An offshore unmanned inspection boat autonomous navigation system based on a marine large model drive

Abstract

The application relates to the technical field of non-electric variable control and water vehicle navigation control, and discloses a near-shore unmanned inspection boat autonomous navigation system based on a marine large model driving. The system comprises a situation awareness fusion unit, an environment prediction and control constraint generation unit, a probability safety trajectory generation unit and a multivariable collaborative anti-disturbance control unit; a short-time environmental disturbance is predicted according to wind, flow, wave, water level and water depth data at a boatborne end, reference trajectory constraints, obstacle probability distribution and disturbance compensation are generated, and rudder angle and main engine speed instructions are synchronously output, so that closed-loop stable control of the position, heading, speed and roll attitude of the near-shore unmanned inspection boat is realized.

Description

Technical Field

[0001] This invention belongs to the field of non-electric variable control or regulation technology, specifically relating to the position, heading, speed and attitude control system of waterborne vehicles, and particularly to an autonomous navigation system for near-shore unmanned patrol boats that uses marine environment prediction results to assist in closed-loop control of unmanned patrol boats. Background Technology

[0002] Nearshore patrol missions typically require unmanned patrol vessels to continuously navigate around ports, waterways, offshore facilities, and shallow water areas. These operational areas are characterized by high target density, rapid changes in wind and current, complex water depth boundaries, and strict safety distance requirements. The ability of unmanned patrol vessels to stably maintain their predetermined position, course, speed, and roll attitude directly affects patrol efficiency and navigation safety.

[0003] Existing autonomous navigation solutions for unmanned surface vessels (USVs) typically handle environmental perception, local obstacle avoidance, and motion control separately. The perception stage mainly outputs the target position or obstacle boundaries, while the control stage adjusts the heading or speed based on a reference route. This approach works in calm water or minimally disturbed environments, but in complex nearshore waters, obstacle movement is affected by advection and turbulence, and the vessel's roll, yaw, and speed fluctuations are influenced by wind, current, and waves. Relying solely on geometric distance or single heading control is insufficient to promptly reflect future disturbances.

[0004] Therefore, an autonomous navigation system with non-electric variable control of watercraft as its core is needed. The environmental prediction results output by the large ocean model are limited to control constraints and disturbance feedforward inputs, so that reference trajectory generation, obstacle avoidance safety constraints, and rudder angle and main engine speed control work together in the same closed loop, thereby improving the track maintenance capability and attitude safety of near-shore unmanned patrol vessels in complex sea conditions. Summary of the Invention

[0005] The purpose of this invention is to provide an autonomous navigation system for near-shore unmanned patrol vessels based on a large ocean model, which solves the problems of insufficient coupling between environmental disturbance prediction and hull motion control, lack of water movement prediction for dynamic obstacle avoidance, and the disconnect between heading, speed and roll attitude control in existing solutions.

[0006] To achieve the above objectives, this invention adopts the following technical solution: an autonomous navigation system for near-shore unmanned patrol vessels driven by a large ocean model, comprising a situational awareness fusion unit, an environmental prediction and control constraint generation unit, a probabilistic safe trajectory generation unit, and a multivariable collaborative disturbance rejection control unit. Each unit forms a closed loop around the unmanned patrol vessel's position, heading, speed, and attitude control. The prediction results from the large ocean model serve only as inputs for reference trajectory constraints, disturbance estimation, and control gain adjustment. The final controlled objects are the rudder angle, main engine speed, and their corresponding hull motion states.

[0007] The situational awareness fusion unit collects shipborne satellite positioning data, inertial measurement data, marine radar point clouds, meteorological and hydrological data, water level or wave data, and multibeam bathymetry data. Through time synchronization, coordinate alignment, anomaly removal, and semantic hierarchical processing, it forms a panoramic situational field under a unified spatiotemporal reference.

[0008] The environmental prediction and control constraint generation unit receives wind, current, wave, water level and water depth information from the panoramic situation field, constructs a local gridded dynamic environmental field, adopts a large ocean model pre-trained with ocean observation data and with physical residual constraints and observation assimilation correction, outputs predicted data of flow field, water level field and wave height field within a preset time window, and uses the predicted eddy current intensity, wave height and flow velocity changes as control constraint parameters.

