US-12619260-B2 - Auto-docking of marine vessels

US12619260B2US 12619260 B2US12619260 B2US 12619260B2US-12619260-B2

Abstract

A computer-implemented method for generating control parameters for auto-docking of a marine vessel, and to a control unit for executing the method, to a marine vessel including the control unit, and to a corresponding computer program product.

Inventors

Hamid Feyzmahdavian
Shiva Sander Tavallaey
Stefan Thorburn

Assignees

ABB SCHWEIZ AG

Dates

Publication Date: 20260505
Application Date: 20240122
Priority Date: 20230203

Claims (13)

1 . A computer-implemented method for generating control parameters for auto-docking of a marine vessel from an initial state including a start position, to a final state including an end position within a target total time, the method comprising: in a prior offline session, retrieving datasets including prior control parameters provided by human operators made for docking a marine vessel from a starting point to an end point; providing a machine learning model configured to predict an initial guess of a set of control parameters and a collision-free path for a marine vessel, the machine learning model being trained on the datasets including the prior control parameters provided by human operators made for docking the marine vessel from the starting point to the end point; training the machine learning model on the datasets to mimic control parameters made by a human operator; deploying the trained machine learning model in a marine vessel auto-docking system; providing an optimization function that describes the total time and energy consumption as a function of control parameters for docking a marine vessel from an initial state to a final state, provided constraints on starting point, ending point, and marine vessel dynamics during maneuvering, retrieving, from a computer interface, a start position and an end position for a present auto-docking process for a marine vessel, and locations of obstacles between start and end positions, processing the start position and end position, and locations of obstacles, in the machine learning model to generate an initial guess set of control parameters and collision-free path; determining output control parameters by minimizing the optimization function using the initial guess set of control parameters and the collision free path as starting input to minimizing the total time and energy consumption for auto-docking the marine vessel; providing output control parameters and an output collision-free path extracted from the minimization of the optimization function to an auto-docking system of the marine vessel to control the maneuvering of the marine vessel to perform an auto-docking process; and maneuvering the marine vessel from the start position to the end position using the output control parameters to control thrusters of the marine vessel.
2 . The method according to claim 1 , wherein the state variables of the optimization function include position of the marine vessel in the earth frame, heading angle, surge and sway velocities, and yaw rate.
3 . The method according to claim 1 , wherein the output control parameters comprise required yaw torque and total forces in surge and sway directions.
4 . The method according to claim 1 , wherein the method is performed for each time step during an auto-docking process to update the output control parameters for each time step.
5 . The method according to claim 4 , wherein a repetition rate for the method is based on the size of the marine vessel.
6 . The method according to claim 1 , wherein the target total time is based on a total time required for manual docking by an operator.
7 . The method according to claim 1 , wherein the optimization function includes environmental disturbance parameters including present wind speed, waves, and local sea current speed and direction.
8 . The method according to claim 1 , wherein data of the marine vessel dynamics during maneuvering is pre-stored on a memory device.
9 . The method according to claim 1 , wherein the optimization function is non-linear and non-convex.
10 . The method according to claim 1 , wherein the machine learning model is accessed from a server of a cloud-based service.
11 . A control unit configured to perform the steps of a computer-implemented method for generating control parameters for auto-docking of a marine vessel from an initial state including a start position, to a final state including an end position within a target total time, the method including the steps of: in a prior offline session, retrieving datasets including prior control parameters provided by human operators made for docking a marine vessel from a starting point to an end point; providing a machine learning model configured to predict an initial guess of a set of control parameters and a collision-free path for a marine vessel, the machine learning model being trained on the datasets including the prior control parameters provided by human operators made for docking the marine vessel from the starting point to the end point; training the machine learning model on the datasets to mimic control parameters made by a human operator; deploying the trained machine learning model in a marine vessel auto-docking system; providing an optimization function that describes the total time and energy consumption as a function of control parameters for docking a marine vessel from an initial state to a final state, provided constraints on starting point, ending point, and marine vessel dynamics during maneuvering, retrieving, from a computer interface, a start position and an end position for a present auto-docking process for a marine vessel, and locations of obstacles between start and end positions, processing the start position and end position, and locations of obstacles, in the machine learning model to generate an initial guess set of control parameters and collision-free path; determining output control parameters by minimizing the optimization function using the initial guess set of control parameters and the collision free path as starting input to minimizing the total time and energy consumption for auto-docking the marine vessel, providing output control parameters and an output collision-free path extracted from the minimization of the optimization function to an auto-docking system of the marine vessel to control the maneuvering of the marine vessel to perform an auto-docking process; and maneuvering the marine vessel from the start position to the end position using the output control parameters to control thrusters of the marine vessel.
12 . A marine vessel comprising a control unit configured to perform the steps of a computer-implemented method for generating control parameters for auto-docking of a marine vessel from an initial state including a start position, to a final state including an end position within a target total time, the method including the steps of: in a prior offline session, retrieving datasets including prior control parameters provided by human operators made for docking a marine vessel from a starting point to an end point; providing a machine learning model configured to predict an initial guess of a set of control parameters and a collision-free path for a marine vessel, the machine learning model being trained on the datasets including the prior control parameters provided by human operators made for docking the marine vessel from the starting point to the end point; training the machine learning model on the datasets to mimic control parameters made by a human operator; deploying the trained machine learning model in a marine vessel auto-docking system; providing an optimization function that describes the total time and energy consumption as a function of control parameters for docking a marine vessel from an initial state to a final state, provided constraints on starting point, ending point, and marine vessel dynamics during maneuvering, retrieving, from a computer interface, a start position and an end position for a present auto-docking process for a marine vessel, and locations of obstacles between start and end positions, processing the start position and end position, and locations of obstacles, in the machine learning model to generate an initial guess set of control parameters and collision-free path; determining output control parameters by minimizing the optimization function using the initial guess set of control parameters and the collision free path as starting input to minimizing the total time and energy consumption for auto-docking the marine vessel; providing output control parameters and an output collision-free path extracted from the minimization of the optimization function to an auto-docking system of the marine vessel to control the maneuvering of the marine vessel to perform an auto-docking process; and maneuvering the marine vessel from the start position to the end position using the output control parameters to control thrusters of the marine vessel.
13 . A computer program product comprising a non-transitory computer-readable medium including program code for generating control parameters for auto-docking of a marine vessel from an initial state including a start position, to a final state including an end position within a target total time, the program code including: code for retrieving, in a prior offline session, datasets including prior control parameters provided by human operators made for docking a marine vessel from a starting point to an end point; code for providing a machine learning model configured to predict an initial guess of a set of control parameters and a collision-free path for a marine vessel, the machine learning model being trained on the datasets including the prior control parameters provided by human operators made for docking the marine vessel from the starting point to the end point; code for training the machine learning model on the datasets to mimic control parameters made by a human operator; code for deploying the trained machine learning model in a marine vessel auto-docking system; code for providing an optimization function that describes the total time and energy consumption as a function of control parameters for docking a marine vessel from an initial state to a final state, provided constraints on starting point, ending point, and marine vessel dynamics during maneuvering, code for retrieving, from a computer interface, a start position and an end position for a present auto-docking process for a marine vessel, and locations of obstacles between start and end positions, code for processing the start position and end position, and locations of obstacles, in the machine learning model to generate an initial guess set of control parameters and collision-free path; code for determining output control parameters by minimizing the optimization function using the initial guess set of control parameters and the collision free path as starting input to minimizing the total time and energy consumption for auto-docking the marine vessel; code for providing output control parameters and an output collision-free path extracted from the minimization of the optimization function to an auto-docking system of the marine vessel to control the maneuvering of the marine vessel to perform an auto-docking process; and code for maneuvering the marine vessel from the start position to the end position using the output control parameters to control thrusters of the marine vessel.

