Deep Learning is a machine learning technique that seeks to define neural networks based on the recognition of patterns in input data and achieve results with a high confidence level and in a time efficient way. In this paper we apply deep learning in two processes that seek to improve the exploration of a mobile robot in interior spaces. The first process allows the recognition of physical structures such as stairs, elevators and corridors from the images that are obtained from a camera installed on the robot. The second process allows to identify the situation of proximity to the obstacles (from the information of a laser sensor) to decide the direction and speed of the robot to navigate in an interior space. This work applies deep learning in both processes by implementing a parametrizable generic neural network system, in order to create several network instances adaptable to the type of information and its integration with a mobile robot that navigates in simulated and experimental environments.