Dual-Branch INS/GNSS Fusion with Inequality and Equality Constraints

Researchers Mor Levenhar and Itzik Klein have published a paper introducing a novel dual-branch information aiding framework for vehicle navigation in challenging urban environments. The system addresses the persistent problem of satellite signal blockages caused by tall buildings by fusing inertial navigation system (INS) readings with global navigation satellite system (GNSS) data through a variance-weighted scheme that incorporates both equality and inequality motion constraints. Unlike traditional approaches that rely on rigid equality assumptions about vehicle motion—which often fail during dynamic urban driving—this method uses physically motivated inequality bounds to create more robust navigation during precisely the conditions where aiding is most needed. The technique requires only software modifications to existing navigation filters, making it a cost-free upgrade path for current systems.

The researchers evaluated their method on four publicly available urban datasets spanning 4.3 hours of recorded data with various inertial sensors, road conditions, and driving dynamics. Under full GNSS availability, the dual-branch approach reduced vertical position error by 16.7% and improved altitude accuracy by 50.1% compared to standard non-holonomic constraint methods. More importantly, during GNSS-denied conditions—when vehicles lose satellite signals—the system reduced vertical drift by 24.2% and improved altitude accuracy by 20.2%. These results demonstrate that replacing hard motion equality assumptions with flexible inequality bounds represents a practical strategy for enhancing navigation resilience without relying on additional sensors, expensive map data, or complex learned models. The approach could significantly improve the performance of autonomous vehicles, delivery drones, and other robotic systems operating in dense urban environments where traditional navigation systems frequently fail.