Localization in next-generation wireless systems (6G) is envisioned as one of the key enablers for applications such as automotive vehicles, efficient resource allocation, or eHealth. Current localization algorithms heavily rely on GPS (or similar) signals, but their performance is severely degraded in scenarios where the GPS receiver cannot capture the signal from four satellites, e.g., in urban canyons or in indoor environments. In those scenarios, machine learning has emerged as one of the main solutions to provide accurate localization. However, recent approaches rely on a measurement campaign to obtain a database that relates the received signal and the true location of users, which is expensive and time-consuming. Some work has been done to avoid the measurement campaign in localization, but the performance is still far from methods based on labelled data. The goal of this project is to improve the performance of existing localization methods that do not rely on labeled data based on machine learning.