A Cooperative Adaptive Kalman Filter for Robust Terrain-Referenced Navigation in Complex Environments

Abstract

Accurate and reliable navigation for unmanned aerial vehicles (UAVs) in GNSS-denied environments remains a significant challenge. Conventional terrain-aided navigation (TAN) systems based on extended Kalman filters (EKFs) are prone to divergence due to their reliance on static noise models, leading to spurious lock syndrome and filter failure. Although conventional adaptive Kalman filters (AKFs) attempt to address this, many are reactive and struggle to handle the abrupt transient nature of TAN challenges. This paper proposes a cooperative adaptive Kalman filter (CAKF) that enhances robustness by dynamically aligning the uncertainty representation of the filter with real-world conditions. The CAKF integrates three adaptive mechanisms: scene-ware measurement noise adaptation, manoeuvre-aware process noise adaptation, and health-monitored state covariance intervention. These components cooperate to prevent covariance collapse and maintain filter consistency under varying terrain and dynamic conditions. Simulations in challenging scenarios, including feature-sparse plains, aggressive manoeuvres, and prolonged measurement outages, demonstrate that the CAKF achieves stable and accurate navigation, whereas the Sage-Husa EKF (SHAEKF), a representative innovation-based adaptive filter, suffers from significant errors or divergence. The results validate the effectiveness of the cooperative adaptation strategy in complex TAN applications.