Self-adaptive sail glider course control method and system based on multi-modal perception

Abstract

The invention discloses a self-adaptive sail glider course control method and system based on multi-mode perception, and relates to the technical field of path planning. The control method comprises the following steps of S1, multi-modal sensor integration and data processing, S2, real-time environment adaptation, S3, intelligent decision support and path planning, S4, adaptive course control, S5, multi-agent cooperation and feedback learning, and S6, fault detection and self-healing mechanism. By integrating various sensors and adopting a data processing technology, the complex marine environment can be comprehensively sensed, key environment indexes can be obtained in real time, the reliability of obstacle detection is improved through the multi-mode sensing capacity, it is ensured that the glider can accurately recognize potential threats under the severe sea condition, and therefore efficient obstacle avoidance path planning is achieved; and in combination with real-time weather forecast and cloud picture information, a navigation strategy can be dynamically adjusted, a navigation path is optimized, and the glider is ensured to safely and efficiently complete tasks in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, specifically to an adaptive windsurf glider heading control method and system based on multimodal perception. Background Technology

[0002] A wind-powered glider is an underwater robot that combines sailboat and traditional underwater glider technologies. It uses wind energy as its propulsion and utilizes buoyancy changes to navigate underwater, thereby achieving long-duration, wide-area autonomous ocean observation.

[0003] However, during actual navigation, the complex and ever-changing sea conditions pose challenges to its safety. To ensure that the wind glider can complete its mission safely and efficiently, it is very important to achieve intelligent obstacle avoidance.

[0004] Existing obstacle avoidance methods for sail gliders mainly use sensors such as radar and sonar to detect obstacles and then plan obstacle avoidance paths based on the location information of the obstacles. However, in severe sea conditions, complex environmental factors such as waves and sea fog can affect the detection accuracy of sensors, reducing the reliability of obstacle recognition. Furthermore, the planning of obstacle avoidance paths often does not fully consider factors such as the flight distance and the real-time status of the glider itself, resulting in low obstacle avoidance efficiency. Therefore, the real-time performance and reliability of obstacle avoidance for sail gliders in the existing technology are poor. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an adaptive heading control method and system for sail gliders based on multimodal perception. This method solves the problems of low obstacle avoidance efficiency caused by the impact of complex environmental factors such as waves and sea fog on sensor detection accuracy in harsh sea conditions, which reduces the reliability of obstacle recognition. Furthermore, obstacle avoidance path planning often fails to fully consider factors such as the flight distance and the real-time status of the glider itself.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive windsurfing glider heading control method based on multimodal perception, comprising the following steps: S1: Multimodal sensor integration and data processing, integrating radar, sonar, vision sensors, meteorological sensors and hydrological sensors for comprehensive environmental perception, and processing sensor data to generate environmental information reports; S2: Real-time environmental adaptation. After obtaining accurate environmental information, it acquires the latest weather forecast data and cloud image information. By combining this forecast information with real-time sensor data, it assesses the impact of waves, sea fog and wind on the glider's flight and adjusts the flight strategy accordingly. S3: Intelligent Decision Support and Path Planning, which uses intelligent path planning algorithms to generate obstacle avoidance paths based on real-time environmental data, weather forecasts and cloud map information; S4: Adaptive heading control. After path planning is completed, the glider's heading and speed are adjusted in real time, and the PID controller is used to dynamically optimize the navigation strategy based on real-time feedback. S5: Multi-agent collaboration and feedback learning. During the flight, a communication protocol is established between multiple sail gliders. Deep reinforcement learning is used to optimize path planning and heading control strategies through mutual feedback information. S6: Fault detection and self-healing mechanism, which monitors the glider's status in real time. Once a potential fault is detected, the system will automatically adjust the navigation strategy and activate the backup sensor when the sensor is damaged.

[0007] Preferably, the processing of sensor data in step S1 includes the following steps: S11: Data cleaning, which cleans the collected sensor data using filtering algorithms to remove outliers and noise caused by environmental interference and sensor malfunctions. S12: Data calibration, calibrating the cleaned data against the standard reference data to eliminate the sensor's systematic errors, and adjusting the sensor output according to the known calibration relationship; S13: Data fusion. After calibration, the Kalman filter algorithm is used to integrate the information from various sensors to generate a unified representation of the environmental state. S14: Feature extraction, using signal processing techniques to extract the required key features from the fused data; S15: Data format conversion, converting the processed data into a unified format; S16: Generate an environmental information report. Based on the processed data, generate a detailed environmental information report.

