This project develops and evaluates robust radio-localization methods for future wireless networks using site-specific digital twins. Three-dimensional urban propagation environments will be simulated with Sionna RT to generate line-of-sight and multipath radio-channel observations under controlled variations in user position, noise level, clock bias, material properties, and propagation-model mismatch. The resulting channel data will be processed using sparse channel estimation, geometric localization, probabilistic data association, robust hypothesis selection, and fingerprinting-based methods. The research focuses on determining when multipath information improves positioning and when incomplete or mismatched digital-twin models lead to unreliable estimates. Large, reproducible experiments will be conducted across multiple urban scenes, user locations, signal-to-noise conditions, and random realizations. CPU resources will be used for sparse recovery, probabilistic inference, Monte Carlo evaluation, parameter sweeps, and statistical analysis, while GPU resources will accelerate ray-tracing simulations. The expected outcomes are validated localization algorithms, quantitative reliability measures, and a systematic characterization of failure modes and generalization across environments. These results will support reliable positioning and sensing functions in future 6G communication systems.