic arms are assembled by PowerCube Modules with cubic outward appearance. The control and power electronics are fully integrated on the connector block inside of the modules. Those modules are connected in series by looping through, and can work completely independently. The communication between robotic arms and PC controller is connected by the Control Area Network bus. CAN protocol detects and corrects transmission errors caused by electromagnetic interference. The local PC can directly control the robotic arm via Visual Basic code, and it can also be treated as server controller. Client PCs can access and control the robotic arm remotely through Socket communication mechanism with certain IP address and port number. A Java3D model is created on the client PC synchronously for customers online monitoring and control. The forward and inverse kinematic analysis is solved by Vector Algebraic Method. The Neutral Network Method is also introduced to improve the kinematic analysis. Multiple-layer networks are capable of approximating any function with finite number of discontinuities. For learning the inverse kinematics neural network needs information about coordinates, joint angles and actuator positions. The desired Cartesian coordinates are given as input to the neural network that returns actuator positions as output. The robot position is simulated using these actuator positions as reference values for each actuator.