[0008] Preferably, in step S14, the key features include wind speed, wind direction, wave height, wave frequency, visibility, water temperature, water salinity data, air pressure, temperature and obstacle characteristics. By calculating the dynamic change rate of the key parameters, the rapid changes in environmental conditions are assessed, and historical data is used for benchmark comparison to identify abnormal situations.

[0009] Preferably, step S2 includes the following steps: S21: Obtain weather forecast data. While acquiring real-time sensor data, the system will also obtain the latest weather forecast data through the network interface. S22: Obtain cloud image information. The system will obtain the latest cloud image information to understand the current cloud distribution and help predict upcoming weather changes. S23: Environmental impact assessment, which involves a comprehensive environmental impact assessment combining real-time sensor data, weather forecasts, and cloud image information; S24: Adjustment of navigation strategy. Based on the results of the environmental impact assessment, the navigation strategy of the glider is dynamically adjusted. S25: Dynamic feedback and monitoring. The system will monitor environmental changes and the glider's status in real time and continue to adjust the navigation strategy according to the changes. In this process, if new obstacles or changes in environmental conditions are detected, the system will respond in a timely manner and optimize the navigation strategy.

[0010] Preferably, step S3 includes the following steps: S31: Environmental data integration, which integrates the real-time environmental data, weather forecast and cloud image information obtained in steps S1 and S2 and aggregates them into a unified environmental data model; S32: Obstacle avoidance path generation. Based on the integrated environmental data, the Dijkstra algorithm is selected to generate obstacle avoidance paths. S33: Route optimization. After generating the initial route, the route details are adjusted based on factors such as travel time, energy consumption, and navigation safety. S34: Decision Support System Evaluation. After the path optimization is completed, the generated navigation path is submitted to the decision support system for evaluation. The decision support system analyzes the generated path based on the current environmental conditions, mission objectives, and risk assessment.

[0011] S35: Path Confirmation and Execution. After the path has been evaluated and confirmed by the decision support system, the generated navigation path is transmitted to the heading control system.

[0012] Preferably, the Dijkstra algorithm formula in step S32 is:

[0013] in, It is from the starting point to the node. The shortest distance, From node To the node edge weights, From the starting point to the node The current known shortest distance.

[0014] Preferably, step S4 includes the following steps: S41: Set the target heading and speed. Based on the obstacle avoidance path generated in step S3, the system will set a new target heading and speed. S42: Real-time feedback adjustment. During the glider's flight, the control system will continuously receive real-time feedback data, including the current heading, speed, and environmental changes. Based on this feedback information, the PID controller will dynamically adjust the corresponding control output. S43: Dynamic optimization of navigation strategy. If a significant change in environmental conditions is detected during navigation, the control system will automatically reassess the current course and speed and adjust the navigation strategy accordingly. S44: Control data recording and analysis. The system will record all data during the heading control process, including real-time heading, speed, PID controller output, and environmental conditions.

[0015] Preferably, step S5 includes the following steps: S51: Status information sharing. Based on the communication protocol, each glider will periodically share its current status information, including position, heading, speed, sensor data, and the latest environmental assessment results. S52: Collaborative Path Planning. Based on feedback learning, when they share environmental information and state data, they can jointly optimize the navigation path to avoid overlapping routes and potential collision risks. S53: Dynamic strategy adjustment. During flight, if a glider encounters an obstacle or a sudden change in weather, other gliders will quickly obtain this information and adjust their course and speed accordingly. S54: Learning Outcome Integration. This involves integrating the learning outcomes acquired by each glider during flight and updating and optimizing the decision-making model for all gliders based on the integrated information.

[0016] Preferably, step S6 includes the following steps: S61: Real-time status monitoring and fault diagnosis, real-time monitoring of various statuses of the glider, including sensor working status and power system operation; S62: Fault type identification and response. Once an anomaly is detected, the system will further analyze the fault type and formulate corresponding response measures. S63: Automatically adjust navigation strategy. After a fault is confirmed, the system will automatically adjust the navigation strategy, activate backup sensors to continue acquiring data, or adjust the course and speed to maintain safe navigation. S64: Fault logging and feedback learning. Record detailed information about fault occurrences and corresponding countermeasures. Optimize fault detection and self-healing mechanisms through analysis and learning.

