A remote control method and system for an underwater rust removal robot

Abstract

This application relates to the field of underwater robot control technology, and in particular to a remote control method and system for an underwater rust removal robot. The method includes: the underwater rust removal robot receiving high-level task commands; the underwater rust removal robot collecting water environment parameters in real time and inputting them into a preset reinforcement learning decision model while moving towards the target work area, matching the optimal communication mode; the underwater rust removal robot dynamically correcting the confidence weights of each sensor in the extended Kalman filter using an evolutionary optimization algorithm, and outputting real-time global perception fusion data; the underwater rust removal robot generating a rust removal trajectory through a hybrid path algorithm combining global planning and local dynamic obstacle avoidance; and when the underwater rust removal robot identifies uncertain and complex working conditions, it sends an intervention request through the matched optimal communication mode, replans the work path, and continues to execute the rust removal operation. This application helps improve the safety and accuracy of remote operation of underwater rust removal robots.

Description

Technical Field

[0001] This application relates to the field of underwater robot control technology, and in particular to a remote control method and system for an underwater rust removal robot. Background Technology

[0002] Ships are constantly submerged in seawater, making their underwater hull structures prone to corrosion and marine organism buildup, necessitating regular rust removal and maintenance. Traditional underwater operations using divers are inefficient, costly, and pose significant safety risks, making underwater rust removal robots an increasingly important alternative to manual labor.

[0003] The existing remote control of underwater rust removal robots has the following technical problems: First, tethered remotely operated underwater vehicles (ROVs) are restricted by the umbilical cable, limiting their range of motion, while untethered autonomous underwater vehicles (AUVs), although free from cables, suffer from low bandwidth and high latency in underwater acoustic communication, making it difficult to balance transmission distance and real-time high-definition video transmission; Second, the complex underwater environment and the use of fixed weights for multi-sensor fusion cannot cope with positioning drift caused by partial sensor failures; Third, remote communication latency makes real-time control difficult and lacks local autonomous decision-making capabilities; Fourth, the system lacks redundancy design, resulting in a high risk of single-point failure.

[0004] Therefore, there is an urgent need for a remote control method and system for underwater rust removal robots that can solve problems such as cable constraints, insufficient communication bandwidth, inaccurate positioning in harsh environments, sensitivity to control delays, and poor system reliability, so as to improve the safety and accuracy of remote operation of underwater rust removal robots. Summary of the Invention

[0005] Therefore, it is necessary to provide a remote control method and system for underwater rust removal robots that can improve the safety and accuracy of remote operation of underwater rust removal robots, in order to address the above-mentioned technical problems.

[0006] In a first aspect, this application provides a remote control method for an underwater rust removal robot, the method comprising: The underwater rust removal robot receives high-level task instructions, including a three-dimensional model of the ship's hull and the work area, from the remote control terminal via a surface relay platform. As the underwater rust removal robot moves toward the target work area, it collects water environment parameters in real time and inputs them into a preset reinforcement learning decision model to dynamically match the optimal communication mode in the current water environment. The underwater rust removal robot uses real-time sensing data collected by the multi-source sensors on board to dynamically correct the confidence weights of each sensor in the extended Kalman filter using an evolutionary optimization algorithm, and outputs real-time global sensing fusion data. The underwater rust removal robot combines the high-level task instructions and the global perception fusion data to generate a rust removal trajectory through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, thereby completing the underwater rust removal operation. When the underwater rust removal robot identifies uncertain and complex working conditions based on the fusion data of the whole-domain perception, it sends an intervention request to the remote control terminal through the matched optimal communication mode. After receiving the high-level adjustment instruction generated by the remote control terminal based on virtual gestures, it replans the operation path and continues to perform the rust removal operation.

[0007] In one embodiment, the underwater rust removal robot, based on real-time sensing data collected by its onboard multi-source sensors, dynamically adjusts the confidence weights of each sensor in the extended Kalman filter using an evolutionary optimization algorithm, and outputs real-time global sensing fusion data, including: The underwater rust removal robot acquires real-time sensing data from the multi-source sensors mounted on the machine; The underwater rust removal robot extracts the measurement values ​​of each sensor at the current moment from the sensing data at a preset frequency, and performs time updates by extended Kalman filtering based on the robot state estimation and robot kinematic model at the previous moment to obtain the predicted measurement values ​​of each sensor at the current moment. The underwater rust removal robot calculates the residual between the measured value of each sensor at the current moment and the corresponding predicted measured value, calculates the data consistency index of each sensor based on the residual, and calculates the data volatility index based on the fluctuation range of the measured value of each sensor within a preset time window. The underwater rust removal robot dynamically corrects the confidence weights of each sensor in the extended Kalman filter based on the data consistency index and the data volatility index, combined with an evolutionary optimization algorithm, and outputs global perception fusion data in real time.

[0008] In one embodiment, the underwater rust removal robot dynamically adjusts the confidence weights of each sensor in the extended Kalman filter based on the data consistency index and the data volatility index, combined with an evolutionary optimization algorithm, and outputs real-time global perception fusion data, including: The underwater rust removal robot inputs the data consistency index and the data volatility index into the evolutionary optimization algorithm, which iteratively updates the observation variance and confidence weight of each sensor. The underwater rust removal robot inputs the updated observation variance and the confidence weight into the extended Kalman filter, performs measurement updates, weights and fuses the measurements of the multi-source sensors at the current moment, and outputs the real-time pose and surrounding environment perception information of the underwater rust removal robot as global perception fusion data.

