Reliability of fingerprint-localisation systems under real-world conditions

Establish the reliability of sensor-based fingerprint-localisation systems under real-world operating conditions involving changes in devices, subjects, sites, and recording periods, beyond evaluation on static benchmark datasets.

Background

The paper situates its staged evaluation protocol within a broader gap in fingerprinting research. Existing work has often evaluated localisation systems on static benchmark datasets or within a single measurement campaign, leaving their reliability under operational distribution shifts insufficiently characterised. The relevant shifts include device replacement, personnel changes, site variation, and long-term operation.

The authors identify reliability under real-world conditions as an open problem named in recent fingerprinting work and present their protocol as an evaluation instrument intended to address that gap. The problem remains broader than the two underground-mine case studies and requires establishing reliability across the diverse conditions encountered by deployed sensor-based AI systems.