Abstract: In the field of embodied intelligence, long-horizon and multi-stage tasks have long been recognized as a central challenge due to their inherent complexity. This article introduces the concept of embodied behavior planning and provides a comparative analysis to distinguish it from several related notions. Furthermore, it examines the key challenges currently facing embodied behavior planning, including the lack of convergence in model architectures, the relative scarcity of high-quality datasets, and the inconsistency in the design of evaluation tasks. In addition, this article reviews representative approaches proposed in recent researches and discusses potential future directions for the development of Embodied Behavior Planning.