[0009] In one embodiment, the underwater rust removal robot inputs the updated observation variance and the confidence weights into the extended Kalman filter, performs measurement updates, weights and fuses the measurements from the multi-source sensors at the current moment, and outputs the real-time pose and surrounding environment perception information of the underwater rust removal robot as global perception fusion data, including: The underwater rust removal robot inputs the updated observation variance and confidence weights of each sensor into the extended Kalman filter, which then performs the measurement update. The extended Kalman filter performs a weighted correction on the residuals between the measured values ​​and predicted measured values ​​of each sensor at the current time based on the confidence weights and observation variances of each sensor, thereby obtaining the posterior state estimate and the corresponding error covariance matrix at the current time, wherein the error covariance matrix characterizes the uncertainty of the posterior state estimate. The underwater rust removal robot extracts real-time pose data from the posterior state estimation, extracts surrounding environment perception data from the real-time data collected by the multi-source sensors, and integrates the real-time pose data and the surrounding environment perception data into global perception fusion data.

[0010] In one embodiment, the underwater rust removal robot combines the high-level task instructions and the global perception fusion data to generate a rust removal trajectory through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, thereby completing the underwater rust removal operation, including: The underwater rust removal robot extracts the work area information from the high-level task instructions and combines it with the real-time pose data in the global perception fusion data. Using the hull 3D model as geometric constraints, it uses an improved genetic algorithm to generate a global macro path covering the target rust removal area. As the underwater rust removal robot travels along the global macro path, it uses the surrounding environment perception data from the global perception fusion data as input and runs the artificial potential field method to calculate in real time the repulsive force generated by local obstacles and the attractive force generated by the target rust removal area. It generates a local obstacle avoidance path to dynamically correct the deviation of the global macro path in the local area, so that the underwater rust removal robot can complete the underwater rust removal operation while maintaining a constant optimal working distance between the rust removal nozzle and the hull wall.

[0011] In one embodiment, the step of combining the real-time pose data from the global perception fusion data, using the hull 3D model as geometric constraints, and employing an improved genetic algorithm to generate a global macropath covering the target rust removal area includes: The underwater rust removal robot extracts the current position from the real-time pose data of the global perception fusion data as the path start point, extracts the working area from the high-level task instructions as the path end point set, and uses the surface of the ship's three-dimensional model as the path search space. The underwater rust removal robot constructs a fitness function with the total path length, energy consumption, and rust removal coverage as optimization objectives. The surface geometric constraints of the three-dimensional model of the hull are introduced into the fitness function. The selection, crossover, and mutation operations of the improved genetic algorithm are used to iteratively optimize the algorithm. Finally, the candidate path with the best fitness is output as the global macro path covering the target rust removal area.

[0012] In one embodiment, as the underwater rust removal robot moves towards the target work area, it collects aquatic environmental parameters in real time and inputs them into a preset reinforcement learning decision model, dynamically matching the optimal communication mode in the current aquatic environment, including: The underwater rust removal robot collects water environment parameters in real time as it moves toward the target work area; The underwater rust removal robot inputs the water environment parameters into a preset reinforcement learning decision model, and the reinforcement learning decision model simultaneously obtains the communication quality index of the current link and the priority parameters of the current task. The reinforcement learning decision model takes the water environment parameters, the communication quality index, and the priority parameters as input states, calculates the cumulative expected value of each candidate communication mode, and outputs the optimal communication mode under the current water environment. The underwater rust removal robot switches communication links according to the optimal communication mode to achieve dynamic matching of communication modes.

[0013] In one embodiment, the reinforcement learning decision model takes the water environment parameters, the communication quality index, and the priority parameter as input states, calculates the cumulative expected value of each candidate communication mode, and outputs the optimal communication mode under the current water environment, including: The reinforcement learning decision model combines the water environment parameters, the communication quality indicators, and the priority parameters into a state vector. The reinforcement learning decision model calculates the cumulative expected value of each candidate communication mode based on the state vector through a preset reward function. The reward function uses the communication success rate as the main positive reward term and the mode switching frequency as the penalty term to suppress frequent switching of communication modes. The reinforcement learning decision model selects the candidate communication mode with the highest cumulative expected value as the optimal communication mode output in the current water environment.

[0014] Secondly, this application also provides a remote control system for an underwater rust removal robot, the system comprising an underwater rust removal robot, a surface relay platform, and a remote control terminal, wherein the underwater rust removal robot comprises: The task instruction receiving module is used to receive high-level task instructions, including a three-dimensional model of the ship hull and the work area, issued by the remote control terminal via the surface relay platform. The communication mode matching module is used to collect water environment parameters in real time and input them into a preset reinforcement learning decision model during the underwater rust removal robot's journey to the target work area, so as to dynamically match the optimal communication mode in the current water environment. The data output module is used to dynamically correct the confidence weights of each sensor in the extended Kalman filter based on the real-time sensing data collected by the multi-source sensors on the machine, combined with the evolutionary optimization algorithm, and output the full-domain sensing fusion data in real time. The rust removal trajectory generation module is used to combine the high-level task instructions and the global perception fusion data to generate a rust removal motion trajectory through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, so as to complete the underwater rust removal operation. The intervention request module is used to send an intervention request to the remote control terminal through the matched optimal communication mode when the uncertain complex working conditions are identified based on the full-domain perception fusion data. After receiving the high-level adjustment instruction generated by the remote control terminal based on virtual gestures, the module replans the operation path and continues to execute the rust removal operation.

