The project focuses on the creation of models of lattice structures. These types of structures are found in numerous constructions and require continuous maintenance. This maintenance can be automated using a mobile robot capable of moving through the structure. However, to undertake this task, the robot must have a model of the structure that allows it to know its position and plan the appropriate trajectory and sequence of movements to reach the target point. To create this model, the robot will collect information as it moves along the structure for the first time, using its equipped sensors (primarily omnidirectional vision systems).

Modeling these types of structures presents several differential aspects compared to other environments, such as their symmetry and the presence of repetitive visual structures, the wide variety of viewpoints from which they can be observed depending on the robot's trajectory, and the changes their appearance may undergo due to repairs carried out by the robot. Considering these characteristics, the model will be given a hierarchical structure, with a high-level layer containing information about the topology of the structure, and one or more low-level layers with data about the bars and nodes, such as their shape, width, planes that compose the bars, and node topology. Artificial intelligence and deep learning techniques will be used for scene description and extraction of relevant information. These tools will separate the information surrounding the structure and its conditions (such as lighting conditions) from the information of the lattice surrounding the robot (bars and nodes). Additionally, algorithms will be implemented for incremental model creation, updating it as the robot advances and captures new information about the structure.