With the progress of innovative technologies, ships in future with different autonomy levels are anticipated to enter the realm of maritime transportation. As a result, the scenarios of multi-ship encounters at sea can become more complex and the risk of potential collisions can be difficult to elevate. To support navigation safety and guarantee the required situation awareness level, it is therefore essential to acquire ship navigation states with a greater degree of precision. The Kalman Filter (KF)-based techniques are one of the popular approaches for deriving the ship navigation state by merging the prior estimates from physics-based models with measurements from onboard sensors. However, many KF-based estimates are calculated by assuming constant system and measurement uncertainties during the iterative process. In this study, an adaptive tuning mechanism in the KF-based techniques is utilized to estimate ship navigation states. This approach enables the estimation processes to skillfully reduce both system and measurement noises estimations. Consequently, it results in the generation of smoother and more responsive estimates of the respective vessel states, particularly when confronted with variations in rudder orders or encountering abnormal measured positions.
Autonomous shipping is expected to exist and show its benefits in maritime transportation in the coming future. With respect to the current development of autonomous ships, onboard sensors are considered as one of the fundamental parts (Perera, 2019;Thombre et al., 2022). These sensors are specifically designed to provide digital navigators of autonomous ships with precise navigation information, ensuring the maintenance of adequate situation awareness. Since not all ship navigation information can be directly measured, and the measurements from sensors may contain measurement noise, it is generally considered to implement KF-based techniques to generate estimated states with higher precision.
Given the convenience of kinematic motion models, using KF-based estimation combined with these models is a favorable choice (Li and Jilkov, 2003). One advantage of employing kinematic motion models is the ignorance of hydrodynamic coefficients which are associated with external forces and moments. The influences of external disturbances can be modeled as system noises in kinematic motion models. The curvilinear motion model (CMM) and the constant turn rate and acceleration model (CTRA) represent two kinematic motion models which can encompass diverse motions exhibited by ships. As a result, these models are widely employed in numerous research studies to provide essential ship navigation states (Perera, 2017;Wang et al., 2023). Nevertheless, it is important to recognize that these kinematic motion models operate under the assumption of constant accelerations and turn rates. Clearly, this assumption becomes invalid when ships execute new rudder orders or adjust engine power. Consequently, the utilization of constant system noises in the CMM and CTRA within the KF-based estimation can lead to less precise estimates. Another crucial factor to consider is the potential for measurement bias when utilizing GNSS systems. To enhance measurement precision, the augmentation of GNSS is actively encouraged. However, it is important to note that while such augmentation can improve precision, it does not guarantee the elimination of measure abnormalities (Baybura et al., 2019). When the measured outliers exceed certain thresholds, the KF-based estimation may yield questionable results.
