This paper presents a novel concept of semi-autonomous navigation where a mobile robot evolves autonomously under the monitoring of a human user. The user provides corrective commands to the robot whenever he disagrees with the robot's navigational choices. These commands are not related to navigational values like directions or goals, but to the relevance of the robot's actions to the overall task. A binary error signal is used to correct the robot's decisions and to bring it to the desired goal location. This simple interface could easily be adapted to input systems designed for disabled people, offering them a convenient alternative to existing assistive systems. After a description of the whole concept, a special focus is given to the decisional process, which takes into account in a Bayesian way the environment perceived by the robot and the user generated signals in order to propose a navigational strategy to the human user. The strength and advantages of the proposed semi-autonomous concept are illustrated with two experiments.