This thesis tackles the problem of Semi-Autonomous Navigation, where the robot is able to propose the navigational actions to the user, and performs them in the absence of negative feedback. Having the brain electroencephalographic (EEG) signals as means of communication between the user and the robot, this negative feedback comes in terms of error-related potentials (ErrP), elicited in the brain after observing erroneous response of the robot or after inappropriate action proposition. Due to the low bandwidth of brain computer interfaces (BCIs), intelligent robotic systems, that have the ability to learn the behavior of the user and thus can predict his/her intentions or rank his/her preferences, are needed to make the best use of the scarce resources.

Among the learning algorithms available in the literature, reinforcement learning stands as a promising approach. Error-related Potentials (ErrP) can be exploited to reward/punish a robot when performing (proposing) an action. The figure below shows an overview of the system. The robot is equipped with a set of sensors (e.g. laser range finder, optical encoders, ... ), by which it gets a feedback from the environment, and increases its autonomy. Both the environment and the user rewards/punishes the robot for the proposed and the carried out actions. After receiving the reward/punishment signals, the robot should update its strategy of taking actions on behalf of the user.