Visual Embodied Brain: Let Multimodal Large Language Models See, Think, and Control in Spaces | Synapse