[0015] In summary, this application includes the following beneficial technical effects: By receiving high-level task commands, including a 3D model of the hull and the work area, from a remote control terminal via a surface relay platform, the underwater rust removal robot eliminates the need for continuous transmission of high-bandwidth real-time video streams or low-level motion commands between the remote control terminal and the underwater robot. Instead, it only needs to transmit semantic-level high-level task descriptions, significantly reducing reliance on the bandwidth and real-time performance of the underwater acoustic communication link. As the underwater rust removal robot moves towards the target work area, it collects real-time water environment parameters and inputs them into a pre-set reinforcement learning decision model. This dynamically matches the optimal communication mode for the current water environment, enabling the robot to intelligently switch between underwater acoustic communication, optical communication, or backup communication methods based on changes in water turbidity, temperature, salinity, and other environmental factors. This ensures communication success rates while suppressing the communication overhead caused by frequent switching, thus maintaining a stable and reliable communication link in complex and changing underwater environments and improving the continuity and safety of remote operations. Through evolutionary optimization algorithms… The method dynamically corrects the confidence weights of each sensor in the extended Kalman filter, solving the problem of positioning drift caused by local sensor failure in harsh underwater environments caused by traditional fixed-weight fusion methods. This allows the robot to maintain positioning accuracy under complex working conditions, ensuring the precise maintenance of a constant optimal working distance between the rust removal nozzle and the hull wall, significantly improving the consistency and quality of rust removal operations. A hybrid path algorithm combining global planning and local dynamic obstacle avoidance generates the rust removal trajectory, guaranteeing both a globally optimal path covering a large rust removal area and real-time dynamic obstacle avoidance of local obstacles. This allows the robot to safely and efficiently complete the rust removal trajectory execution without relying on remote commands during autonomous operation. Through an interactive intervention mechanism, the method fully leverages human decision-making advantages in complex working conditions, forming a human-machine collaborative operation mode with local autonomous decision-making as the primary method and remote high-level intervention as a supplement, significantly improving the overall safety and operational accuracy of the underwater rust removal robot's remote operation. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a remote control method for an underwater rust removal robot in one embodiment; Figure 2 This is a flowchart illustrating the remote control method for an underwater rust removal robot in another embodiment; Figure 3 This is a structural block diagram of the remote control system of an underwater rust removal robot in one embodiment. Detailed Implementation

[0017] This invention provides a remote control method and system for an underwater rust removal robot.

[0018] The embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] In the description of the embodiments disclosed in this invention, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0020] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the remote control method for the underwater rust removal robot in this invention includes: The S100 underwater rust removal robot receives high-level task instructions, including a 3D model of the ship's hull and the work area, from a remote control terminal via a surface relay platform.

[0021] Specifically, the underwater rust removal robot receives high-level task instructions from the remote control terminal via a surface relay platform. These instructions include a 3D model of the hull and the work area. The surface relay platform acts as the communication hub between the underwater robot and the remote control terminal, handling communication protocol conversion, data relay, and link pre-detection. The remote control terminal (shore-based or cloud-based) only handles high-level, non-real-time decision-making and does not require real-time operation of the robot's actuators, thus avoiding the impact of remote communication delays on lower-level control. The 3D hull model provides prior geometric information about the work environment, while the work area defines the spatial scope of the rust removal task, providing target boundaries for subsequent path planning.

[0022] The S200 underwater rust removal robot collects water environment parameters in real time as it moves toward the target work area and inputs them into a preset reinforcement learning decision model to dynamically match the optimal communication mode in the current water environment.

[0023] Specifically, the underwater robot is equipped with water environment sensors that monitor environmental parameters such as turbidity, temperature, and depth in real time. These parameters directly affect the transmission quality of different communication methods. The reinforcement learning decision model takes environmental parameters, current communication quality indicators, and task priorities as input. Through a pre-trained decision network, it evaluates the cumulative expected value of each candidate communication mode and outputs the optimal communication mode that matches the current water environment. Using these techniques, the system can utilize high-bandwidth optical communication in clear water environments to ensure real-time transmission of high-definition video streams, and switch to underwater acoustic communication in turbid or long-distance conditions to ensure command delivery. This predictive intelligent switching of communication modes avoids operational interruptions and safety accidents caused by communication failures or lost commands.

[0024] The S300 underwater rust removal robot uses real-time sensing data collected by its onboard multi-source sensors and combines this data with an evolutionary optimization algorithm to dynamically adjust the confidence weights of each sensor in the extended Kalman filter, and outputs real-time global sensing fusion data.

[0025] Specifically, underwater robots are equipped with various navigation and perception sensors (such as velocimeters, attitude sensors, depth gauges, positioning systems, sonar, and underwater cameras), each with its own advantages and disadvantages in different underwater environments. Extended Kalman filtering (EKF), as the core framework for multi-source data fusion, fuses data from various sensors according to optimal estimation theory, outputting robot pose and surrounding environment information. However, the underwater environment is complex and variable; visual sensors are prone to failure due to water turbidity, and velocimeters are susceptible to inaccurate readings due to bubble interference. If fixed fusion weights are used, data from failed sensors will severely contaminate the fusion results. Therefore, an evolutionary optimization algorithm is introduced to monitor the consistency and volatility indices of each sensor's data in real time, dynamically adjusting the observation variance and confidence weights of each sensor in the EKF. This ensures that the system maintains high-precision positioning even when sensors experience partial failure, guaranteeing the accurate maintenance of a constant working distance between the rust removal nozzle and the hull surface.

[0026] The S400 underwater rust removal robot combines high-level task instructions and global perception fusion data to generate rust removal motion trajectories through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, thus completing underwater rust removal operations.

[0027] Specifically, the hybrid path algorithm consists of a global planning layer and a local planning layer. The global planning layer uses the work area in the high-level task instructions as constraints, the real-time pose in the global perception fusion data as the starting point, and the 3D model of the hull as the geometric boundary. It uses an improved genetic algorithm to generate a global macro-path covering the target rust removal area. This path satisfies both energy optimization and rust removal coverage. As the robot travels along the global macro-path, the local planning layer uses the surrounding environment information in the global perception fusion data as input. It runs an artificial potential field method to perceive local obstacles such as welds, weld scars, and sacrificial anodes in real time, dynamically adjusting the direction of travel to avoid obstacles, while maintaining a constant optimal rust removal distance between the robot and the hull wall. Through this hybrid strategy, the global path ensures the overall optimization of large-area operations, while local obstacle avoidance ensures real-time safety. The collaboration of the two allows the robot to autonomously complete the rust removal trajectory without relying on remote commands.

[0028] When the S500 underwater rust removal robot identifies uncertain and complex working conditions based on the fusion of global perception data, it sends an intervention request to the remote control terminal through the matched optimal communication mode. After receiving the high-level adjustment instructions generated by the remote control terminal based on virtual gestures, it replans the operation path and continues to perform the rust removal operation.