[0009] The probabilistic safety trajectory generation unit correlates the obstacle's own motion with the predicted water motion at its location to generate a probability distribution of the obstacle's future state transition. Then, under the constraints of hull kinematics, safe collision avoidance distance, and roll safety, a reference obstacle avoidance trajectory with the lowest overall risk of deviation from the course, excessive roll, energy consumption, and collision risk is selected.

[0010] The multivariable cooperative disturbance rejection control unit decomposes the reference obstacle avoidance trajectory into reference position, reference heading, reference speed, and reference allowable roll angle. Based on extended state observation and adaptive gain adjustment, it generates a cooperative control law and synchronously outputs rudder angle commands and main engine speed commands, enabling the unmanned inspection vessel to maintain cooperative stability of position, heading, speed, and roll attitude when performing obstacle avoidance and inspection tasks.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: First, this invention uses the predicted results of wind, current, waves, water level, and water depth as the constraint input and disturbance compensation input of the unmanned patrol boat control system, so that the output of the large ocean model directly serves the position, heading, speed, and attitude control, which helps to limit the technical focus to non-electric variable control or regulation systems.

[0012] Second, this invention introduces water motion vectors and uncertainty propagation processes into obstacle prediction, so that the obstacle avoidance trajectory no longer depends solely on the current geometric distance, but can reserve a safe distance based on the possible range of the obstacle's drift with the current, thereby improving the predictability of dynamic obstacle avoidance.

[0013] Third, the present invention adopts a multi-input multi-output cooperative control method to simultaneously solve the rudder angle and main engine speed commands, and incorporates the roll angle safety margin into the reference trajectory and control gain adjustment process, which can reduce the conflict between heading control, speed control and attitude safety, and reduce the risk of overshoot and roll amplification under complex sea conditions. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall closed-loop control architecture of the autonomous navigation system for nearshore unmanned patrol vessels based on a large ocean model, as proposed in this invention. Figure 2 This is a schematic diagram of the collaborative computing architecture and data flow between the shipborne edge computing platform and the shore-based computing platform in this invention; Figure 3 This is a comparative data analysis diagram of the dynamic obstacle state transition probability distribution and the reference obstacle avoidance trajectory in this invention; Figure 4 This is a data analysis diagram of the predicted flow field vorticity, effective wave height, and reference allowable roll angle in this invention; Figure 5 This is a data analysis diagram of heading error, speed error and roll angle response under cooperative control in this invention; Figure 6 This is a comparison chart of the trajectory error and control output before and after the disturbance compensation is enabled in this invention. Detailed Implementation

[0015] The embodiments of the present invention will be described below with reference to the accompanying drawings. These embodiments are used to explain the technical solutions of the present invention and do not constitute a limitation on the scope of protection. Without departing from the control logic of the present invention, the sensor model, grid resolution, prediction time window, control cycle, and actuator interface can all be adapted according to the specifications of the unmanned inspection vessel.

[0016] like Figure 1 As shown, the system in this embodiment centers on the navigation control of the unmanned patrol vessel, and sequentially includes a situational awareness fusion unit, an environmental prediction and control constraint generation unit, a probabilistic safe trajectory generation unit, and a multivariable cooperative disturbance rejection control unit. The situational awareness fusion unit provides the current water area and hull status, the environmental prediction and control constraint generation unit outputs future disturbance constraints through a large ocean model, the probabilistic safe trajectory generation unit outputs a reference obstacle avoidance trajectory, and the multivariable cooperative disturbance rejection control unit ultimately drives the servo motors and main engine to perform control.

[0017] The situational awareness fusion unit comprises four processing stages: multi-source sensor access, time synchronization, spatial alignment, and situational field generation. The shipborne satellite positioning terminal provides the unmanned patrol vessel's position data; the inertial measurement unit provides roll, pitch, heading, and angular velocity data; marine radar or equivalent sensing equipment provides point clouds of surrounding targets; meteorological and hydrological sensors provide wind speed, wind direction, current velocity, current direction, water level, or wave data; and a multibeam echo sounder provides underwater topographic data for shallow water areas. While the sampling frequencies of each sensor may differ, they are all converted to a unified time reference before entering the fusion stage.

