The smart meter plays an increasingly important role in power grids. However, operation failures, caused by different faults during power metering, are often observed. Therefore, improving the robustness of smart meters becomes important and has been focused by many studies. In this paper, based on an introduction of the general structure and function of a smart meter, we summarize different techniques for improving the smart meter robustness into two categories, i.e. the inherent reliability and the use reliability respectively. The inherent reliability includes the structure design, hardware reliability analysis, key module development, software testing and so on. While the use reliability focuses on measurement error compensation and operating status detection for smart meters. The latter category involves evaluation systems based on artificial intelligence technologies, such as the full life cycle theory, neural networks, Bayesian algorithm, K-means algorithm, etc. Future development trends towards a more robust smart meter are discussed.