Probabilistic Semantic Mapping for Vision-Based Systems: Integrating Perception and Uncertainty

This study presents a system implemented on a quadruped robot, designed to generate a multi-layered map that incorporates semantic information and uncertainty, specifically tailored for Search \& Rescue robotics. The article details the development of a machine learning model based on visual data for recognizing people and environmental features and its integration into an architecture for mapping and navigation. Extensive testing of the system in two different locations, conducted using Boston Dynamics' Spot robot with an external ZED2 depth camera, is also described.