[0029] Specifically, during autonomous operation, the robot continuously monitors the working environment. When the fusion data of the entire perception domain shows complex conditions such as hull welds, highly corroded steel plates, or unknown underwater attachments, and the onboard AI system's confidence level in recognizing these conditions is below a preset threshold, the robot automatically reduces its operating speed and sends an intervention request signal to the remote control terminal through the current optimal communication mode. After receiving the intervention request through the digital twin somatosensory interaction system, the operator observes the digital twin scene through VR / AR somatosensory devices, uses virtual gestures to delineate target areas or avoidance positions in virtual space, generates high-level adjustment instructions, and issues them to the robot. Upon receiving the instructions, the robot uses the target points or avoidance areas contained in the instructions as new path constraints, calls the local path planning module to replan the work path, and then continues to perform rust removal operations according to the replanned path. Through this mechanism, the robot only requests remote assistance in a few complex situations where it cannot make decisions independently, and the remote terminal only provides high-level target guidance without real-time control of the underlying movement. This fundamentally isolates the interference of communication delays on operational stability, while ensuring operational safety and decision-making accuracy in complex environments.

[0030] In one embodiment, such as Figure 2 As shown, S300 includes: S310, an underwater rust removal robot, acquires real-time sensing data from the multi-source sensors on board. S320, the underwater rust removal robot extracts the measurement values ​​of each sensor at the current moment from the sensing data at a preset frequency, and performs time updates by extended Kalman filtering based on the robot state estimation and robot kinematic model at the previous moment to obtain the predicted measurement values ​​of each sensor at the current moment. The S330 underwater rust removal robot calculates the residual between the measured value of each sensor at the current moment and the corresponding predicted measured value, calculates the data consistency index of each sensor based on the residual, and calculates the data volatility index based on the fluctuation range of the measured value of each sensor within a preset time window. The S340 underwater rust removal robot dynamically adjusts the confidence weights of each sensor in the extended Kalman filter based on data consistency and data volatility indicators, combined with an evolutionary optimization algorithm, and outputs real-time global perception fusion data.

[0031] Specifically, firstly, the underwater rust removal robot acquires real-time sensing data from its onboard multi-source sensors, including a Doppler velocimeter, attitude reference system, depth gauge, ultra-short baseline positioning system, side-scan sonar, and underwater high-definition camera. These sensors provide information on the robot's motion state and surrounding environment from different dimensions, such as velocity, attitude, depth, position, obstacle distribution, and visual images. Then, the underwater rust removal robot extracts the current-time measurements from the aforementioned sensing data at a preset frequency. Based on the robot's state estimate from the previous moment and the robot's kinematic model, it performs an extended Kalman filter time update to obtain the predicted measurements of each sensor at the current moment. Specifically, the extended Kalman filter employs a recursive estimation architecture. Each filtering cycle is divided into two stages: time update and measurement update. In the time update stage, the system calculates the prior state estimate for the current moment based on the posterior state estimate of the previous moment (including multi-dimensional state variables such as the robot's three-dimensional position, attitude angle, linear velocity, and angular velocity) and the robot's kinematic model. Subsequently, the prior state estimate is mapped to the observation space of each sensor through the observation equation to obtain the predicted measurement value of each sensor at the current moment. This predicted measurement value is essentially the value that the system "predicts" based on the motion model, which should be measured by each sensor at the current moment, and serves as a reference benchmark for subsequent comparison with the actual sensor readings. Next, the underwater rust removal robot calculates the residual between the current measured value and the corresponding predicted measured value of each sensor. Based on the residual, it calculates the data consistency index of each sensor and the data volatility index based on the fluctuation range of the measured values ​​of each sensor within a preset time window. The consistency index reflects the degree of deviation between the sensor measured value and the system expectation, while the volatility index reflects the stability of the sensor measured value. Finally, based on the data consistency index and the data volatility index, the underwater rust removal robot dynamically corrects the confidence weight of each sensor in the extended Kalman filter using an evolutionary optimization algorithm, and outputs the global perception fusion data in real time. When the consistency index of a sensor decreases or the volatility index increases, the evolutionary optimization algorithm automatically increases the observation variance of that sensor and decreases its confidence weight, reducing its contribution to the fusion result. Conversely, it increases its weight, so that the extended Kalman filter can automatically reduce the impact of failed sensors when sensor performance degrades. It can maintain high-precision positioning even in complex underwater environments such as turbid water and bubble interference. The final output global perception fusion data includes the robot's real-time pose and surrounding environment perception information, providing an accurate state reference for subsequent path planning and autonomous operation.

[0032] In one embodiment, the underwater rust removal robot dynamically adjusts the confidence weights of each sensor in the extended Kalman filter based on data consistency and data volatility indices, combined with an evolutionary optimization algorithm, and outputs real-time global perception fusion data, including: The underwater rust removal robot inputs data consistency and data volatility indices into an evolutionary optimization algorithm, which iteratively updates the observation variance and confidence weights of each sensor. The underwater rust removal robot then inputs the updated observation variance and confidence weights into an extended Kalman filter to perform measurement updates. It performs weighted fusion of the measurements from multiple sensors at the current moment and outputs the real-time pose of the underwater rust removal robot and its surrounding environment perception information as global perception fusion data.

[0033] Specifically, the underwater rust removal robot inputs the calculated data consistency index and data volatility index into an evolutionary optimization algorithm. This algorithm uses the observation variance and confidence weight of each sensor as optimization variables, and minimizes the mean square error of the state estimation after extended Kalman filtering as the optimization objective. It searches for the optimal weight combination under the current operating conditions in each filtering cycle through a population-based iterative optimization approach. Specifically, when the consistency index of a sensor decreases or the volatility index increases, the algorithm automatically increases its observation variance and decreases its confidence weight to weaken the sensor's influence on the fusion result; conversely, it decreases the observation variance and increases the weight to allow the sensor to play a greater role in the fusion. Then, the underwater rust removal robot updates the evolutionary optimization algorithm... The observation variances and confidence weights of each sensor are input into the Extended Kalman Filter (EKF) to perform a measurement update phase. In this phase, the EKF performs weighted fusion of the current sensor measurements based on the updated confidence weights of each sensor. The higher the confidence weight of a sensor, the greater its weight in the fusion, and vice versa, thus obtaining the optimal state estimate after fusion at the current moment. Finally, the EKF outputs the real-time pose data of the underwater rust removal robot (including three-dimensional position coordinates and attitude angles) and surrounding environment perception information (including obstacle distribution and wall distance). The two together constitute global perception fusion data, providing an accurate and reliable state benchmark for the robot's subsequent path planning and autonomous operation.