[0017] The present invention also provides an adaptive windsurfing glider heading control system based on multimodal perception, comprising: The multimodal sensor integration and data processing module integrates radar, sonar, visual sensors, meteorological sensors and hydrological sensors to achieve comprehensive environmental perception, and processes the sensor data to generate environmental information reports. The real-time environment adaptation module, after obtaining accurate environmental information, acquires the latest weather forecast data and cloud image information. By combining this forecast information with real-time sensor data, it assesses the impact of waves, sea fog and wind on the glider's flight and adjusts the flight strategy accordingly. The intelligent decision support and path planning module generates obstacle avoidance paths using intelligent path planning algorithms based on real-time environmental data, weather forecasts, and cloud map information. Adaptive heading control adjusts the glider's heading and speed in real time after path planning is completed, and uses a PID controller to dynamically optimize the navigation strategy based on real-time feedback. The multi-agent collaboration and feedback learning module establishes a communication protocol between multiple sail gliders during the flight, and optimizes path planning and heading control strategies by using deep reinforcement learning through mutual feedback information. The fault detection and self-healing mechanism module monitors the glider's status in real time. Once a potential fault is detected, the system will automatically adjust the navigation strategy and activate the backup sensor when the sensor is damaged.

[0018] This invention discloses an adaptive heading control method and system for a wind-powered glider based on multimodal perception, which has the following beneficial effects: (1) By integrating multiple sensors and using data processing technology, this invention can comprehensively perceive the complex marine environment and obtain key environmental indicators in real time, such as wind speed, wave height, and visibility. The multimodal perception capability greatly improves the reliability of obstacle detection, ensuring that the glider can accurately identify potential threats in harsh sea conditions, thereby achieving efficient obstacle avoidance path planning. At the same time, combined with real-time weather forecasts and cloud map information, it can dynamically adjust the navigation strategy and optimize the navigation path, ensuring that the glider can safely and efficiently complete its mission in a complex environment. (2) This invention improves the glider’s flight performance in changing environments through adaptive heading control and multi-agent cooperation mechanism. The real-time feedback adjustment capability of the PID controller enables the heading and speed to respond quickly to environmental changes, thereby improving the flexibility and safety of flight. In addition, the communication and feedback learning mechanism between multiple gliders allows them to share state information and environmental assessment results during flight, jointly optimize path planning, and avoid overlapping routes and potential collision risks. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0020] Figure 1 This is a flowchart of the adaptive windsurf glider heading control method based on multimodal perception according to the present invention; Detailed Implementation To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The applicant discovered: Achieving highly reliable autonomous obstacle avoidance and heading control for sail gliders in highly dynamic, uncertain, and harsh marine environments is currently the core technological challenge in the field of sail glider heading control. Existing technologies mainly rely on single or limited types of sensors (such as radar or sonar) for obstacle detection and employ static path planning algorithms. This results in poor performance of the system under complex sea conditions (such as large waves, sea fog, and sudden weather changes). The fundamental problem does not stem solely from insufficient sensor accuracy or low algorithm efficiency, but rather from three deeper technical bottlenecks: the limitations of physical perception, the static nature of control strategies, and the looseness of multi-agent cooperation.

[0022] This application provides an adaptive heading control method for sail-gliders based on multimodal perception. It addresses the problems of complex environmental factors, such as waves and fog, affecting sensor detection accuracy and reducing obstacle identification reliability in harsh sea conditions. Furthermore, obstacle avoidance path planning often fails to adequately consider factors like flight distance and the glider's real-time status, leading to low obstacle avoidance efficiency. This method integrates multiple sensors and employs data processing technology to comprehensively perceive the complex marine environment and acquire key environmental indicators such as wind speed, wave height, and visibility in real time. The multimodal perception capability significantly improves obstacle detection reliability, ensuring accurate identification of potential threats in harsh sea conditions and enabling efficient obstacle avoidance path planning. Simultaneously, by combining real-time weather forecasts and cloud imagery, the method dynamically adjusts navigation strategies and optimizes flight paths, ensuring the glider completes its mission safely and efficiently in complex environments.

[0023] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0024] This invention discloses an adaptive heading control method for a windsurfing glider based on multimodal perception.

[0025] According to the appendix Figure 1 As shown, it includes the following steps: S1: Multimodal sensor integration and data processing, integrating radar, sonar, vision sensors, meteorological sensors and hydrological sensors for comprehensive environmental perception, and processing sensor data to generate environmental information reports; S2: Real-time environmental adaptation. After obtaining accurate environmental information, the system acquires the latest weather forecast data and cloud image information. By combining this forecast information with real-time sensor data, the system assesses the impact of waves, sea fog, and wind on the glider's flight and adjusts the flight strategy accordingly. S3: Intelligent Decision Support and Path Planning, which uses intelligent path planning algorithms to generate obstacle avoidance paths based on real-time environmental data, weather forecasts and cloud map information; S4: Adaptive heading control. After path planning is completed, the glider's heading and speed are adjusted in real time, and the PID controller is used to dynamically optimize the navigation strategy based on real-time feedback. S5: Multi-agent collaboration and feedback learning. During the flight, a communication protocol is established between multiple sail gliders. Deep reinforcement learning is used to optimize path planning and heading control strategies through mutual feedback information. S6: Fault detection and self-healing mechanism, which monitors the glider's status in real time. Once a potential fault is detected, the system will automatically adjust the navigation strategy and activate the backup sensor when the sensor is damaged.

