Redes neurais artificiais na otimização da computação de caminhos seguros em terrenos com topografia

Abstract

The computation of shortest paths (pathfinding) optimized for the topographical features of real world terrains has been investigated in different problems in the areas of robotics, computer games and simulation. Pathfinding for topographical realistic terrain models involves finding a least-cost path evaluated in terms of distance traveled according to the slopes of the terrain. In path computations performed to solve tactical and strategic problems, the pathfinding of safe paths and with lower topographical costs is a subject still little explored in the literature. Among other aspects, path safety may be related to the visual concealment of an agent traveling along the path in relation to an observer located at a position on the terrain. The objective of this work is to investigate how the computations of a heuristic pathfinding algorithm A∗ should be performed so that an agent can use a path that allows the agent’s stealth locomotion in relation to an observer. This indicates that the terrain relief features are important both for obtaining a low-cost path and for computing the observer’s field of view and the consequent agent stealth according to the terrain relief. For the use of heuristic pathfinding algorithms, determining an efficient heuristic in terms of execution time and node expansion is a challenging task in this secure topographic pathfinding problem. The proposal of this work, therefore, also involves using and evaluating the learning capacity of a deep neural network in the approximation of the heuristic to be used by the A∗ algorithm, thus allowing to optimize the computations of safe routes with lower costs.