[0034] In one embodiment, the underwater rust removal robot inputs the updated observation variance and confidence weights into an extended Kalman filter to perform measurement updates. It then performs weighted fusion of measurements from multiple sensors at the current moment, outputting the underwater rust removal robot's real-time pose and surrounding environment perception information. This output data, as global perception fusion data, includes: The underwater rust removal robot inputs the updated observation variances and confidence weights of each sensor into an extended Kalman filter, which then performs measurement updates. The extended Kalman filter, based on the confidence weights and observation variances of each sensor, performs weighted correction on the residuals between the current sensor measurements and the predicted measurements, obtaining the posterior state estimate and the corresponding error covariance matrix for the current moment. The error covariance matrix characterizes the uncertainty of the posterior state estimate. The underwater rust removal robot extracts real-time pose data from the posterior state estimate and extracts surrounding environment perception data from the real-time data acquired by the multi-source sensors. Finally, it integrates the real-time pose data and the surrounding environment perception data into a global perception fusion data set.

[0035] Specifically, the underwater rust removal robot inputs the updated observation variances and confidence weights of each sensor into the Extended Kalman Filter (EKF) to initiate the measurement update process. The observation variance characterizes the uncertainty of each sensor's current measurement values, while the confidence weights characterize the reliability priority of each sensor in the fusion process. Together, they constitute the core parameters for the EKF to perform weighted fusion. Then, the EKF calculates the Kalman gain of each sensor at the current moment based on its corresponding confidence weight and observation variance. This Kalman gain determines the degree to which the actual measurement values ​​of each sensor at the current moment correct the final fusion result. Sensors with higher confidence weights have larger Kalman gains, resulting in stronger correction effects on the fusion result, and vice versa. Next, the EKF uses the calculated Kalman gains to perform weighted correction on the residuals between the current sensor measurements and predicted measurements, using the updated observation variance as a quantitative constraint on the correction magnitude. Larger residuals and higher Kalman gains result in larger correction magnitudes. However, the larger the observation variance, the less likely the constraint correction magnitude will be to exceed a reasonable range. This yields the posterior state estimate and the corresponding error covariance matrix for the current moment. The posterior state estimate is the optimal state output after fusing the current measurements from all sensors, while the error covariance matrix quantifies the uncertainty of the posterior state estimate. The smaller the value, the more reliable the estimation result. This matrix will be stored and used for the time update recursion of the extended Kalman filter in the next moment. Finally, the underwater rust removal robot extracts position and attitude information from the posterior state estimate as real-time pose data, and extracts obstacle distribution and wall distance information from the real-time data collected by multi-source sensors as surrounding environment perception data. The real-time pose data and surrounding environment perception data are integrated into unified global perception fusion data. This global perception fusion data includes both the robot's own motion state and the spatial information of the external environment, providing a unified and accurate data benchmark for subsequent path planning, obstacle avoidance decisions, and rust removal operations. This ensures that the robot can autonomously and stably complete the rust removal task in a complex underwater environment.

[0036] In one embodiment, the underwater rust removal robot combines high-level task instructions and global perception fusion data to generate a rust removal trajectory through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, thereby completing the underwater rust removal operation, including: The underwater rust removal robot extracts work area information from high-level task instructions and combines it with real-time pose data from the global perception fusion data. Using the 3D model of the hull as geometric constraints, it uses an improved genetic algorithm to generate a global macro path covering the target rust removal area. As the underwater rust removal robot travels along the global macro path, it uses the surrounding environment perception data from the global perception fusion data as input and runs the artificial potential field method to calculate in real time the repulsive force generated by local obstacles and the attractive force generated by the target rust removal area. This generates a local obstacle avoidance path to dynamically correct the deviation of the global macro path in local areas, enabling the underwater rust removal robot to complete the underwater rust removal operation while maintaining a constant optimal working distance between the rust removal nozzle and the hull wall.

[0037] Specifically, firstly, the underwater rust removal robot extracts work area information from high-level task instructions, including the spatial boundary coordinates of the target rust removal area and rust removal process parameters, and obtains real-time pose data from global perception fusion data, including the robot's current 3D position coordinates and attitude angles. Then, the robot uses the surface of the hull 3D model as the path search space, discretizing the hull surface into walkable mesh nodes. Path points are constrained to the hull surface to ensure the robot always travels along the wall. Based on this, an improved genetic algorithm is used to generate a global macro-path covering the target rust removal area. This improved genetic algorithm constructs a fitness function with the total path length, energy consumption, and rust removal coverage as optimization objectives, and incorporates the surface geometric constraints of the hull 3D model as a penalty function into the fitness function. These geometric constraints include surface normal information and curvature information. Through iterative optimization via selection, crossover, and mutation operations, the algorithm finally outputs a global macro-path that starts from the current position, traverses the target rust removal area, and has the optimal overall cost. This global macro-path serves as the macroscopic motion benchmark for the robot's movement from the current position to the target rust removal area, ensuring the overall optimality of large-area operations. Subsequently, as the underwater rust removal robot travels along the aforementioned global macropath towards the target rust removal area, it uses real-time ambient environment perception data from the global perception fusion data as input to perform local dynamic obstacle avoidance using the artificial potential field method. The ambient environment perception data includes obstacle positions, wall distances, and local structural information such as welds, detected in real-time by side-scan sonar and underwater high-definition cameras. The basic mechanism of the artificial potential field method is to treat the target rust removal area as a virtual gravitational source, generating an attractive force towards the robot, while simultaneously treating the real-time detected local obstacles as virtual repulsive sources, generating a repulsive force away from the obstacles. The direction of the vector resultant of these two forces is the robot's optimal direction of travel at the current moment. Through this mechanism, when encountering local obstacles, the robot can deviate from the global macropath in real-time to avoid them, and after overcoming the obstacle, return to the macropath to continue traveling, while always maintaining a constant optimal working distance between the rust removal nozzle and the hull wall, typically 8 to 12 centimeters. Ultimately, the global macropath ensures the overall coverage and energy efficiency of large-scale rust removal operations, while the artificial potential field law responds in real time to local environmental changes to ensure safe movement. Together, they form a complete rust removal trajectory, enabling the underwater rust removal robot to autonomously, safely, and efficiently complete underwater rust removal operations from its current location to the target rust removal area without any remote real-time control.

