Navwareset: um conjunto de dados de navegação robótica socialmente conformista e não conformista

Abstract

This work presents the development and application of NavWareSet, a new dataset designed for the study of socially-aware robot navigation in indoor environments. The dataset was collected through controlled experiments involving two robots (Toyota HSR and Clearpath Jackal) and 17 human participants, covering seven distinct social interaction scenarios. The recordings include multimodal data such as 3D point clouds, RGB-D images, video, odometry, and velocity commands, along with manually annotated pedestrian trajectories, totaling over 1000 human tracks and 600 robot tracks. To enable the modeling of human-robot interactions, the Social Force Model (SFM) was implemented within the UAIbot simulator, allowing physical simulation of pedestrians in environments with obstacles. Subsequently, the model parameters were calibrated based on real trajectories extracted from NavWareSet, through an optimization process that minimizes the error between simulated and observed trajectories. The results show that the optimized parameters accurately reproduce human behavior and are consistent—with justifiable variations—with values reported in the literature. The main contributions of this dissertation include the public release of a novel dataset, the practical implementation of a classical social navigation model, and the development of an objective methodology for parameter identification using real-world data. This work aims to strengthen the empirical foundation for research in socially-aware robot navigation, promoting reproducibility and standardized algorithm comparison.