The Embodied World Model Based on LLM with Visual Information and Prediction-Oriented Prompts