[0018] During spatial alignment, the system transforms radar point clouds, depth point clouds, wind vectors, and hull status to a unified coordinate reference based on the installation position and attitude parameters of each sensor relative to the hull reference coordinate system. Anomalies caused by sea clutter, occasional obstructions, or communication delays are eliminated through continuous frame consistency and measurement range constraints. The processed panoramic situational field includes hull motion status, target list, near-shore wind field, sea surface current field, underwater topography grid, and mission semantic region.

[0019] To illustrate the data input and verification methods of the situational awareness fusion unit, this implementation uses nearshore open observation and open mode data from April 4, 2026 to May 3, 2026 as an example. After resampling at a ten-minute control verification cycle, 4210 fusion records were generated. The uses of each data source in the system are shown in the table below.

[0020] Table 1. Validation Data Sources and Fusion Inputs Combination Figure 2 The environmental prediction and control constraint generation unit runs on an onboard edge computing platform. This unit does not require large-scale offline training on the ship's deck; instead, it uses a pre-trained, compressed, and differentially updated large-scale ocean model for inference. The model input consists of current and historical wind, current, wave, water level, and water depth topography data, while the output is predicted local flow field, water level field, and wave height field data for a predetermined future time window. To maintain physical plausibility, physical residual constraints such as mass conservation, boundary continuity, and shallow water topography constraints are introduced during model training and inference.

[0021] To reduce the accumulation of prediction errors, the environmental prediction and control constraint generation unit performs assimilation corrections upon the arrival of new measured flow velocity, direction, water level, or wave data. Assimilation corrections can employ variational assimilation, ensemble Kalman filtering, or other methods capable of fusing observed and predicted states. The corrected environmental prediction results serve two purposes: first, they provide the probabilistic safety trajectory generation unit with environmental inputs regarding obstacle drift and hull roll response; second, they provide the multivariable collaborative disturbance rejection control unit with a basis for disturbance compensation and control gain adjustment.

[0022] like Figure 3 As shown, the probabilistic safety trajectory generation unit first extracts the centroid position, geometric envelope, and motion vector of the dynamic obstacle from the panoramic situation field, and then extracts the predicted water flow velocity, direction, and turbulence intensity of the obstacle's location and its neighborhood from the environmental prediction data. The system decomposes the obstacle's motion into its own motion and the influence of water advection, and combines historical trajectories to establish an uncertain motion model including drift and diffusion, thus obtaining the probability distribution of the obstacle's future state transition.

[0023] During the reference trajectory generation process, the system simultaneously evaluates track deviation, roll exceedance, energy consumption, and collision risk within the candidate trajectory set. Track deviation reflects the degree of deviation of the local obstacle avoidance trajectory relative to the global inspection path; roll exceedance is estimated by the hull response proxy model based on changes in speed, heading, wave height, and current velocity at candidate waypoints; collision risk is determined based on the degree to which the candidate trajectory enters a high-probability obstacle distribution area. Through the above evaluations, the system generates a reference obstacle avoidance trajectory that satisfies the minimum turning radius, maximum rate of change of speed, maximum angular velocity, and safe collision avoidance distance.

[0024] In order to make Figure 3 and Figure 4 The data analysis has corresponding numerical sources. In this implementation, the environmental prediction accuracy, obstacle probability extrapolation, and candidate obstacle avoidance trajectory evaluation results are listed in Table 2. The data in the table are obtained from the above-mentioned fused data and probabilistic trajectory generation process, and are used to support the availability of the ocean large model output as trajectory constraints.

[0025] Table 2 Key Results of Environmental Prediction and Probabilistic Obstacle Avoidance like Figure 4 As shown, the environmental disturbance data analysis uses distance or time as the horizontal axis to display the correlation between predicted flow field vorticity, significant wave height, and reference allowable roll angle. When vorticity or wave height increases, the reference allowable roll angle is correspondingly tightened, and the trajectory generation unit tends to select a path with gentler turning and less lateral acceleration; when environmental disturbances weaken, the system allows the reference trajectory to return to the global inspection path more quickly.

[0026] After receiving the reference obstacle avoidance trajectory, the multivariable cooperative disturbance rejection control unit converts it into a synchronized sequence of reference position, reference heading, reference speed, and reference allowable roll angle. The control unit uses position deviation, heading deviation, speed deviation, roll angle deviation, and their rate of change as state variables, and simultaneously generates rudder angle commands and main engine speed commands to avoid mutual cancellation or over-response that would result from completely separating heading control and speed control.

