A multi-source perception fusion obstacle avoidance path planning method for motor-driven unmanned ship

Abstract

The application discloses a multi-source perception fusion obstacle avoidance path planning method for motor-driven unmanned ships, which comprises the following steps: obtaining multi-modal data related to the passage of the unmanned ship, performing fusion processing to form the fusion perception data of the unmanned ship, generating at least two candidate obstacle avoidance motion trajectory clusters of the unmanned ship, analyzing and generating corresponding passage influence data, predicting and obtaining the obstacle avoidance motion trajectory parameter set of the unmanned ship after screening according to the corresponding passage influence data and in combination with the travel intention data of the unmanned ship, determining the corresponding obstacle avoidance motion trajectory cluster, generating the path navigation path of the unmanned ship, and completing the real-time guidance obstacle avoidance instruction of the unmanned ship; by constructing the target optimization function of the unmanned ship and the fitness function, the optimal obstacle avoidance motion trajectory of the unmanned ship can be selected, and the intelligentization and obstacle avoidance efficiency of the unmanned ship operation are improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned vessel transportation technology, and in particular to a multi-source perception fusion obstacle avoidance path planning method for motor-driven unmanned vessels. Background Technology

[0002] With the development of intelligent transportation, unmanned ships are iterating towards autonomous navigation technology towards multimodal perception fusion, intelligent cross-domain decision-making, integrated water and land control, and highly reliable system integration. Complex cross-domain environment perception, water and land mode switching control, multi-constraint path planning, and perception-decision-control link collaboration have become the core links to ensure the safety, traffic efficiency, and operational continuity of ships in cross-domain navigation.

[0003] Currently, when unmanned ships are navigating, control still mainly relies on manual driving combined with traditional rule-based auxiliary control. Although this can meet basic driving needs, it lacks deep fusion of cross-domain multimodal data, adaptive perception of complex waterway environments, and processing of traffic impact characteristics. In terms of economic energy consumption, operational risk, and path complexity, there is no complete target optimization scheme for unmanned ships, nor a corresponding unmanned ship control strategy. Summary of the Invention

[0004] In view of the problems existing in the above-mentioned multi-source perception fusion obstacle avoidance path planning method of motor-driven unmanned ships, this invention is proposed.

[0005] Therefore, one of the objectives of this invention is to provide a multi-source perception fusion obstacle avoidance path planning method for motor-driven unmanned surface vessels (USVs). This method generates traffic impact data by processing the multimodal perception data of the USV, constructs the target optimization function and fitness function of the USV, and can select the optimal obstacle avoidance trajectory of the USV. At the same time, it generates corresponding USV body and power control feedback strategy schemes through machine autonomous learning regression models, further improving the intelligence and obstacle avoidance efficiency of USV operation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The present invention provides a multi-source perception fusion obstacle avoidance path planning method for motor-driven unmanned surface vessels, which acquires multimodal data related to the passage of the unmanned surface vessel, and the multimodal data is collected based on the perception devices installed on the unmanned surface vessel; The multimodal data is fused with multimodal and multi-source sensing data to form fused sensing data for the unmanned vessel. In response to the fused perception data of the unmanned vessel, at least two candidate obstacle avoidance trajectory clusters of the unmanned vessel are generated, and based on each candidate obstacle avoidance trajectory cluster, corresponding traffic impact data are generated through analysis. Based on the relevant traffic impact data and the unmanned vessel's (UV) travel intention data, the parameter prediction of the UV after screening is obtained through the parameter prediction of the machine autonomous learning regression model. The corresponding obstacle avoidance trajectory cluster is determined, and the path navigation path of the UV is generated. At the same time, the data of the UV body and power control feedback strategy is generated. The traffic impact data comprehensively considers the motor energy consumption, path complexity and traffic risk of the UV. The objective optimization function and fitness function of the UV are constructed based on the traffic impact data. Among the candidate obstacle avoidance trajectory clusters, the minimum value of the fitness function is selected to obtain the parameter set of the UV obstacle avoidance trajectory. The corresponding UV body and power control feedback strategy scheme is generated through the machine autonomous learning regression model. In response to the path navigation path of the target unmanned vessel and the feedback strategy data of the unmanned vessel's hull and power control, the data is received in real time by the sensing devices on the unmanned vessel, and then output to form the adjustment and control data of the unmanned vessel, thus completing the real-time guidance and obstacle avoidance command of the unmanned vessel.

[0007] As a preferred embodiment of the present invention, the multimodal data is subjected to multimodal and multi-source sensing data fusion processing to form fused sensing data of the unmanned vessel. The multimodal data includes environmental perception multimodal data, navigation and pose multimodal data, hull and power control multimodal data, mission payload multimodal data, and communication and collaboration multimodal data.

