Embedded Robotics must find internal rewards to motivate the robot and allow it to measure its progress. Some approaches investigate the predictive accuracy of the world model (see Intrinsic Motivations, Oudeyer et al).
The PhD will investigate how the exploration of the sensori-motor space can be sustained by measuring and maximizing the quantity of information of the interaction between the robot and the world - before building any World Model.
The approach will rely on the Monte-Carlo Tree Search and extend the same principles used in the Computer-Go champion program, MoGo.