[0027] The extended state observation subunit is used to estimate the total disturbance that is difficult to model directly. This total disturbance includes the yaw disturbance moment caused by the predicted flow field, the lateral disturbance caused by wind pressure changes, the roll disturbance caused by wave excitation, and unmodeled higher-order terms of the hull's hydrodynamics. Since the environmental prediction results already provide the trends of flow velocity, vorticity, and wave height changes in the near future, the control unit can use this information as a feedforward disturbance input, reducing the hysteresis that may occur when relying solely on feedback observations.

[0028] The adaptive gain adjustment subunit smoothly adjusts the control gain based on the predicted vorticity intensity, significant wave height, and roll safety margin. When the unmanned patrol vessel is about to enter waters with significant changes in current direction or increased wave height, the system improves the response capability of the course maintenance and roll suppression channels; when the water disturbance is small or the actuators are close to their operating limits, the system appropriately reduces the gain to reduce frequent servo motor movements and large fluctuations in main engine speed.

[0029] The control command synthesis subunit synthesizes the feedback control quantity and disturbance compensation quantity into rudder angle command and main engine speed command. The rudder angle command is sent to the servo system via the onboard control network, and the main engine speed command is sent to the propulsion system via the main engine electronic control interface. The two commands are updated within the same control cycle, enabling the unmanned inspection vessel to maintain its heading, speed, and roll attitude within a preset safe range while avoiding obstacles.

[0030] like Figure 5 As shown, the control performance data analysis displays the response curves of heading error, speed error, and roll angle over time, used to verify the convergence and safety of the cooperative control law under disturbance changes. Figure 6 As shown, the comparative data analysis of track holding error and control output changes before and after disturbance compensation is used to illustrate the improvement effect of environmental prediction results as disturbance feedforward input on track holding and actuator stability.

[0031] For the corresponding Figure 5 and Figure 6 In this implementation method, the control performance is analyzed by comparing the controller with the large ocean model disturbance feedforward enabled with the baseline controller without feedforward enabled. The evaluation indicators are shown in the table below.

[0032] Table 3 Comparison of Cooperative Disturbance Rejection Control Performance During engineering deployment, the onboard terminal performs sensor fusion, local environment prediction, trajectory generation, and control output to meet real-time navigation control requirements; the shore-based terminal is used to collect post-mission data, train or calibrate large-scale ocean models offline, and update the onboard model in the form of differential parameter packages or configuration files. Even if ship-shore communication is interrupted, the onboard terminal can still maintain closed-loop control by relying on the local model and real-time observation data.

[0033] Through the above implementation methods, this invention limits the environmental prediction output of the large-scale ocean model to auxiliary inputs for the position, heading, speed, and attitude control of the unmanned patrol vessel, rather than simple meteorological and hydrological forecasting or general artificial intelligence computing tasks. The final technical effect of the system is reflected in the coordinated control of rudder angle and main engine speed, track maintenance, dynamic obstacle avoidance, and roll safety constraints, and has a clear engineering implementation path and feasibility.

Claims

1. An autonomous navigation system for near-shore unmanned patrol vessels driven by a large ocean model, characterized in that, include: The situational awareness fusion unit is used to collect and fuse multimodal perception data in real time to generate a panoramic situational field with a unified spatiotemporal reference. The multimodal perception data includes at least shipborne satellite positioning data, ship attitude data, surrounding target point cloud data, real-time wind data, flow velocity and direction data, water level or wave data, and underwater topography data in shallow water areas. The environmental prediction and control constraint generation unit is used to construct a gridded dynamic environmental field corresponding to the current nearshore operating water area based on the wind conditions, current velocity and direction, water level or waves and underwater topography data in the panoramic situation field. It also generates predicted data of the current field, water level field and wave height field within a future preset time window through a large ocean model pre-trained with ocean observation data and with physical residual constraints and observation assimilation correction. The predicted data is used as the reference trajectory constraint and control disturbance feedforward input. The probabilistic safety trajectory generation unit is used to identify the geometric contour, centroid position and motion vector of dynamic obstacles based on the surrounding target point cloud data and the hull attitude data, extract the predicted water flow velocity and direction information at the obstacle position, associate the obstacle motion vector with the water motion vector to obtain the probability distribution of the obstacle's future state transition, and generate a reference obstacle avoidance trajectory under the conditions of satisfying the hull kinematic constraints, safe collision avoidance distance constraints and roll safety constraints. A multivariable cooperative disturbance rejection control unit is used to receive the reference obstacle avoidance trajectory, decompose it into reference position, reference heading, reference speed, and reference allowable roll angle, and use the real-time position deviation, heading deviation, speed deviation, roll angle deviation, and corresponding change rate of the hull as state variables, and rudder angle command and main engine speed command as control variables to construct a multi-input multi-output cooperative control law for position, heading, speed, and attitude. The cooperative control law generates environmental disturbance compensation quantities based on the predicted data of the flow field, water level field, and wave height field, and synchronously outputs rudder angle command and main engine speed command to the onboard servo system and the main engine electronic control system to control the unmanned inspection vessel to navigate along the reference obstacle avoidance trajectory and keep the roll angle within a preset safe range.

