Recent advancements in AI, and specifically Machine Learning, are enabling robots to more seamlessly integrate into our everyday routines. The objective of this thesis is to take a further step towards the development of intelligent autonomous agents that can be embedded in our daily environment, such as houses, hospitals, shopping malls, and so forth. These agents ought to possess the capability to effectively navigate their surroundings to achieve a certain target, such as reaching a certain place in the environment or finding a certain object. Therefore, we examine a wide range of existing techniques for building an embodied navigation agent. These techniques can be fully learned by neural networks
(learned-based techniques) or they can be based on geometry techniques that rely on explicit modeling of the agent and its environment. In this thesis, we build hybrid approaches that use both techniques in such a way that they can work not
only in a simulation but also in a real physical environment. This is a common goal for all the contributions to this thesis.