[0038] In one embodiment, combining real-time pose data from global perception fusion data, using the 3D model of the hull as geometric constraints, and employing an improved genetic algorithm to generate a global macropath covering the target rust removal area, includes: The underwater rust removal robot extracts the current position from the real-time pose data of the global perception fusion data as the starting point of the path, extracts the working area from the high-level task instructions as the set of path endpoints, and uses the surface of the 3D model of the hull as the path search space. The underwater rust removal robot constructs a fitness function with the total path length, energy consumption and rust removal coverage as optimization objectives, and introduces the surface geometric constraints of the 3D model of the hull into the fitness function. Through the improved selection, crossover and mutation operations of the genetic algorithm, iterative optimization is carried out, and finally the candidate path with the best fitness is output as the global macro path covering the target rust removal area.

[0039] Specifically, firstly, the underwater rust removal robot extracts its current position from the real-time pose data of the global perception fusion data as the starting point for path planning. At the same time, it extracts the working area from the high-level task instructions as the endpoint set for path planning. This endpoint set is not a single coordinate point, but a series of spatial nodes covering the boundary of the target rust removal area to ensure that the path can eventually traverse the entire working area. Then, the robot uses the surface of the 3D model of the hull as the path search space, reducing the path planning problem from three-dimensional free space to two-dimensional curved surface space. Specifically, the surface of the 3D model of the hull is discretized into a digital curved surface composed of triangular or quadrilateral meshes. Path points are constrained to these mesh nodes and their connections, thereby ensuring that each planned candidate path always fits the hull wall and avoids the path from deviating from the hull surface and losing engineering feasibility. After the search space and path endpoints are determined, the underwater rust removal robot constructs a multi-objective fitness function with the total path length, energy consumption, and rust removal coverage as optimization objectives. The total path length directly affects the operation time efficiency; energy consumption reflects the energy consumption differences of the robot moving on different slopes of the hull surface; and rust removal coverage evaluates whether the path fully covers the target area and whether there are any missed areas. These three objectives are not entirely consistent; for example, the shortest path may not have the lowest energy consumption or the optimal coverage. Therefore, the fitness function uses a weighted summation method to unify the three into a single-objective fitness value. The weight coefficients can be preset or adjusted online according to the specific requirements of the task. Simultaneously, the underwater rust removal robot incorporates the surface geometric constraints of the hull's 3D model into the fitness function. These geometric constraints include the curvature and normal information of the hull surface. Specifically, if a candidate path traverses areas with excessive curvature or abrupt changes in normal direction, it means that the robot needs to make significant attitude adjustments at that location, which not only increases energy consumption and operational difficulty but may also damage the hull surface. The aforementioned geometric constraints are incorporated into the fitness function as a penalty function: when a candidate path violates the geometric constraints, the fitness function imposes a corresponding penalty value on the path, reducing its overall fitness and thus gradually eliminating it in subsequent iterations. In this way, geometric constraints and the three optimization objectives are evaluated uniformly within the same framework, ensuring both the optimality and physical feasibility of the path.Building upon this foundation, the underwater rust removal robot employs an improved genetic algorithm for iterative optimization through selection, crossover, and mutation operations. In the initialization phase, the algorithm randomly generates multiple candidate paths within the path search space to form an initial population. Each candidate path is represented as an ordered sequence of nodes starting from the origin, passing through intermediate nodes, and finally reaching a node in the destination set. In the selection operation, the algorithm uses a roulette wheel or tournament selection strategy based on the fitness function values ​​of each candidate path to retain high-fitness, superior individuals. In the crossover operation, the algorithm randomly selects two parent paths from the retained superior individuals and exchanges some path segments between them with a certain probability, generating new offspring paths that inherit the superior characteristics of their parents and explore new path combinations. In the mutation operation, the algorithm randomly fine-tunes some intermediate nodes in the path with a small probability, introducing new genetic diversity and preventing the algorithm from prematurely converging to a local optimum. These selection, crossover, and mutation operations are iterated repeatedly, with the population evolving generation by generation and the fitness function value increasing with each generation. When the preset maximum number of iterations is reached or the fitness function value no longer significantly improves over multiple generations, the algorithm terminates the iteration and outputs the candidate path with the best fitness as a global macropath covering the target rust removal area. Through the complete process of the improved genetic algorithm described above, the underwater rust removal robot finally obtains a global macropath that starts from the current position, traverses the target rust removal area, achieves an optimal balance between path length, energy consumption, and rust removal coverage, and perfectly conforms to the hull surface. This global macropath serves as the macroscopic motion benchmark for subsequent local obstacle avoidance and autonomous operation, laying the path foundation for the robot to efficiently and safely complete large-area underwater rust removal operations.