[0026] Step S1 involves processing the sensor data, including the following steps: S11: Data cleaning, which cleans the collected sensor data using filtering algorithms to remove outliers and noise caused by environmental interference and sensor malfunctions. S12: Data calibration, calibrating the cleaned data against the standard reference data to eliminate the sensor's systematic errors, and adjusting the sensor output according to the known calibration relationship; S13: Data fusion. After calibration, the Kalman filter algorithm is used to integrate the information from various sensors to generate a unified representation of the environmental state. S14: Feature extraction, using signal processing techniques to extract the required key features from the fused data; S15: Data format conversion, converting the processed data into a unified format; S16: Generate an environmental information report. Based on the processed data, generate a detailed environmental information report.

[0027] Step S1 effectively removes outliers caused by environmental interference and sensor malfunctions through data cleaning, ensuring data quality. Data calibration eliminates systematic errors and enhances measurement accuracy. Data fusion uses the Kalman filter algorithm to integrate information from multiple sensors into a unified environmental state representation, improving the comprehensiveness of environmental perception. Feature extraction ensures the effectiveness of key indicators, providing an important basis for subsequent decision-making. Unified data format conversion facilitates subsequent processing and analysis. The final generated environmental information report provides detailed and reliable information support for glider navigation strategy adjustment and obstacle avoidance path planning. The above steps together improve the adaptability and navigation safety of the sail glider in complex marine environments.

[0028] In step S14, key features include wind speed, wind direction, wave height, wave frequency, visibility, water temperature, water salinity data, air pressure, temperature, and obstacle characteristics. By calculating the dynamic change rate of these key parameters, rapid changes in environmental conditions are assessed, and historical data is used for benchmark comparison to identify anomalies. Real-time monitoring of these key features and their dynamic change rates allows for timely identification of rapid changes in environmental conditions, helping the system respond quickly to potential risks or obstacles, thereby improving the glider's navigation safety. Secondly, by combining historical data for benchmark comparison, the system can effectively identify anomalies, such as sudden weather changes or unexpected obstacles, providing an important basis for adjusting navigation strategies. Furthermore, the comprehensive analysis of these key features supports intelligent decision-making, enabling the glider to perform more precise path planning and obstacle avoidance in complex environments, thereby improving overall navigation efficiency and mission success rate.

[0029] Step S2 includes the following steps: S21: Obtain weather forecast data. While acquiring real-time sensor data, the system will also obtain the latest weather forecast data through the network interface. S22: Obtain cloud image information. The system will obtain the latest cloud image information to understand the current cloud distribution and help predict upcoming weather changes. S23: Environmental impact assessment, which involves a comprehensive environmental impact assessment combining real-time sensor data, weather forecasts, and cloud image information; S24: Adjustment of navigation strategy. Based on the results of the environmental impact assessment, the navigation strategy of the glider is dynamically adjusted. S25: Dynamic feedback and monitoring. The system will monitor environmental changes and the glider's status in real time and continue to adjust the navigation strategy according to the changes. In this process, if new obstacles or changes in environmental conditions are detected, the system will respond in a timely manner and optimize the navigation strategy.

[0030] Real-time access to weather forecasts and cloud imagery enables gliders to stay informed about weather changes and predict environmental factors that may affect flight, thereby reducing the risks caused by sudden weather changes. Secondly, the comprehensive environmental impact assessment combines sensor data with meteorological information, providing a more comprehensive understanding of the current environmental conditions and a scientific basis for adjusting flight strategies. This ensures that gliders can select the optimal route in complex sea conditions. The dynamic feedback and monitoring functions enable the system to respond to environmental changes and glider status in real time, ensuring that flight strategies can be quickly adjusted when encountering new obstacles or sudden weather conditions, thereby improving flight safety and efficiency and reducing the possibility of accidents.

[0031] Step S3 includes the following steps: S31: Environmental data integration, which integrates the real-time environmental data, weather forecasts and cloud image information obtained in steps S1 and S2 and converges them into a unified environmental data model, providing a comprehensive and accurate basis for subsequent decision-making; S32: Obstacle avoidance path generation. Based on the integrated environmental data, the Dijkstra algorithm is selected to generate obstacle avoidance paths. S33: Path optimization. After generating the initial path, the path details are adjusted based on factors such as travel time, energy consumption, and navigation safety. The obstacle avoidance path generated by Dijkstra's algorithm ensures effective avoidance of potential obstacles in complex environments. At the same time, the path optimization process comprehensively considers factors such as travel time, energy consumption, and safety, making the selected path both efficient and economical. S34: Decision Support System Evaluation. After the route optimization is completed, the generated navigation route is submitted to the decision support system for evaluation. The decision support system analyzes the generated route based on the current environmental conditions, mission objectives, and risk assessment. The evaluation of the decision support system further verifies the reliability of the route and ensures that the generated navigation route meets the current environmental conditions and risk assessment.

