2D Grid Map Generation for Deep-Learning-based Navigation Approaches

Abstract

In the last decade, autonomous navigation for robotics has been leveraged by deep learning and other approaches based on machine learning. These approaches have demonstrated significant advantages in robotics performance. But they have the disadvantage that they require a lot of data to infer knowledge. In this paper, we present an algorithm for building 2D maps with attributes that make them useful for training and testing machine-learning-based approaches. The maps are based on dungeons environments where several random rooms are built and then those rooms are connected. In addition, we provide a dataset with 10,000 maps produced by the proposed algorithm and a description with extensive information for algorithm evaluation. Such information includes validation of path existence, the best path, distances, among other attributes. We believe that these maps and their related information can be useful for robotics enthusiasts and researchers who want to test deep learning approaches. The dataset is available at https://github.com/gbriel21/map2D dataSet.git.