The rapid development of power artificial intelligence (AI) technology have brought significant benefits to various aspects of the grid. However, a critical challenge remains: the lack of automated actuators for AI-generated strategies, which currently rely heavily on human execution. This paper addresses the need for intelligent robots in power grid applications, particularly in inspection and live-line work, to enhance operational safety and efficiency. We explore the potential implementation of embodied intelligence (EI) technology, focusing on two primary approaches: the integration of large and small models for vision, language, and action, and end-to-end visual-language-action (VLA) models. The first approach, exemplified by the PixelNav framework, demonstrates effective navigation and manipulation capabilities, while the second approach, represented by models like RT-2 and 3D-VLA, shows promise in generalization and robustness. Additionally, we discuss the application of these EI technologies in specific power grid scenarios, such as substation and distribution station inspections, and highlight the importance of developing more intelligent and adaptable robotic systems. The aim of this paper is to provide a comprehensive overview of the latest advancements in EI technology and to offer practical insights and directions for its deployment in the power grid, ultimately contributing to the stability, reliability, and safety of the power system.