Embodied cognitive robotics has made remarkable advances in modelling natural cognitive processes and thus speculating about the content of the black box of the mind. The field’s tackling of one of the biggest questions in cognitive science, the question of meaning, has been met with considerable acclaim. The latter was gained by modelling the grounding of meaning through an agent’s use of language as a cognitive tool in its active interaction with the environment via sensorimotor behavior. Many different approaches with such an agent have been labelled as successful in modelling meaning comprehension, which begs the question of what natural understanding and meaning really are. This points to such embodied cognitive models lacking – their grounding is neither meaningful nor significant to them, and it is not internally necessary, but is rather forced onto them by externally programmed fundamental goals. What is missing is a non-trivial, constitutive autonomy, which would make robots create their own goals in order to self-regulate and preserve their internal organization; in order to survive. Only then can meaning truly be assigned.