2. The autonomous navigation system for near-shore unmanned patrol vessels based on a large ocean model as described in claim 1, characterized in that, The situation awareness fusion unit includes a time synchronization subunit, a coordinate alignment subunit, and a situation field generation subunit. The time synchronization subunit is used to attach a unified clock tag to the data from each sensor. The coordinate alignment subunit is used to convert data from different sensor coordinate systems to the hull reference coordinate system or the geographic reference coordinate system. The situation field generation subunit is used to filter point cloud noise, abnormal measurement values, and delayed data, and output a panoramic situation field that includes the hull status, target list, near-shore wind field, sea surface current field, water depth topography, and mission semantic region.

3. The autonomous navigation system for near-shore unmanned patrol vessels based on a large ocean model as described in claim 1, characterized in that, When constructing a gridded dynamic environmental field, the environmental prediction and control constraint generation unit uses measured flow velocity, flow direction, water level, wave, and underwater topography data as initial or boundary conditions, and uses variational assimilation or ensemble filtering to correct the predicted state, so that the predicted flow field, water level field, and wave height field data are updated with new observation data. The large ocean model is a lightweight inference model deployed on a shipboard edge computing platform, and the shore-based computing platform is only used for non-real-time training, verification, or differential updates of model parameters.

4. The autonomous navigation system for near-shore unmanned patrol vessels based on a large ocean model as described in claim 1, characterized in that, When the probability distribution of the future state transition of the obstacle is obtained, the probabilistic safe trajectory generation unit uses an uncertain motion model containing drift and diffusion terms to describe the future position distribution of the obstacle after being affected by water advection, turbulence and its own motion, and uses steady-state flow velocity, flow direction change rate, turbulence intensity and the historical motion trajectory of the obstacle as input parameters of the uncertain motion model.

5. The autonomous navigation system for near-shore unmanned patrol vessels based on a large ocean model as described in claim 1 or 4, characterized in that, When generating a reference obstacle avoidance trajectory, the probabilistic safe trajectory generation unit uses the weighted result of track deviation cost, roll over-limit penalty, energy consumption cost, and collision risk cost as the optimization objective. The track deviation cost is determined based on the distance between the candidate trajectory point and the preset global inspection path. The roll over-limit penalty is determined based on the degree to which the simulated roll angle corresponding to the candidate trajectory exceeds the reference allowable roll angle. The collision risk cost is determined based on the probability that the candidate trajectory falls into the high-risk area in the state transition probability distribution.

6. The autonomous navigation system for near-shore unmanned patrol vessels based on a large ocean model as described in claim 1, characterized in that, The multivariable collaborative disturbance rejection control unit includes a reference trajectory decomposition subunit, an extended state observation subunit, an adaptive gain adjustment subunit, and a control command synthesis subunit. The extended state observation subunit estimates the predicted flow field disturbance, wind pressure change, wave excitation, and unmodeled hull hydrodynamic terms as the total disturbance. The adaptive gain adjustment subunit adjusts the control gain based on the predicted flow field vorticity intensity, wave height, and roll safety margin. The control command synthesis subunit synthesizes the feedback control quantity and disturbance compensation quantity into rudder angle command and main engine speed command to synchronously adjust the position, heading, speed, and roll attitude of the unmanned patrol boat.

Images