[0008] In a preferred embodiment of the present invention, a target optimization function for the unmanned vessel and a fitness function are constructed based on the traffic impact data, wherein the planned path of the unmanned vessel is assumed to be... It consists of an ordered number of waypoints, represented as ,in For the first The location and heading of each waypoint are calculated using three types of indicators as follows: Motor energy consumption Represented as: ;in, The energy consumption for a single flight segment is calculated using the following formula: ; In the above formula, The basic energy consumption coefficient per unit distance is obtained from the calibration of the unmanned vessel's power system. For the first The sailing distance of the segment, The flow resistance energy consumption coefficient is... For the first The relative speed difference between the ocean current and the course reflects the gain or loss of energy consumption due to flow resistance. Additional energy consumption coefficient for steering, For the first Change in heading angle for a flight segment; Path complexity Represented as: ; in, This is the total path length. The straight-line distance from the starting point to the ending point. The cumulative turning angle for the entire path is obtained by summing the absolute values ​​of the turning angles for each segment. The effective number of turns is obtained by comparing and analyzing the number of turns with an angle greater than a preset threshold. To preset the maximum number of turns, These are the weighting coefficients for the three sub-dimensions: path detour rate, cumulative turning angle, and number of effective turns. ; Traffic risk Represented as: ,in, The overall risk level for a single flight segment is calculated using the following formula: In the above formula, The static obstacle risk is inversely proportional to the obstacle distance and is calculated using a Gaussian decay model. To assess the collision risk of dynamic targets, the collision probability is calculated based on the nearest encounter distance and time. Hydrometeorological risks are assessed based on wave height, wind speed, and visibility. Risk assessment of waterway boundaries is generated by comparing and analyzing distances to the edge of navigable areas. These are the weighting coefficients for four sub-dimensions: static obstacles, dynamic vessels, hydrometeorology, and channel boundaries. .

[0009] As a preferred embodiment of the present invention, the motor energy consumption is... Path complexity and the degree of traffic risk Normalization is performed, and the values ​​are normalized to the interval [0, 1] using the extreme value method, as follows: , ,as well as , The optimized distance value for feasible trajectory clusters, To optimize the economic speed of ships, The objective optimization function for optimizing the safety coefficient is expressed as: ; The three objectives are aggregated into a single unmanned vessel comprehensive traffic cost function using weighting coefficients, as follows: ; in, These are the weighting coefficients for energy consumption, complexity, and risk, respectively. All values ​​are greater than 0, and the weights can be dynamically adjusted according to the task scenario. Constraints can be added, including motion constraints, safety constraints, battery life constraints, and boundary constraints. If the path violates the constraints, a penalty term is added to the objective function, transforming the constrained problem into an unconstrained optimization problem, as shown below: ; in, The penalty coefficient is for large values. To constrain the degree of violation.

[0010] In a preferred embodiment of the present invention, the fitness function is constructed using a Pareto-dominated fitness function, outputting a set of Pareto optimal paths for decision-making. Specifically, the individual waypoints of all paths within the candidate obstacle avoidance trajectory cluster are non-dominated and hierarchically stratified. Within the same level, the congestion distance is calculated to avoid the solution converging to a local region. Finally, the fitness calculation formula is used for analysis, as follows: ; in, For fitness value, This represents the total number of levels. The level at which the individual waypoints on the path are located. This represents the normalized crowding distance.

[0011] As a preferred embodiment of the present invention, when aggregating the comprehensive passage cost function of unmanned vessels, the method further includes dynamic adaptive adjustment of the weights of unmanned vessels. Specifically, by acquiring multimodal fusion perception data in real time, and through a closed-loop strategy logic of working condition identification, weight linkage correction, constraint verification, and function update, the weight coefficients of energy consumption, path complexity, and passage risk in the objective function are dynamically adjusted, and the system autonomously switches to balance according to the working conditions, adapting to the path planning requirements of the entire navigation scenario of unmanned vessels.

[0012] In a preferred embodiment of the present invention, the obstacle avoidance trajectory parameter set of the unmanned vessel is generated through a machine autonomous learning regression model to produce a corresponding unmanned vessel hull and power control feedback strategy scheme, specifically as follows: Feature extraction is performed on the obstacle avoidance trajectory parameter set of the unmanned vessel. The K-nearest neighbor algorithm is used to establish a traffic impact feature dataset for each feature point. At the same time, based on the obstacle avoidance trajectory parameter set of the unmanned vessel, the energy consumption, complexity, and risk features in the traffic impact feature dataset are analyzed by principal component analysis. In response to energy consumption, complexity, and risk characteristics, and combined with the direction and path trajectory turning angle of the unmanned vessel's travel intention data, the current curvature trajectory waypoint path of the unmanned vessel's obstacle avoidance trajectory is fitted by the moving least squares method. The Gaussian curvature and average curvature at each feature point of the unmanned vessel's obstacle avoidance trajectory are calculated, and the absolute value of the average curvature is used as the curvature feature value of that point. The system retrieves the current unmanned vessel and power control type and parameter information from the pre-set unmanned vessel and power control database. It then combines this information with the pre-set unmanned vessel and power control feedback strategy schemes to analyze the corresponding energy consumption, complexity, and risk characteristics, and determines different unmanned vessel and power control feedback strategy schemes.

[0013] As a preferred embodiment of the present invention, the unmanned hull and power control database is stored based on a three-level architecture of hull body, power system and control attributes, and directly calls the full parameters of the corresponding type as the input benchmark for subsequent correlation analysis. The full parameters include hull body parameters, power system parameters and control attribute parameters. The power control feedback strategy scheme includes an energy-saving cruise strategy, a safety and stability strategy, a maneuvering obstacle avoidance strategy, and a balanced adaptation strategy.

[0014] As a preferred embodiment of the present invention, it further includes displaying at least two sets of the obstacle avoidance trajectory clusters on the terminal control display device of the unmanned vessel, determining the obstacle avoidance trajectory clusters by controlling the terminal control display device of the unmanned vessel, and generating the path navigation path of the unmanned vessel.