[0032] S35: Path Confirmation and Execution. After the path has been evaluated and confirmed by the decision support system, the generated navigation path is transmitted to the heading control system. The path confirmation and execution step ensures that the generated path is quickly transmitted to the heading control system after rigorous evaluation, achieving efficient connection from decision-making to execution.

[0033] The formula for Dijkstra's algorithm in step S32 is:

[0034] in, It is from the starting point to the node. The shortest distance, From node To the node edge weights, From the starting point to the node The current known shortest distance.

[0035] Dijkstra's algorithm effectively finds the optimal path from start to finish, which is crucial for obstacle avoidance in complex environments. By progressively expanding the known shortest paths, the algorithm ensures that each step selects the lowest-cost option among currently reachable paths, thus guaranteeing the optimality of the final path. Dijkstra's algorithm is highly scalable and flexible, allowing adjustments based on different weighting factors (such as flight time, energy consumption, and safety) to optimize path selection according to specific mission requirements. This flexibility ensures that gliders can not only find effective paths in complex marine environments but also achieve an optimal balance between safety and efficiency.

[0036] Step S4 includes the following steps: S41: Set the target heading and speed. Based on the obstacle avoidance path generated in step S3, the system will set a new target heading and speed. S42: Real-time feedback adjustment. During the glider's flight, the control system will continuously receive real-time feedback data, including the current heading, speed, and environmental changes. Based on this feedback information, the PID controller will dynamically adjust the corresponding control output. S43: Dynamic optimization of navigation strategy. If a significant change in environmental conditions is detected during navigation, the control system will automatically reassess the current course and speed and adjust the navigation strategy accordingly. S44: Control data recording and analysis. The system will record all data during the heading control process, including real-time heading, speed, PID controller output, and environmental conditions.

[0037] Step S4, by setting the target heading and speed, enables the system to formulate a clear navigation plan based on the obstacle avoidance path, ensuring efficient navigation. The real-time feedback adjustment function allows the PID controller to dynamically optimize its output according to changes in heading, speed, and environment, maintaining stable navigation and responding promptly to emergencies. Dynamic optimization of the navigation strategy automatically adjusts the heading and speed when environmental conditions change significantly, ensuring optimal navigation conditions. Finally, the recording and analysis of control data provides important information for subsequent tasks, promotes the optimization of future navigation algorithms, and comprehensively improves the glider's autonomous decision-making ability and navigation safety.

[0038] Step S5 includes the following steps: S51: Status information sharing. Based on the communication protocol, each glider will periodically share its current status information, including position, heading, speed, sensor data, and the latest environmental assessment results. S52: Collaborative Path Planning. Based on feedback learning, when they share environmental information and state data, they can jointly optimize the navigation path to avoid overlapping routes and potential collision risks. S53: Dynamic strategy adjustment. During flight, if a glider encounters an obstacle or a sudden change in weather, other gliders will quickly obtain this information and adjust their course and speed accordingly. S54: Learning Outcome Integration. This involves integrating the learning outcomes acquired by each glider during flight and updating and optimizing the decision-making model for all gliders based on the integrated information.

[0039] Step S5 significantly enhances the collaborative capabilities and navigation safety among gliders through status information sharing, cooperative path planning, dynamic strategy adjustment, and integration of learning outcomes. Regularly sharing status information allows gliders to understand each other's dynamics in real time, optimizing flight paths to avoid overlapping routes and collision risks. Simultaneously, dynamic strategy adjustment ensures that all gliders can respond quickly to obstacles or sudden weather events, adjusting their course and speed to enhance overall navigation safety.

[0040] Step S6 includes the following steps: S61: Real-time status monitoring and fault diagnosis, real-time monitoring of various statuses of the glider, including sensor working status and power system operation; S62: Fault type identification and response. Once an anomaly is detected, the system will further analyze the fault type and formulate corresponding response measures. S63: Automatically adjust navigation strategy. After a fault is confirmed, the system will automatically adjust the navigation strategy, activate backup sensors to continue acquiring data, or adjust the course and speed to maintain safe navigation. S64: Fault logging and feedback learning. Record detailed information about fault occurrences and corresponding countermeasures. Optimize fault detection and self-healing mechanisms through analysis and learning.

