Exploração autônoma de ambientes baseada em ganho de informação

Abstract

The exploration of unknown environments using autonomous mobile robots is essential for different applications, for example, search and rescue missions after natural disasters. This is due to the performance capability of these robots in unknown environments even when the human intervention is deprived. The main objective of this task is to efficiently transverse the environment and build a complete and accurate map. However, different applications may demand different exploration strategies. The simplest strategy is a greedy approach which visits the closest frontier without considering whether it will yield a significant reduction in map uncertainty. In this dissertation, we proposed two main contributions. First, we elaborated a novel method to predict information beyond the candidate frontiers by analysing the local structures in a building map. In this way, it turns out possible to estimate the information gain of each frontier candidate to be the next destination using Shanon entropy. Afterwards, the second contribution was developed with the intention to create a unified planning which allows the robot to identify the best destinations and, at the same time, its own paths. The methodology was evaluated through several experiments in a simulated environment, showing that our exploration approach is better suited for rapid exploration than the classic Near-Frontier Exploration (NFE).