This research presents a novel method for identifying hydrodynamic model parameters that are essential for the deployment and recovery of deep-sea mining vehicles. The proposed methodology utilizes a gradient descent (GD)-based optimization technique to minimize a loss function specifically formulated based on the governing motion equations relevant to these processes. This approach yields more accurate hydrodynamic parameters, thereby enhancing predictive capabilities. A key aspect of this article is the comparative analysis between the proposed method and the Unscented Kalman Filter method, demonstrating that GD optimization reduces local time-series errors and exhibits greater stability. Importantly, this technique does not require evenly spaced datasets, thus simplifying data input and enhancing practical applicability. Furthermore, the ease of reusing both the GD optimizer and the loss function facilitates efficient computation of motion data for deep-sea mining vehicles, surpassing traditional filtering methods in terms of input sample data requirements. This article represents a significant advancement in the field of deep-sea vehicle dynamics by introducing a GD-based methodology for hydrodynamic parameter identification.

Gradient descent-based hydrodynamic parameter identification for deep-sea mining vehicle deployment and recovery