Neural-assisted in-motion self-heading alignment for marine vehicles

Abstract

Autonomous platforms operating in the oceans require accurate navigation to successfully complete their missions. In this regard, the initial heading estimation accuracy and the time required to achieve it play a critical role. The initial heading is commonly estimated using model-based orientation decomposition methods based on inertial measurements. These methods rely on two assumptions: small platform velocity and known approximate position. Under these assumptions, the inertial quantities in the navigation frame can be determined. However, methods such as dual-vector decomposition and optimization-based attitude decomposition achieve satisfactory heading accuracy only after long alignment times. To address this limitation and enable fast and accurate initial heading estimation, we propose an end-to-end, model-free, neural-assisted framework that uses the same inputs as the model-based approaches. That is, the platform is moored and thus quasi-stationary, satisfying the small velocity condition, while the position is provided by the global navigation satellite system. The proposed approach was trained and validated using a real-world dataset captured by an autonomous surface vehicle. The proposed method demonstrates an average accuracy improvement of 51% and up to an 83% reduction in alignment time. It estimates the initial heading using only inertial measurements as input, producing the heading angle at the end of the alignment period. Thus, the proposed approach enables reduced alignment time and improved accuracy, allowing shorter deployment times and enhanced navigation performance during the mission.