We propose a model of robot “consciousness” that can be used for controlling hyperredundant robots. The model is based on a resident real-time self simulator. It resembles human/animate consciousness in the sense that the robot “imagines” and manipulates a model of itself. This model can be used for evaluating hypothetical situations, without physical trials. We discuss and illustrate by an experiment how consciousness can be used to accelerate learning for snake-like robot locomotion.