[0015] In a preferred embodiment of the present invention, in response to obtaining a 360° surround view of the current position and time of the unmanned vessel, a mixed reality view is displayed in the graphical user interface of the terminal control display device of the unmanned vessel, the mixed reality view including a real unmanned vessel passage scene; Based on the current location and time of the unmanned vessel and its path navigation path, a path navigation guide line of the unmanned vessel is superimposed and displayed on the actual unmanned vessel passage screen. The path navigation guide line of the unmanned vessel points to the end of the path navigation path of the unmanned vessel's current location and time.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention generates traffic impact data by processing multimodal perception data of the unmanned surface vessel (USV), constructs the USV's objective optimization function and fitness function, and establishes a three-dimensional comprehensive evaluation model considering energy consumption, path complexity, and comfort. This model takes into account the USV's endurance, control difficulty, and navigation stability, enabling the selection of the optimal obstacle avoidance trajectory. Simultaneously, it generates corresponding USV hull and power control feedback strategies through a machine learning regression model, further improving the intelligence and obstacle avoidance efficiency of the USV's operation. Furthermore, the dual-mode approach of global automatic optimization and manual visual intervention, along with the overlay of navigation lines on a mixed reality scenario, enhances the intuitiveness of control, making operation more intuitive and intelligent. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the process by which the machine autonomously learns to generate an unmanned hull and power control feedback strategy scheme in the method of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0019] Reference Figures 1-2 This is one embodiment of the present invention, which provides a multi-source perception fusion obstacle avoidance path planning method for motor-driven unmanned surface vessels, such as... Figure 1 As shown, the details are as follows: Acquire multimodal data related to unmanned vessel traffic; the multimodal data is collected based on sensing devices installed on the unmanned vessel. Multimodal data is fused with multimodal and multi-source sensing data to form fused sensing data for unmanned vessels; In response to the fused perception data of the unmanned vessel, at least two candidate obstacle avoidance trajectory clusters of the unmanned vessel are generated, and based on each candidate obstacle avoidance trajectory cluster, corresponding traffic impact data are generated through analysis. Based on the relevant traffic impact data and the unmanned vessel's (UV) travel intention data, the parameter prediction of the UV's obstacle avoidance trajectory is obtained through machine learning regression model to determine the corresponding obstacle avoidance trajectory clusters, generate the UV's path navigation path, and generate UV body and power control feedback strategy data. The traffic impact data comprehensively considers the UV's motor energy consumption, path complexity, and traffic risk. Based on the traffic impact data, the UV's objective optimization function and fitness function are constructed. Among the candidate obstacle avoidance trajectory clusters, the minimum value of the fitness function is selected to obtain the filtered UV obstacle avoidance trajectory parameter set. The corresponding UV body and power control feedback strategy scheme is then generated through machine learning regression model. In response to the path navigation path and the feedback strategy data of the unmanned vessel's hull and power control, the data is received in real time by the sensing devices on the unmanned vessel, and then output to form the adjustment and control data of the unmanned vessel, thus completing the real-time guidance and obstacle avoidance command of the unmanned vessel.