Description

TECHNICAL FIELD The present invention generally relates to a computer-implemented method for generating control parameters for auto-docking of a marine vessel, and to a control unit for executing the method, to a marine vessel comprising the control unit, and to a corresponding computer program product. BACKGROUND Docking with marine vessels in a port is efficiently performed by human pilots. However, there is a desire to allow auto-docking control systems to perform the docking process in place of human pilots. However, solutions for auto-docking are too slow compared to experienced human pilots. Accordingly, there is room for improvement with regards solutions for auto-docking of marine vessels. SUMMARY In view of the above-mentioned and other drawbacks of the prior art, it is an object of the present invention to provide a method for auto-docking of marine vessels that at least partly alleviates the drawbacks of prior art. According to a first aspect of the invention, there is provided a computer-implemented method for generating control parameters for auto-docking of a marine vessel from an initial state including a start position, to a final state including an end position, within a target total time, the method comprising: providing a machine learning model configured to predict an initial guess of a set of control parameters and a collision-free path for a marine vessel; providing an optimization function that describes the total time and energy consumption as a function of control parameters for docking a marine vessel from an initial state to a final state, provided constraints on starting point, ending point, and marine vessel dynamics during maneuvering, retrieving, from a computer interface, a start position and an end position for a present auto-docking process for a marine vessel, and locations of obstacles between start and end positions, processing the start position and end position, and locations of obstacles, in the machine learning model to generate an initial guess set of control parameters and collision-free path; determining output control parameters by minimizing the optimization function using the initial guess set of control parameters and the collision free path as starting input to minimizing the total time and energy consumption for auto-docking the marine vessel, and providing output control parameters and an output collision-free path extracted from the minimization of the optimization function to an auto-docking system of the marine vessel to control the maneuvering of the marine vessel to perform an auto-docking process. The present invention is at least partly based on the realization to employ a machine learning model for providing an accurate initial guess for the optimization function. By providing an accurate initial guess to the optimization function, the time for solving the optimization function is reduced, and the accuracy of the output therefrom is improved. The machine learning model is trained on datasets including prior control parameters provided by human operators made for docking a marine vessel from a starting point to an end point. This provides for an accurate initial guess based on control parameters used by experienced human pilots. The datasets of prior control parameters are generated off-line in a simulator mimicking a real marine vessel used by human pilots in different environments, e.g., at different ports, scenarios, and in different environmental conditions. This data is used as input to train the machine learning model. By inputting the start position, end position, and positions of obstacles to the machine learning model, for the present scenario, the machine learning model provides an initial guess set of control parameters and an initial guess for a collision-free path that can be used in solving the optimization function. The machine learning model may be a supervised learning model employing polygonal regression. Preferably, the machine learning model may be a multi-output regression model. Unlike normal regression where a single value is predicted for each sample of data, multi-output regression requires specialized machine learning algorithms that support two or more outputs for each data point and the outputs are required simultaneously. In embodiments herein, the control parameter is a vector including a number of decision variables which should be calculated simultaneously. Thus, the machine learning model is advantageously a Multi-output Regression model. There are several algorithms for solving Multi-output Regression problems, for example, (i) Multi-output support vector regression, (ii) Multi-target regression trees, (iii) Multioutput Regression Neural Network. Each of these algorithms are suitable for embodiments of the present invention. If the data set is of medium size, multi-output support vector regression may have higher performance in terms of better generalization and lower computational time. For larger data sets Multiout