Abstract: Robot task planning bridges high-level decision-making and low-level control, enabling autonomy in long-horizon scenarios. However, as environments become increasingly unstructured and dynamic, resource-constrained single robots struggle to meet complex demands. In contrast, multi-robot systems leverage collaborative operation, heterogeneous resources, and concurrent execution to decouple tasks, significantly enhancing efficiency, robustness, and adaptability. This paper systematically reviews multi-robot task planning research, analyzing modeling and optimization methods for task decomposition, allocation, and planning. Furthermore, it explores integration paradigms combining Large Language Models (LLMs) with traditional symbolic algorithms, synthesizing strengths in semantic understanding and logical reasoning. Finally, the paper outlines future research directions for this evolving field.