Cloud robotic network (CRN) is an extension of the traditional robots, and its important significance is to use the Internet and cloud computing to overcome the limitations of single robot and enhance its computing power. As the core technology of CRN, how to achieve the full potential of computation off‐loading is a critical issue. Starting from the basic models of CRN, this paper studies the network architecture and the computation off‐loading method to solve this problem. Firstly, a relatively detailed system model is designed based on the characteristics of CRN to provide theoretical and quantitative basis for subsequent research. Then, based on the theory of connected dominating set, a distributed CDS construction algorithm PCDS is proposed, which provides a virtual backbone with more power for the network. Finally, an energy sensitive computation off‐loading method is proposed to improve the lifetime of the whole network while ensuring the execution time and energy loss of computational tasks. Simulation results show that the proposed framework can effectively improve the performance of CRN and prolong the lifetime to a certain extent.