[0020] In this preferred embodiment, multimodal data is fused from multiple sources to form fused perception data for the unmanned surface vessel (USV). This multimodal data includes environmental perception, navigation and pose, hull and power control, mission payload, and communication and collaboration data. The USV's multimodal data in this embodiment can be categorized by function into environmental perception, navigation and pose, hull and power control, mission payload, and communication and collaboration. Environmental perception data forms the basis of the USV's autonomous capabilities, enabling it to recognize its surroundings and autonomously avoid obstacles. It utilizes a multi-sensor complementary system adapted to all-weather, all-space scenarios. Visual image data sources include visible light high-definition cameras, infrared thermal imagers, and low-light night vision cameras. Data formats include video streams, single-frame images, and structured semantic features. Its core function is to identify surface targets (ships, navigation marks, personnel, reefs) and shore targets. Line detection, search and rescue target recognition, and scene semantic segmentation are the most intuitive environmental perception modalities. Radar detection data comes from navigation radar, millimeter-wave radar, and solid-state marine radar. The data includes target point and track data, range-azimuth echo maps, and point cloud data. Its core function is all-weather, long-range detection of water surface obstacles and ships, unaffected by rain, fog, or lighting conditions. It outputs the target's range, azimuth, and relative speed, making it a core data source for long-range collision avoidance. LiDAR point cloud data comes from shipborne 2D and 3D LiDAR. The data format is three-dimensional spatial point clouds and obstacle feature clusters. Its core function is... High-precision 3D environment modeling at close range is used for dock berthing, precise obstacle avoidance near the shore, and accurate ranging of small floating objects, supplementing the accuracy shortcomings of vision and radar. Underwater sonar data sources include forward-looking collision avoidance sonar, multibeam echo sonar, side-scan sonar, and shallow seismic profilers. Data formats include sonar echo images, depth point clouds, and seismic profile data. Its core function is to detect obstacles below the water surface such as reefs, shipwrecks, and underwater pipelines, and to measure water depth and seabed topography; it is the core perception mode of the unmanned surface vessel's underwater dimension. Meteorological and hydrological data sources include shipborne miniature weather stations, current meters, wave sensors, and temperature, salinity, and depth gauges. The data, in the form of time-series values ​​such as wind speed and direction, air temperature and pressure, wave height and direction, surface current speed and direction, and water temperature and salinity, serves as the core function of providing environmental parameters for route planning and motion control compensation, predicting the interference of wind and waves on the hull, and improving navigation stability in complex sea conditions. The data source for the ship identification signal is the AIS (Automatic Identification System) receiver and VHF maritime radio. The data format is a structured message containing the identification code, position, course, speed, and ship type of surrounding ships. Its core function is to identify the identity of cooperative targets at long distances, cooperate with radar to complete target matching and collision avoidance decisions, and comply with the requirements of maritime navigation rules.Navigation and pose multimodal data are used to determine the spatiotemporal position and motion state of unmanned surface vessels (USVs), serving as the baseline data for autonomous navigation. Multi-sensor fusion is commonly employed to improve reliability. Satellite positioning data originates from GNSS receivers and RTK differential positioning modules, and its data format includes latitude and longitude, geodetic height, ground velocity, and positioning accuracy factor. Its core function is to provide an absolute position reference. RTK mode can achieve centimeter-level high-precision positioning, meeting the needs of fine-grained operations such as berthing and surveying. Inertial motion data originates from IMU (Inertial Measurement Unit) and Inertial Navigation System (INS), and its data format includes three-axis acceleration, three-axis angular velocity, roll angle, pitch angle, and heading. The azimuth sensor's core function is to output high-frequency hull attitude and motion parameters, compensating for the low-frequency limitations and signal interruption issues of satellite positioning, and achieving continuous and stable navigation when combined with GNSS. Navigational auxiliary data sources include electronic compasses, Doppler logs, and barometric depth sounders. Data formats include true and magnetic headings, speeds over water and on land, and real-time depth. Its core function is to correct heading errors, provide independent speed references, and perform redundancy checks for the navigation system. Route and track data sources include the route planning system and the navigation log system. Data formats include preset waypoints, planned routes, historical tracks, and yaw distances. Its core function is to serve as a target benchmark for navigation control and record the entire voyage. Process data; multimodal data related to hull and power control reflects the operating status of the unmanned surface vessel's (USV) hardware systems, supporting closed-loop control, fault diagnosis, and health management. Power system data sources include the engine ECU, battery management system (BMS), and propeller controller. Data formats include engine speed, oil temperature and pressure, voltage and current, remaining battery power and fuel quantity, and thrust output. Its core function is to monitor the power system's operating status, providing a basis for speed control and energy consumption optimization; this is the core safety data for electric USVs. Actuator feedback data sources include servo motors, azimuth thrusters, and jet pump rudder angle sensors. Data formats include rudder angle, propeller steering angle, and propeller speed. Feedback values ​​serve a core function of achieving closed-loop control of navigation attitude and verifying the effectiveness of control commands. Hull safety status data originates from draft sensors, tilt sensors, and compartment leak detection units. Data includes draft depth, trim and heel angles, compartment liquid levels, and leak alarm signals. Its core function is to monitor hull safety status, provide early warnings of water ingress and capsizing risks, and ensure operational safety under extreme sea conditions. Equipment health and alarm data originate from the status monitoring units of each subsystem. Data includes equipment online status, fault codes, alarm levels, and cumulative runtime. Its core function is to support remote health management and fault prediction, and optimize the maintenance-free cycle of unmanned surface vessels.Multimodal data in mission payloads belong to the business function layer. They are customized according to the application scenarios of unmanned vessels and are the core carriers of business value. For marine surveying scenarios, they correspond to multibeam bathymetry point clouds, side-scan sonar images, shallow seismic profile data, and orthophoto data. For water environment monitoring scenarios, they correspond to time-series data of multiple water quality parameters (pH, dissolved oxygen, turbidity, ammonia nitrogen, heavy metals, etc.) and water sample collection records. For security and search and rescue scenarios, they correspond to photoelectric tracking videos, target locking coordinates, VHF voice data, and life-saving equipment deployment records. For scientific research sampling scenarios, they correspond to hydrological profile data, sediment sampling data, and biological sampling records. Multimodal data related to communication and collaboration supports data interaction between unmanned surface vessels (USVs) and shore-based and other unmanned platforms. It forms the foundation for remote control and swarm collaboration. Shore-based interaction data includes remote control commands, task scheduling messages, uplink monitoring audio and video streams, and downlink parameter configuration data. Transmission links include 4G and 5G, microwave, and satellite communication. Multi-platform collaboration data includes location sharing, task allocation, and perception fusion data with other USVs, drones, and unmanned underwater vehicles (UUVs), supporting swarm collaborative operations. Maritime regulatory interaction data includes navigation dynamic data reported to the maritime system and received channel control and weather warning information.

[0021] In this embodiment, it is important to emphasize that the objective optimization function and fitness function of the unmanned vessel are constructed based on the traffic impact data. Here, the planned path of the unmanned vessel is assumed to be... It consists of an ordered number of waypoints, represented as ,in For the first The location and heading of each waypoint are calculated using three types of indicators as follows: Motor energy consumption Represented as: ;in, The energy consumption for a single flight segment is calculated using the following formula: ; In the above formula, The basic energy consumption coefficient per unit distance is obtained from the calibration of the unmanned vessel's power system. For the first The sailing distance of the segment, The flow resistance energy consumption coefficient is... For the first The relative speed difference between the ocean current and the course reflects the gain or loss of energy consumption due to flow resistance. Additional energy consumption coefficient for steering, For the first Change in heading angle for a flight segment; Path complexity Represented as: ; in, This is the total path length. The straight-line distance from the starting point to the ending point. The cumulative turning angle for the entire path is obtained by summing the absolute values ​​of the turning angles for each segment. The effective number of turns is obtained by comparing and analyzing the number of turns with an angle greater than a preset threshold. To preset the maximum number of turns, These are the weighting coefficients for the three sub-dimensions: path detour rate, cumulative turning angle, and number of effective turns. ; Traffic risk Represented as: ,in, The overall risk level for a single flight segment is calculated using the following formula: In the above formula, The static obstacle risk is inversely proportional to the obstacle distance and is calculated using a Gaussian decay model. To assess the collision risk of dynamic targets, the collision probability is calculated based on the nearest encounter distance and time. Hydrometeorological risks are assessed based on wave height, wind speed, and visibility. Risk assessment of waterway boundaries is generated by comparing and analyzing distances to the edge of navigable areas. These are the weighting coefficients for four sub-dimensions: static obstacles, dynamic vessels, hydrometeorology, and channel boundaries. .