[0041] Step S6 focuses on real-time status monitoring and fault diagnosis, ensuring that the glider can promptly detect and handle faults during operation. Through fault type identification and response, the system can quickly analyze anomalies and formulate corresponding measures, automatically adjusting navigation strategies to maintain safe flight. Simultaneously, fault recording and feedback learning provide crucial data support for future fault detection and self-healing mechanisms, thereby optimizing the system's reliability and intelligence.

[0042] As one implementation method, sensor data acquisition in harsh environments suffers from significant uncertainties and errors, particularly in highly nonlinear and non-Gaussian marine environments. For example, visual sensors fail in sea fog, sonar experiences multipath interference under complex hydrological conditions, and meteorological sensor forecasts suffer from latency and accuracy issues. This uncertainty at the perception level directly leads to distortion in environmental modeling, thereby affecting the reliability of path planning. A more fundamental problem is that existing methods lack a perception reconstruction mechanism capable of dynamically adapting to sensor failure or data degradation.

[0043] To address this perception bottleneck, this embodiment introduces a multimodal adaptive sensing framework based on quantum sensing enhancement. Quantum sensors (such as quantum radar and quantum gyroscopes) possess ultra-high precision and anti-interference capabilities, providing more stable environmental data even in harsh sea conditions. Simultaneously, by introducing a hybrid model combining variational autoencoders (VAEs) and Bayesian neural networks (BNNs), probabilistic fusion and uncertainty quantification of multimodal sensor data are achieved. Specifically, VAEs can be used to learn the underlying distribution of sensor data and generate robust feature representations, while BNNs provide the confidence interval for prediction results. When a sensor data point becomes abnormal, the system can automatically weight or switch data sources based on uncertainty indicators, thereby achieving dynamic reconstruction of the perception layer. This enhancement scheme not only improves data reliability but also provides a probabilistic environmental representation for subsequent path planning and control, working in conjunction with the Kalman filtering and feature extraction steps of the original scheme to form a more robust perception-decision closed loop.

[0044] Furthermore, regarding the mismatch between the static nature of path planning and the dynamic nature of the environment, without online learning or real-time replanning capabilities, the glider may frequently recalculate its path during flight, increasing computational load and energy consumption, and even causing control lag.

[0045] Therefore, this embodiment also includes a hybrid real-time path planner based on deep reinforcement learning (DRL) and model predictive control (MPC). The DRL agent (such as using PPO or SAC algorithms) learns dynamic obstacle avoidance strategies through interaction with the environment, and its reward function comprehensively considers travel time, energy consumption, and safety factors. Simultaneously, the MPC module continuously optimizes short-term paths based on the current environmental state and predictive models (such as weather forecasts and hydrodynamic models) to ensure real-time performance. This scheme collaborates with the original Dijkstra algorithm in the following way: Dijkstra is used to generate a global reference path, while the DRL-MPC hybrid module is responsible for local real-time adjustments. Furthermore, spatiotemporal semantic mapping technology is introduced to associate environmental information (such as wave height and wind speed) with spatiotemporal attributes, enabling path planning to consider not only geometric obstacle avoidance but also environmental dynamic constraints. This significantly improves the smoothness and executability of the path and reduces frequent controller adjustments.

[0046] In some embodiments, state sharing and deep reinforcement learning mechanisms among multiple gliders have been proposed. However, in real-world marine environments, communication may be limited by distance, bandwidth, and interference, leading to delays in collaborative decision-making. Furthermore, deep reinforcement learning requires a large amount of interactive data, and the sample efficiency and policy consistency among distributed agents are difficult to guarantee. Therefore, the communication latency and learning efficiency issues in multi-agent collaboration have long plagued the field.

[0047] Therefore, this embodiment proposes a multi-agent collaborative architecture based on federated learning and edge computing. Each glider, acting as an edge node, locally trains the DRL model and uploads the model parameters (rather than the raw data) to the central server via encrypted communication for federated aggregation, generating a global model which is then distributed to each node. This not only reduces communication load but also protects data privacy. Simultaneously, a variant of the Asynchronous Advantage Actor-Commentator (A3C) algorithm is introduced to support asynchronous policy updates, mitigating policy lag caused by communication latency. This architecture collaborates with the original multi-agent communication protocol and feedback learning module, improving learning efficiency through federated learning and reducing the central server load through edge computing. Ultimately, this system achieves more efficient collaborative path planning and dynamic obstacle avoidance, demonstrating its advantages, especially in large-scale cluster tasks.

