Reliability-governed scaling of mobile crowdsensing-based 3D radio mapping for ubiquitous indoor positioning: An uncertainty-aware lifecycle review and deployment-oriented pipeline

Mobile crowdsensing (MCS)-based 3D Radio Mapping (3D-RM) is a primary pathway toward ubiquitous Wi-Fi fingerprinting-based indoor positioning systems (IPS). Its scalability depends on replacing site surveys with weakly labeled, in-the-wild, user-governed logs in a realistic deployment setting where generic RM construction must be infrastructure-free, opportunistic, and largely self-calibrating, without assuming device-specific calibration or skilled users, while remaining robust across cities without overfitting. In deployment-realistic conditions, the central challenge shifts from data acquisition to reliability: namely, deciding which uncertain evidence can be safely promoted into a persistent RM structure without inducing error-amplifying feedback over the long term, particularly in multi-floor buildings where vertical ambiguity and cross-floor mis-association are common. Existing reviews survey major waves of progress, yet they do not provide a deployment-realistic synthesis of the RM lifecycle under weak supervision that connects cold-start bootstrapping, progressive globalization, 3D assembly, and maintenance through reliability governance. This paper addresses this gap by providing, to the best of our knowledge, the first survey that frames autonomous MCS-based 3D-RM as a reliability-governed lifecycle, adopts an uncertainty-aware evidence lens and a pipeline-coupled uncertainty stack, explicitly tracing how sensing noise and weak-label attachment errors propagate into anchoring and alignment uncertainty, graph-level structural aggregation, including multi-floor 3D assembly, and long-term temporal drift. Through this lens, it investigates cold-start generation and RM updating as high-stakes, gated renovation problems. It further discusses deep and continuous learning, sequence and graph models, and LLM-enabled data operations as reliability-bounded learning processes. Finally, it outlines a research agenda toward globally scalable, 3D-RM-governed reliability.