Route-constrained GIS method keeps vehicle position within about 3 m without GNSS fixes

Researchers have published a localisation framework in Scientific Reports that keeps an autonomous vehicle's position estimate going when GNSS position updates are unavailable. It tracks one-dimensional progress along a known route, built from OpenStreetMap road centrelines with the Open-Source Routing Machine. Incremental distance comes from onboard motion data, estimated with either an LSTM network or a Kalman filter, and mapped landmarks (intersections and parking-lot access points) reset accumulated error.

Tested on a 2.53 km urban corridor with 23 landmarks, the full method gave a mean error of 3.445 m with the Kalman filter and 3.410 m with the LSTM. Unconstrained dead reckoning gave 32.631 m. In simulated outages the fallback averaged 2.723 m, against 242.800 m when the last valid position was held. Delayed or wrongly matched landmark corrections degraded results far more than missed ones, with the worst misassociation case reaching 66.933 m.

Why it matters: Shows that cheap road-network data and a few mapped landmarks can bridge GNSS outages at metre level, useful for urban navigation where signals degrade. Limits apply: one corridor, predefined landmark events, and recorded GNSS speed still used.