Sensorimotor Self-Recognition in Multimodal Large Language Model-Driven Robots