The advent of the large language model (LLM) has undoubtedly brought robotics to a focal point of academic research and industrial applications. The main advantage of LLM technology lies in its exceptional data processing and pattern recognition capabilities. Through continuous optimization of deep learning and neural network technologies, LLMs can efficiently process and analyze vast datasets. Therefore, LLM-based robotic systems can learn new skills in a shorter time, adapt to new tasks more quickly, and even solve problems autonomously without being explicitly programmed. Moreover, with the aid of LLMs, the perception capabilities of robots have become more accurate, which will undoubtedly greatly expand their application fields and enhance their adaptability and flexibility in complex environments. However, the integration of LLMs and robotics also brings challenges, such as how to ensure the reliability and interpretability of the robot due to LLM integration and how to apply LLM-based robots in practical scenarios. This paper covers a comprehensive and in-depth perspective from theoretical research to practical applications, from hardware design to software algorithms, from perception to task/motion planning, to provide insights into the latest developments and future trends in this field.