[0022] This embodiment further illustrates the motor energy consumption. Path complexity and the degree of traffic risk Normalization is performed, and the values ​​are normalized to the interval [0, 1] using the extreme value method, as follows: , ,as well as , The optimized distance value for feasible trajectory clusters, To optimize the economic speed of ships, The objective optimization function for optimizing the safety coefficient is expressed as: ; The three objectives are aggregated into a single unmanned vessel comprehensive traffic cost function using weighting coefficients, as follows: ; in, These are the weighting coefficients for energy consumption, complexity, and risk, respectively. All values ​​are greater than 0, and the weights can be dynamically adjusted according to the task scenario. Constraints can be added, including motion constraints, safety constraints, battery life constraints, and boundary constraints. If the path violates the constraints, a penalty term is added to the objective function, transforming the constrained problem into an unconstrained optimization problem, as shown below: ; in, The penalty coefficient is for large values. To constrain the degree of violation.

[0023] In this embodiment, the fitness function is constructed using a Pareto-dominated fitness function, outputting a set of Pareto optimal paths for decision-making. Specifically, the individual waypoints of all paths within the candidate obstacle avoidance trajectory cluster are non-dominated and hierarchically stratified. Within the same level, the congestion distance is calculated to avoid solutions converging to local regions. Finally, the fitness calculation formula is used for analysis, as follows: ; in, For fitness value, This represents the total number of levels. The level at which the individual waypoints on the path are located. This represents the normalized crowding distance.

[0024] In this preferred embodiment, when aggregating the comprehensive passage cost function of unmanned vessels, the weights of the unmanned vessels are also dynamically and adaptively adjusted. Specifically, by acquiring multimodal fusion perception data in real time, and through a closed-loop strategy logic of working condition identification, weight linkage correction, constraint verification, and function update, the weight coefficients of energy consumption, path complexity, and passage risk in the objective function are dynamically adjusted, and the balance is switched autonomously according to the working condition, which is adapted to the path planning requirements of the entire navigation scenario of unmanned vessels.

[0025] This embodiment further explains that the core driving factors for motor energy consumption weight are remaining power and fuel, remaining range, mission timeliness requirements, and ocean current resistance, with data sources including the Battery Management System (BMS), GNSS trajectory planning, and current sensors; the core driving factors for path complexity weight are channel curvature, steering constraints, and operating mode, with data sources including electronic navigation charts, ship maneuvering parameters, and mission instructions; the core driving factors for traffic risk weight are obstacle density, dynamic number of ships, channel width, sea state level, and visibility, with data sources including navigation radar, visual and AIS fusion target list, electronic navigation charts, and meteorological and hydrological sensors; each weight is set with upper and lower limits to ensure that all three dimensions are considered to a minimum under any operating condition, avoiding extreme optimization of a single dimension that could lead to other risks: the risk weight range is set to [0.25, 0.70]; the energy consumption weight range is set to [0.15, 0.60]; and the complexity weight range is set to [0.10, 0.40].

[0026] refer to Figure 2In this embodiment, for the machine to autonomously learn and generate unmanned vessel hull and power control feedback strategy schemes, preferably, the obstacle avoidance trajectory parameter set of the unmanned vessel is used to generate the corresponding unmanned vessel hull and power control feedback strategy scheme through a machine autonomous learning regression model, as follows: Feature extraction is performed on the obstacle avoidance trajectory parameter set of the unmanned vessel. The K-nearest neighbor algorithm is used to establish a traffic impact feature dataset for each feature point. At the same time, based on the obstacle avoidance trajectory parameter set of the unmanned vessel, the energy consumption, complexity, and risk features in the traffic impact feature dataset are analyzed by principal component analysis. In response to energy consumption, complexity, and risk characteristics, and combined with the direction and path trajectory turning angle of the unmanned vessel's travel intention data, the current curvature trajectory waypoint path of the unmanned vessel's obstacle avoidance trajectory is fitted by the moving least squares method. The Gaussian curvature and average curvature at each feature point of the unmanned vessel's obstacle avoidance trajectory are calculated, and the absolute value of the average curvature is used as the curvature feature value of that point. The system retrieves the current unmanned vessel and power control type and parameter information from the pre-set unmanned vessel and power control database. It then combines this information with the pre-set unmanned vessel and power control feedback strategy schemes to analyze the corresponding energy consumption, complexity, and risk characteristics, and determines different unmanned vessel and power control feedback strategy schemes.