[0048] The present invention also provides an adaptive windsurfing glider heading control system based on multimodal perception, comprising: The multimodal sensor integration and data processing module integrates radar, sonar, visual sensors, meteorological sensors and hydrological sensors to achieve comprehensive environmental perception, and processes the sensor data to generate environmental information reports. The real-time environment adaptation module, after obtaining accurate environmental information, acquires the latest weather forecast data and cloud image information. By combining this forecast information with real-time sensor data, it assesses the impact of waves, sea fog and wind on the glider's flight and adjusts the flight strategy accordingly. The intelligent decision support and path planning module generates obstacle avoidance paths using intelligent path planning algorithms based on real-time environmental data, weather forecasts, and cloud map information. Adaptive heading control adjusts the glider's heading and speed in real time after path planning is completed, and uses a PID controller to dynamically optimize the navigation strategy based on real-time feedback. The multi-agent collaboration and feedback learning module establishes a communication protocol between multiple sail gliders during the flight, and optimizes path planning and heading control strategies by using deep reinforcement learning through mutual feedback information. The fault detection and self-healing mechanism module monitors the glider's status in real time. Once a potential fault is detected, the system will automatically adjust the navigation strategy and activate the backup sensor when the sensor is damaged.

[0049] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive heading control method for a sail-glide aircraft based on multimodal sensing, characterized in that, Includes the following steps: S1: Multimodal sensor integration and data processing, integrating radar, sonar, vision sensors, meteorological sensors and hydrological sensors for comprehensive environmental perception, and processing sensor data to generate environmental information reports; S2: Real-time environmental adaptation. After obtaining accurate environmental information, the system acquires the latest weather forecast data and cloud image information. By combining this forecast information with real-time sensor data, the system assesses the impact of waves, sea fog, and wind on the glider's flight and adjusts the flight strategy accordingly. S3: Intelligent Decision Support and Path Planning, which uses intelligent path planning algorithms to generate obstacle avoidance paths based on real-time environmental data, weather forecasts and cloud map information; S4: Adaptive heading control. After path planning is completed, the glider's heading and speed are adjusted in real time, and the PID controller is used to dynamically optimize the navigation strategy based on real-time feedback. S5: Multi-agent collaboration and feedback learning. During the flight, a communication protocol is established between multiple sail gliders. Deep reinforcement learning is used to optimize path planning and heading control strategies through mutual feedback information. S6: Fault detection and self-healing mechanism, which monitors the glider's status in real time. Once a potential fault is detected, the system will automatically adjust the navigation strategy and activate the backup sensor when the sensor is damaged.

2. The adaptive wind-powered glider heading control method based on multimodal perception according to claim 1, characterized in that, The processing of sensor data in step S1 includes the following steps: S11: Data cleaning, which cleans the collected sensor data using filtering algorithms to remove outliers and noise caused by environmental interference and sensor malfunctions. S12: Data calibration, calibrating the cleaned data against the standard reference data to eliminate the sensor's systematic errors, and adjusting the sensor output according to the known calibration relationship; S13: Data fusion. After calibration, the information from each sensor is integrated using the Kalman filter algorithm to generate a unified representation of the environmental state. S14: Feature extraction, using signal processing techniques to extract the required key features from the fused data; S15: Data format conversion, converting the processed data into a unified format; S16: Generate an environmental information report. Based on the processed data, generate a detailed environmental information report.

3. The adaptive sail glider heading control method based on multimodal perception according to claim 2, characterized in that, In step S14, key features include wind speed, wind direction, wave height, wave frequency, visibility, water temperature, water salinity data, air pressure, temperature and obstacle characteristics. By calculating the dynamic change rate of key parameters, rapid changes in environmental conditions are assessed, and historical data is used for benchmark comparison to identify abnormal situations.

4. The adaptive wind-powered glider heading control method based on multimodal perception according to claim 1, characterized in that, Step S2 includes the following steps: S21: Obtain weather forecast data. While acquiring real-time sensor data, the system will also obtain the latest weather forecast data through the network interface. S22: Obtain cloud image information. The system will obtain the latest cloud image information to understand the current cloud distribution and help predict upcoming weather changes. S23: Environmental impact assessment, which involves a comprehensive environmental impact assessment combining real-time sensor data, weather forecasts, and cloud image information; S24: Adjustment of navigation strategy. Based on the results of the environmental impact assessment, the navigation strategy of the glider is dynamically adjusted. S25: Dynamic feedback and monitoring. The system will monitor environmental changes and the glider's status in real time and continue to adjust the navigation strategy according to the changes. In this process, if new obstacles or changes in environmental conditions are detected, the system will respond in a timely manner and optimize the navigation strategy.

