Goal-directed synergistic fusion navigation for robust localization under cross-scenario spatial uncertainty

Abstract

Abstract A goal-directed synergistic fusion navigation framework is essential for enabling accurate and reliable navigation of autonomous unmanned systems across diverse operational scenario environments, especially under challenging conditions such as Global Navigation Satellite System (GNSS)-denied and visual-degraded/Light Detection and Ranging (LiDAR)-degraded environments. However, it is almost inevitable to encounter spatial uncertainty arising from the spatiotemporal heterogeneity of perceptual information, which may lead to the degradation and the unreliability of navigation. To address this issue, this article proposes a synergistic fusion navigation method against spatial uncertainty inspired by human navigation. Within a unified framework, the system transforms multi-source sensor observations into navigation landmarks, and an online quantitative assessment of sensor availability analysis is performed by analyzing their effective quantity and distribution to define the Degree of Sufficiency (DoS). Based on this, we further design an adaptive state estimation strategy that dynamically adjusts the filtering and optimization processes according to the DoS. To improve the fusion performance, Degree of Observability (DoO)-driven information credibility analysis, opinion dynamics modeling, and factor graph optimization are integrated for the optimal fusion of multi-source navigation information. Real-world experiments show that in complex cross-scenario environments, the proposed method achieves localization Root Mean Square (RMS) errors ranging from 0.32 m to 0.97 m over test trajectories spanning 419.91–1429.80 m. It outperforms State of the Art (SOTA) algorithms and ablation models in both accuracy and reliability.