[0027] To further illustrate this embodiment, the obstacle avoidance trajectory parameter set of the unmanned surface vessel (USV) can be generated through a machine autonomous learning regression model to produce a corresponding USV hull and power control feedback strategy scheme, and the closed-loop execution process is as follows: Parameter retrieval: Match the hull, power, and control parameters of the current unmanned vessel from the database and output a standardized feature vector; Initial matching: Based on the ship's dynamic type, the default primary strategy is matched from the solution library, and the baseline weights of the objective function are initialized; Real-time correlation analysis: Access real-time multimodal sensing data, calculate the measured values ​​of energy consumption, complexity, and risk under the current operating conditions, and compare them with the expected features of the strategy; Strategy correction: If the measured indicators deviate from the expectations, fine-tune the parameter thresholds of the feedback strategy (such as power limit and rudder angle limit) and adjust the weights of the objective function simultaneously; Strategy switching determination: When the operating conditions undergo a step change (such as entering a port from open water or a sudden rise in sea state), the main strategy switching is triggered to complete the synchronous update of power control parameters and objective function; Effect write-back: Write the actual running effects under different strategies back to the database to continuously optimize parameter mapping relationships and strategy adaptation accuracy.

[0028] In this preferred embodiment, the unmanned hull and power control database is stored based on a three-level architecture of hull body, power system, and control attributes. It directly calls the full parameters of the corresponding type as the input benchmark for subsequent correlation analysis. The full parameters include hull body parameters, power system parameters, and control attribute parameters. The dynamic control feedback strategy scheme includes an energy-saving cruise strategy, a safety and stability strategy, a maneuvering obstacle avoidance strategy, and a balanced adaptation strategy. Each type of feedback strategy modifies the power and control parameters to produce a targeted adjustment effect on three dimensions, with the adjustment range based on the ship's inherent parameters: Energy-saving strategy: Energy consumption decreases by 10%–20%, path complexity tolerance decreases by 15%, and passage risk increases slightly by 5%–10%; Safety strategy: Passage risk decreases by 20%–30%, energy consumption increases by 10%–15%, and path complexity tolerance decreases by 10%; Maneuvering strategy: Path complexity tolerance increases by more than 30%, energy consumption increases by 15%–25%, and dynamic collision avoidance risk decreases; Balanced strategy: All three dimensions maintain baseline values ​​without directional deviation.

[0029] Based on the above, this implementation generates traffic impact data by processing the multimodal perception data of the unmanned vessel, constructs the objective optimization function and fitness function of the unmanned vessel, and establishes a three-dimensional comprehensive evaluation model of energy consumption, path complexity and comfort, taking into account the unmanned vessel's endurance, control difficulty and navigation stability, and can select the optimal obstacle avoidance trajectory of the unmanned vessel.

[0030] This embodiment combines the above method with an example operation, providing two exemplary schemes. The first is a pure electric single-unit water-displacement type, employing an energy-saving cruise-oriented and maneuver obstacle avoidance-assisted strategy, as follows: Suitable for vessel power: small to medium tonnage pure electric single unmanned vessel, propulsion by fixed-pitch propeller, with limited range, mostly used for routine inspections in inland waterways and near shore. Core control logic: By default, the energy-saving cruise strategy is enabled, the economic speed is locked, the motor output curve is optimized, and ineffective steering is limited; when dense obstacles or narrow channels are detected, the system automatically switches to maneuver assist mode and briefly releases the power and rudder angle restrictions. Three key characteristics: low baseline energy consumption, moderate risk, and moderate tolerance for path complexity; energy consumption is prioritized in energy-saving mode, while complexity adaptability is temporarily increased in maneuver mode. The baseline weights for the objective function are: energy consumption, complexity, and risk, with weights of 0.45, 0.30, and 0.25, respectively. When the power is low, the weight of energy consumption is further increased to 0.55.

[0031] The second is a high-speed planing boat, with maneuverability obstacle avoidance adaptation and speed graded control strategies, as follows: Suitable for ship propulsion: high-speed planing unmanned surface vessel, water jet propulsion, high-power diesel and pure electric power, mostly used in emergency search and rescue and rapid response scenarios; Core control logic: The default is to enable the maneuverability adaptation strategy, which leverages the high maneuverability of water jet propulsion and supports high-frequency turning and emergency obstacle avoidance; at the same time, a speed classification mechanism is set up, which allows high-speed cruising in areas with sparse obstacles and low-speed maneuvering in areas with dense obstacles. Three key characteristics: extremely high tolerance for path complexity, high energy consumption at high speeds, and low risk during low-speed maneuvers; its core advantage is its ability to navigate complex scenarios. The baseline weights for the objective function are: energy consumption, complexity, and risk, with weights of 0.35, 0.35, and 0.30, respectively. For emergency missions, the weight of energy consumption is reduced to 0.20, while the weights of mobility and risk are increased.

[0032] For example, the core driving factors are graded and used as trigger conditions for weight adjustment. Typical scenario weight configuration examples are shown in the table below: This embodiment aims to achieve mixed reality visualization guidance during the operation of the unmanned surface vessel (USV): It acquires a real-time 360° surround view of the vessel and generates a mixed reality view on the terminal GUI interface; based on the current vessel position and the planned navigation path, navigation guidance lines are overlaid on the real-view image, pointing to the navigation endpoint, providing intuitive assistance to operators in monitoring and intervening in the USV's navigation. Specifically, at least two sets of obstacle avoidance trajectory clusters are displayed on the USV's terminal control display device. By controlling the USV's terminal control display device, the obstacle avoidance trajectory clusters are determined, and the USV's path navigation path is generated.

[0033] In this embodiment, in response to obtaining the 360° surround view of the current position and time of the unmanned vessel, a mixed reality view is displayed in the graphical user interface of the terminal control display device of the unmanned vessel. The mixed reality view includes the actual unmanned vessel passage scene. Based on the current location and time of the unmanned vessel and its path navigation path, a path navigation guide line of the unmanned vessel is superimposed and displayed on the actual unmanned vessel passage scene. The path navigation guide line of the unmanned vessel points to the end of the path navigation path of the unmanned vessel's current location and time.