[0040] In one embodiment, as the underwater rust removal robot moves towards the target work area, it collects aquatic environmental parameters in real time and inputs them into a preset reinforcement learning decision model, dynamically matching the optimal communication mode under the current aquatic environment, including: As the underwater rust removal robot moves towards the target work area, it collects water environment parameters in real time. The robot then inputs these parameters into a pre-set reinforcement learning decision model, which simultaneously acquires the communication quality index of the current link and the priority parameters of the current task. Using the water environment parameters, communication quality index, and priority parameters as input, the reinforcement learning decision model calculates the cumulative expected value of each candidate communication mode and outputs the optimal communication mode under the current water environment. The underwater rust removal robot then switches communication links according to the optimal communication mode, achieving dynamic matching of communication modes.

[0041] Specifically, as the underwater rust removal robot moves towards the target work area, it collects real-time water environment parameters through onboard water environment sensors. These sensors include a turbidity sensor, a temperature sensor, and a depth gauge, used to detect the current water turbidity, water temperature, and the robot's diving depth, respectively. Then, the underwater rust removal robot inputs the collected water environment parameters into a pre-set reinforcement learning decision model. This reinforcement learning decision model preferably uses a deep Q-network model, containing an input layer, several hidden layers, and an output layer. While receiving the water environment parameters, the model simultaneously acquires the communication quality indicators of the current communication link and the priority parameters of the current task. The communication quality indicators include the bit error rate and transmission delay of the current communication link, and the task priority parameters include real-time control priority and data feedback priority. Next, the reinforcement learning decision model combines the water environment parameters, communication quality indicators, and priority parameters into an input state vector for the current moment. This state vector comprehensively describes the physical conditions of the current communication environment, the link status, and the service requirements. The model calculates the cumulative expected value of each candidate communication mode through forward propagation. Candidate communication modes include at least pure optical communication, pure underwater acoustic communication, and optical-acoustic parallel redundant communication. The cumulative expected value reflects the comprehensive reward that can be obtained over a period of time after selecting a certain communication mode in the current state. This comprehensive reward takes into account factors such as communication success rate, transmission bandwidth utilization, and mode switching smoothness. The model selects the candidate communication mode with the highest cumulative expected value as the optimal communication mode in the current water environment and outputs it. Finally, the underwater rust removal robot performs communication link switching according to the optimal communication mode, switching the current working communication mode to the optimal communication mode, achieving dynamic matching of communication modes. Through the closed-loop mechanism of real-time acquisition, reinforcement learning decision-making, and link switching, the robot can proactively predict and adjust the communication mode according to changes in the water environment during its movement, ensuring optimal communication transmission performance under any water conditions. This provides reliable communication support for high-definition video transmission and control command transmission in subsequent rust removal operations.

[0042] In one embodiment, the reinforcement learning decision model takes aquatic environmental parameters, communication quality indicators, and priority parameters as input states, calculates the cumulative expected value of each candidate communication mode, and outputs the optimal communication mode under the current aquatic environment, including: The reinforcement learning decision model combines water environment parameters, communication quality indicators, and priority parameters into a state vector. Based on the state vector, the reinforcement learning decision model calculates the cumulative expected value of each candidate communication mode through a preset reward function. The reward function uses the communication success rate as the main positive reward term and the mode switching frequency as the penalty term to suppress frequent switching of communication modes. The reinforcement learning decision model selects the candidate communication mode with the largest cumulative expected value as the optimal communication mode output in the current water environment.

[0043] Specifically, the reinforcement learning decision model first combines the current water environment parameters, communication quality indicators, and task priority parameters into a state vector for the current moment. This state vector fully describes all information in three dimensions: the environmental physical conditions, the real-time status of the link, and the business requirements of the current communication scenario. The model inputs the above state vector into a deep Q-network and calculates the cumulative expected value (Q-value) of each candidate communication mode through the forward propagation of the network. In the process of calculating the cumulative expected value of the reward, the model uses a preset reward function to evaluate each candidate communication mode. This reward function takes the communication success rate as the main positive reward term, that is, the higher the communication success rate, the greater the positive reward assigned to the mode by the model. At the same time, it uses the frequency of mode switching as a penalty term. That is, when the communication mode is frequently switched recently, the model applies an additional negative reward penalty to the mode, thereby effectively suppressing the frequent oscillation of the communication mode and avoiding the impact on transmission stability caused by small fluctuations in water environment parameters leading to back-and-forth switching of the communication mode. Finally, the model selects the one with the largest Q-value among the cumulative expected values ​​of the candidate communication modes as the optimal communication mode in the current water environment and outputs it, so that the underwater rust removal robot can perform communication link switching to achieve dynamic matching of communication modes.

[0044] In one embodiment, such as Figure 3 As shown, a remote control system for an underwater rust removal robot is provided, including the underwater rust removal robot, a surface relay platform, and a remote control terminal. The underwater rust removal robot includes: The task instruction receiving module is used to receive high-level task instructions, including a three-dimensional model of the ship's hull and the work area, issued by the remote control terminal via the surface relay platform. The communication pattern matching module is used to collect water environment parameters in real time and input them into a preset reinforcement learning decision model as the underwater rust removal robot moves toward the target work area, so as to dynamically match the optimal communication mode in the current water environment. The data output module is used to dynamically correct the confidence weights of each sensor in the extended Kalman filter based on the real-time sensing data collected by the multi-source sensors on the machine, combined with the evolutionary optimization algorithm, and output the full-domain sensing fusion data in real time. The rust removal trajectory generation module is used to combine high-level task instructions and global perception fusion data to generate rust removal motion trajectories through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, so as to complete underwater rust removal operations. The intervention request module is used to send an intervention request to the remote control terminal through the optimal communication mode when the uncertain and complex working conditions are identified based on the fusion data of the whole domain perception. After receiving the high-level adjustment instructions generated by the remote control terminal based on virtual gestures, the module replans the operation path and continues to execute the rust removal operation.

