One can design a visual language in which such a robot would present to the user what exactly it’s going to do.

Then a user can say a command, and the resulting action would be displayed, waiting for confirmation by, say, pressing a button.