[0034] Based on the above, this embodiment transforms abstract navigation path data into visual guidance elements that fit the real scene, significantly reducing the situational awareness load of remote operators and improving the intuitiveness and intervention efficiency of navigation monitoring. At the same time, the dual mode of global automatic optimization and manual visual intervention, which combines real-world scene overlay with navigation lines, enhances the intuitiveness of operation, making operation more intuitive and intelligent. It balances the efficiency of autonomous navigation of unmanned vessels with the reliability of human decision-making, effectively improving navigation safety in complex waters and obstacle-dense scenarios.

[0035] In summary, this invention generates traffic impact data by processing multimodal perception data of unmanned surface vessels (USVs), constructs a target optimization function and a fitness function for the USV, and establishes a three-dimensional comprehensive evaluation model that considers energy consumption, path complexity, and comfort. This model takes into account the USV's endurance, control difficulty, and navigation stability, enabling the selection of the optimal obstacle avoidance trajectory. Furthermore, it generates corresponding USV body and power control feedback strategies through a machine learning regression model, further improving the intelligence and obstacle avoidance efficiency of the USV. In addition, the dual-mode approach of global automatic optimization and manual visual intervention, along with the overlay of navigation lines on a mixed reality scenario, enhances the intuitiveness of control, making operation more intuitive and intelligent.

[0036] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0037] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM and RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-source perception fusion obstacle avoidance path planning method for an electric motor-driven unmanned surface vessel, characterized in that: include: Acquire multimodal data related to unmanned vessel traffic, the multimodal data being collected based on sensing devices installed on the unmanned vessel; The multimodal data is fused with multimodal and multi-source sensing data to form fused sensing data for the unmanned vessel. In response to the fused perception data of the unmanned vessel, at least two candidate obstacle avoidance trajectory clusters of the unmanned vessel are generated, and based on each candidate obstacle avoidance trajectory cluster, corresponding traffic impact data are generated through analysis. Based on the relevant traffic impact data and combined with the unmanned vessel's travel intention data, the parameter prediction of the machine autonomous learning regression model is used to obtain the set of obstacle avoidance trajectory parameters of the unmanned vessel after screening, determine the corresponding obstacle avoidance trajectory cluster, generate the path navigation path of the unmanned vessel, and generate unmanned vessel body and power control feedback strategy data at the same time. The traffic impact data mentioned above comprehensively considers the motor energy consumption, path complexity, and traffic risk of the unmanned vessel. Based on the traffic impact data, an objective optimization function and a fitness function of the unmanned vessel are constructed. Among the candidate obstacle avoidance trajectory clusters, the minimum value of the fitness function is selected to obtain the set of obstacle avoidance trajectory parameters of the unmanned vessel after screening. The corresponding unmanned vessel body and power control feedback strategy scheme is generated through a machine autonomous learning regression model. In response to the path navigation path of the target unmanned vessel and the feedback strategy data of the unmanned vessel's hull and power control, the data is received in real time by the sensing devices on the unmanned vessel, and then output to form the adjustment and control data of the unmanned vessel, thus completing the real-time guidance and obstacle avoidance command of the unmanned vessel.

2. The multi-source perception fusion obstacle avoidance path planning method for a motor-driven unmanned surface vessel as described in claim 1, characterized in that, The multimodal data is fused with multimodal and multi-source sensing data to form fused sensing data for the unmanned vessel. The multimodal data includes environmental perception multimodal data, navigation and pose multimodal data, hull and power control multimodal data, mission payload multimodal data, and communication and collaboration multimodal data.

3. The multi-source perception fusion obstacle avoidance path planning method for an electric motor-driven unmanned surface vessel as described in claim 1, characterized in that, Based on the traffic impact data, an objective optimization function and a fitness function are constructed for the unmanned surface vessel (USV), where the planned path of the USV is assumed to be... It consists of an ordered number of waypoints, represented as ,in For the first The location and heading of each waypoint are calculated using three types of indicators as follows: Motor energy consumption Represented as: ;in, The energy consumption for a single flight segment is calculated using the following formula: ; In the above formula, The basic energy consumption coefficient per unit distance is obtained from the calibration of the unmanned vessel's power system. For the first The sailing distance of the segment, The flow resistance energy consumption coefficient is... For the first The relative speed difference between the ocean current and the course reflects the gain or loss of energy consumption due to flow resistance. Additional energy consumption coefficient for steering, For the first Change in heading angle for a flight segment; Path complexity Represented as: ; in, This is the total path length. The straight-line distance from the starting point to the ending point. The cumulative turning angle for the entire path is obtained by summing the absolute values ​​of the turning angles for each segment. The effective number of turns is obtained by comparing and analyzing the number of turns with an angle greater than a preset threshold. To preset the maximum number of turns, These are the weighting coefficients for the three sub-dimensions: path detour rate, cumulative turning angle, and number of effective turns. ; Traffic risk Represented as: ,in, The overall risk level for a single flight segment is calculated using the following formula: In the above formula, The static obstacle risk is inversely proportional to the obstacle distance and is calculated using a Gaussian decay model. To assess the collision risk of dynamic targets, the collision probability is calculated based on the nearest encounter distance and time. Hydrometeorological risks are assessed based on wave height, wind speed, and visibility. Risk assessment of waterway boundaries is generated by comparing and analyzing distances to the edge of navigable areas. These are the weighting coefficients for four sub-dimensions: static obstacles, dynamic vessels, hydrometeorology, and channel boundaries. .

