ThisSelf-evolving embodied AI chapter introduces self-evolving embodied Artificial Intelligence (AI), a critical frontier for developing autonomous agents like robots that can learn and adapt directly within the physical world. We define these agents as embodied systems that continuously and systematically refine their own internal models and policies through dynamic, real-world interaction. The chapter analyzes three key technologies enabling this evolution: Meta Learning, which allows rapid task adaptation; Curriculum Learning, which organizes the learning process progressively from simple to complex tasks, and Lifelong Learning, which facilitates knowledge accumulation over time without catastrophic forgetting. We then deconstruct this paradigm into its fundamental components: Self-evolving World Models and Self-evolving Agents. This latter concept encompasses intelligence emerging from embodied cognition and dynamic interaction. This work provides a comprehensive roadmap toward a new generation of AI, shifting from autonomous pattern recognition to self-evolving, physically grounded intelligence.