5. The adaptive wind-powered glider heading control method based on multimodal perception according to claim 1, characterized in that, Step S3 includes the following steps: S31: Environmental data integration, which integrates the real-time environmental data, weather forecast and cloud image information obtained in steps S1 and S2 and aggregates them into a unified environmental data model; S32: Obstacle avoidance path generation. Based on the integrated environmental data, the Dijkstra algorithm is selected to generate obstacle avoidance paths. S33: Route optimization. After generating the initial route, the route details are adjusted based on factors such as travel time, energy consumption, and navigation safety. S34: Decision Support System Evaluation. After the path optimization is completed, the generated navigation path is submitted to the decision support system for evaluation. The decision support system analyzes the generated path based on the current environmental conditions, mission objectives, and risk assessment. S35: Path Confirmation and Execution. After the path has been evaluated and confirmed by the decision support system, the generated navigation path is transmitted to the heading control system.

6. The adaptive wind-powered glider heading control method based on multimodal perception according to claim 5, characterized in that, The Dijkstra algorithm formula in step S32 is: in, It is from the starting point to the node. The shortest distance, From node To the node edge weights, From the starting point to the node The current known shortest distance.

7. The adaptive wind-powered glider heading control method based on multimodal perception according to claim 1, characterized in that, Step S4 includes the following steps: S41: Set the target heading and speed. Based on the obstacle avoidance path generated in step S3, the system will set a new target heading and speed. S42: Real-time feedback adjustment. During the glider's flight, the control system will continuously receive real-time feedback data, including the current heading, speed, and environmental changes. Based on this feedback information, the PID controller will dynamically adjust the corresponding control output. S43: Dynamic optimization of navigation strategy. If a significant change in environmental conditions is detected during navigation, the control system will automatically reassess the current course and speed and adjust the navigation strategy accordingly. S44: Control data recording and analysis. The system will record all data during the heading control process, including real-time heading, speed, PID controller output, and environmental conditions.

8. The adaptive wind-powered glider heading control method based on multimodal perception according to claim 1, characterized in that, Step S5 includes the following steps: S51: Status information sharing. Based on the communication protocol, each glider will periodically share its current status information, including position, heading, speed, sensor data, and the latest environmental assessment results. S52: Collaborative Path Planning. Based on feedback learning, when they share environmental information and state data, they can jointly optimize the navigation path to avoid overlapping routes and potential collision risks. S53: Dynamic strategy adjustment. During flight, if a glider encounters an obstacle or a sudden change in weather, other gliders will quickly obtain this information and adjust their course and speed accordingly. S54: Learning Outcome Integration. This involves integrating the learning outcomes acquired by each glider during flight and updating and optimizing the decision-making model for all gliders based on the integrated information.

9. The adaptive wind-powered glider heading control method based on multimodal perception according to claim 1, characterized in that, Step S6 includes the following steps: S61: Real-time status monitoring and fault diagnosis, real-time monitoring of various statuses of the glider, including sensor working status and power system operation; S62: Fault type identification and response. Once an anomaly is detected, the system will further analyze the fault type and formulate corresponding response measures. S63: Automatically adjust navigation strategy. After a fault is confirmed, the system will automatically adjust the navigation strategy, activate backup sensors to continue acquiring data, or adjust the course and speed to maintain safe navigation. S64: Fault logging and feedback learning. Record detailed information about fault occurrences and corresponding countermeasures. Optimize fault detection and self-healing mechanisms through analysis and learning.

10. A multimodal sensing-based adaptive windsurfing glider heading control system, applied to the multimodal sensing-based adaptive windsurfing glider heading control method as described in any one of claims 1-9, characterized in that, include: The multimodal sensor integration and data processing module integrates radar, sonar, visual sensors, meteorological sensors and hydrological sensors to achieve comprehensive environmental perception, and processes the sensor data to generate environmental information reports. The real-time environment adaptation module, after obtaining accurate environmental information, acquires the latest weather forecast data and cloud image information. By combining this forecast information with real-time sensor data, it assesses the impact of waves, sea fog and wind on the glider's flight and adjusts the flight strategy accordingly. The intelligent decision support and path planning module generates obstacle avoidance paths using intelligent path planning algorithms based on real-time environmental data, weather forecasts, and cloud map information. Adaptive heading control adjusts the glider's heading and speed in real time after path planning is completed, and uses a PID controller to dynamically optimize the navigation strategy based on real-time feedback. The multi-agent collaboration and feedback learning module establishes a communication protocol between multiple sail gliders during the flight, and optimizes path planning and heading control strategies by using deep reinforcement learning through mutual feedback information. The fault detection and self-healing mechanism module monitors the glider's status in real time. Once a potential fault is detected, the system will automatically adjust the navigation strategy and activate the backup sensor when the sensor is damaged.

Images