4. The multi-source perception fusion obstacle avoidance path planning method for an electric motor-driven unmanned surface vessel as described in claim 3, characterized in that, Motor energy consumption Path complexity and the degree of traffic risk Normalization is performed, and the values ​​are normalized to the interval [0, 1] using the extreme value method, as follows: , ,as well as , The optimized distance value for feasible trajectory clusters, To optimize the economic speed of ships, The objective optimization function for optimizing the safety coefficient is expressed as: ； The three objectives are aggregated into a single unmanned vessel comprehensive traffic cost function using weighting coefficients, as follows: ； in, These are the weighting coefficients for energy consumption, complexity, and risk, respectively. All values ​​are greater than 0, and the weights can be dynamically adjusted according to the task scenario. Constraints can be added, including motion constraints, safety constraints, battery life constraints, and boundary constraints. If the path violates the constraints, a penalty term is added to the objective function, transforming the constrained problem into an unconstrained optimization problem, as shown below: ； in, The penalty coefficient is for large values. To constrain the degree of violation.

5. The multi-source perception fusion obstacle avoidance path planning method for an electric motor-driven unmanned surface vessel as described in claim 4, characterized in that, The fitness function is constructed using a Pareto-dominated fitness function, outputting a set of Pareto optimal paths for decision-making. Specifically, all individual waypoints within the candidate obstacle avoidance trajectory cluster are non-dominated and hierarchically stratified. Within the same level, congestion distance is calculated to avoid solutions converging to local regions. Finally, the fitness calculation formula is analyzed, as follows: ； in, For fitness value, This represents the total number of levels. The level at which the individual waypoints on the path are located. This represents the normalized crowding distance.

6. The multi-source perception fusion obstacle avoidance path planning method for an electric motor-driven unmanned surface vessel as described in claim 5, characterized in that, When aggregating the comprehensive passage cost function of unmanned vessels, it also includes dynamic adaptive adjustment of the weights of unmanned vessels. Specifically, by acquiring multimodal fusion perception data in real time, and through a closed-loop strategy logic of working condition identification, weight linkage correction, constraint verification, and function update, the weight coefficients of energy consumption, path complexity, and passage risk in the objective function are dynamically adjusted. The system autonomously switches to balance according to the working conditions and adapts to the path planning requirements of all unmanned vessel navigation scenarios.

7. The multi-source perception fusion obstacle avoidance path planning method for a motor-driven unmanned surface vessel as described in claim 1, characterized in that... At, The obstacle avoidance trajectory parameter set of the unmanned vessel is used to generate corresponding unmanned vessel hull and power control feedback strategy schemes through a machine autonomous learning regression model, as follows: Feature extraction is performed on the obstacle avoidance trajectory parameter set of the unmanned vessel. The K-nearest neighbor algorithm is used to establish a traffic impact feature dataset for each feature point. At the same time, based on the obstacle avoidance trajectory parameter set of the unmanned vessel, the energy consumption, complexity, and risk features in the traffic impact feature dataset are analyzed by principal component analysis. In response to energy consumption, complexity, and risk characteristics, and combined with the direction and path trajectory turning angle of the unmanned vessel's travel intention data, the current curvature trajectory waypoint path of the unmanned vessel's obstacle avoidance trajectory is fitted by the moving least squares method. The Gaussian curvature and average curvature at each feature point of the unmanned vessel's obstacle avoidance trajectory are calculated, and the absolute value of the average curvature is used as the curvature feature value of that point. The system retrieves the current unmanned vessel and power control type and parameter information from the pre-set unmanned vessel and power control database. It then combines this information with the pre-set unmanned vessel and power control feedback strategy schemes to analyze the corresponding energy consumption, complexity, and risk characteristics, and determines different unmanned vessel and power control feedback strategy schemes.

8. The multi-source perception fusion obstacle avoidance path planning method for an electric motor-driven unmanned surface vessel as described in claim 7, characterized in that, The unmanned vessel and power control database is stored in a three-level architecture based on the vessel body, power system, and control attributes. It directly calls the full parameters of the corresponding type as the input benchmark for subsequent correlation analysis. The full parameters include vessel body parameters, power system parameters, and control attribute parameters. The power control feedback strategy scheme includes an energy-saving cruise strategy, a safety and stability strategy, a maneuvering obstacle avoidance strategy, and a balanced adaptation strategy.

9. A multi-source perception fusion obstacle avoidance path planning method for an electric motor-driven unmanned surface vessel as described in any one of claims 1-8, characterized in that, It also includes displaying at least two sets of obstacle avoidance trajectory clusters on the terminal control display device of the unmanned vessel, determining the obstacle avoidance trajectory clusters by controlling the terminal control display device of the unmanned vessel, and generating the path navigation path of the unmanned vessel.

10. The multi-source perception fusion obstacle avoidance path planning method for an electric motor-driven unmanned surface vessel as described in claim 9, characterized in that, In response to obtaining a 360° surround view of the current position and time of the unmanned vessel, a mixed reality view is displayed in the graphical user interface of the terminal control display device of the unmanned vessel. The mixed reality view includes a real unmanned vessel passage scene. Based on the current location and time of the unmanned vessel and its path navigation path, a path navigation guide line of the unmanned vessel is superimposed and displayed on the actual unmanned vessel passage screen. The path navigation guide line of the unmanned vessel points to the end of the path navigation path of the unmanned vessel's current location and time.

Images