State representation learning using robotic priors in continuous action spaces for mobile robot navigation

State representation learning in the context of robotics can be used to increase the efficiency of learning a task, while keeping a high performance.

One way of learning such a state representation in the context of robotics is to make use of robotic priors. Learning a state representation using robotic priors has mainly been applied in fully observable, Markovian, environments. In this thesis the case of a partially observable environment will be considered. To make the state Markovian, a recurrent neural network will be used to map the partially observable observations to a fully observable state.

This will be used to discover the limits of the robotic priors proposed in literature, and possible extensions to these robotic priors could be proposed.