[0045] Specifically, the task instruction receiving module is used to receive high-level task instructions, including the 3D model of the hull and the work area, issued by the remote control terminal via the surface relay platform. The surface relay platform acts as the communication hub between the underwater robot and the remote control terminal, undertaking communication protocol conversion, data relay, and link pre-detection functions. The remote control terminal is only responsible for high-level non-real-time decision-making and does not need to operate the robot's actuators in real time, thus avoiding the impact of remote communication delays on the underlying control. The 3D model of the hull provides prior geometric information about the work environment, while the work area defines the spatial scope of the rust removal task, providing target boundaries for subsequent path planning.

[0046] The communication mode matching module is used to collect water environment parameters in real time as the underwater rust removal robot moves towards the target work area and input them into a preset reinforcement learning decision model to dynamically match the optimal communication mode for the current water environment. Specifically, the water environment sensors on the underwater robot monitor environmental parameters such as turbidity, temperature, and depth in real time. The reinforcement learning decision model takes environmental parameters, current communication quality indicators, and task priorities as inputs, evaluates the cumulative expected value of each candidate communication mode through a pre-trained decision network, and outputs the optimal communication mode that matches the current water environment. The system can use high-bandwidth optical communication in clear water to ensure real-time transmission of high-definition video streams, and switch to underwater acoustic communication in turbid or long-distance working conditions to ensure command reachability, realizing predictive intelligent switching of communication modes.

[0047] The data output module is used to dynamically adjust the confidence weights of each sensor in the extended Kalman filter based on the real-time sensing data collected by the multi-source sensors on the underwater robot, combined with an evolutionary optimization algorithm, to output real-time global sensing fusion data. Specifically, the underwater robot is equipped with multiple navigation and sensing sensors. The extended Kalman filter, as the core framework for multi-source data fusion, fuses the data from each sensor according to the optimal estimation theory, outputting the robot's pose and surrounding environment information. The evolutionary optimization algorithm monitors the consistency and volatility indicators of the data from each sensor in real time, dynamically adjusting the observation variance and confidence weights of each sensor in the extended Kalman filter, so that the system can still maintain high-precision positioning when sensors fail locally, ensuring the accurate maintenance of a constant working distance between the rust removal nozzle and the hull surface.

[0048] The rust removal trajectory generation module combines high-level task instructions and global perception fusion data to generate a rust removal motion trajectory using a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, thus completing the underwater rust removal operation. Specifically, the global planning layer uses the work area in the high-level task instructions as constraints, the real-time pose in the global perception fusion data as the starting point, and the 3D model of the hull as the geometric boundary. It uses an improved genetic algorithm to generate a global macro path covering the target rust removal area. As the robot moves along the global macro path, the local planning layer uses the surrounding environment information in the global perception fusion data as input, runs an artificial potential field method to perceive local obstacles in real time, and dynamically adjusts the direction of travel to avoid obstacles. At the same time, it keeps the robot at a constant optimal rust removal distance from the hull wall. The collaboration of these two layers allows the robot to autonomously complete the rust removal motion trajectory without relying on remote commands.

[0049] The intervention request module is used to send an intervention request to the remote control terminal when uncertain and complex working conditions are identified based on the fusion data of the full-domain perception system. It then sends the request through the optimal communication mode and, upon receiving a high-level adjustment instruction generated by the remote control terminal based on virtual gestures, replans the work path and continues the rust removal operation. Specifically, when the fusion data of the full-domain perception system shows complex working conditions such as welds, highly corroded steel plates, or unknown underwater attachments in the current area, and the confidence level of the onboard AI system is lower than a preset threshold, the robot automatically slows down and sends an intervention request to the remote control terminal through the current optimal communication mode. After receiving the request through the digital twin somatosensory interaction system, the operator uses virtual gestures to delineate the target area or avoidance position and issues a high-level adjustment instruction. After receiving the instruction, the robot uses the target point or avoidance area as the new path constraint, calls the local path planning module to replan the work path, and continues the rust removal operation.

[0050] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A remote control method for an underwater rust removal robot, characterized in that, include: The underwater rust removal robot receives high-level task instructions, including a three-dimensional model of the hull and the work area, from the remote control terminal via a surface relay platform. As the underwater rust removal robot moves toward the target work area, it collects water environment parameters in real time and inputs them into a preset reinforcement learning decision model to dynamically match the optimal communication mode in the current water environment. The underwater rust removal robot uses real-time sensing data collected by the multi-source sensors on board to dynamically correct the confidence weights of each sensor in the extended Kalman filter using an evolutionary optimization algorithm, and outputs real-time global sensing fusion data. The underwater rust removal robot combines the high-level task instructions and the global perception fusion data to generate a rust removal trajectory through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, thereby completing the underwater rust removal operation. When the underwater rust removal robot identifies uncertain and complex working conditions based on the fusion data of the whole-domain perception, it sends an intervention request to the remote control terminal through the matched optimal communication mode. After receiving the high-level adjustment instruction generated by the remote control terminal based on virtual gestures, it replans the operation path and continues to perform the rust removal operation.

2. The remote control method for an underwater rust removal robot according to claim 1, characterized in that, The underwater rust removal robot uses real-time sensing data collected by its onboard multi-source sensors, combined with an evolutionary optimization algorithm to dynamically adjust the confidence weights of each sensor in the extended Kalman filter, to output real-time global sensing fusion data, including: The underwater rust removal robot acquires real-time sensing data from the multi-source sensors mounted on the machine; The underwater rust removal robot extracts the measurement values ​​of each sensor at the current moment from the sensing data at a preset frequency, and performs time updates by extended Kalman filtering based on the robot state estimation and robot kinematic model at the previous moment to obtain the predicted measurement values ​​of each sensor at the current moment. The underwater rust removal robot calculates the residual between the measured value of each sensor at the current moment and the corresponding predicted measured value, calculates the data consistency index of each sensor based on the residual, and calculates the data volatility index based on the fluctuation range of the measured value of each sensor within a preset time window. The underwater rust removal robot dynamically corrects the confidence weights of each sensor in the extended Kalman filter based on the data consistency index and the data volatility index, combined with an evolutionary optimization algorithm, and outputs global perception fusion data in real time.

3. The remote control method for an underwater rust removal robot according to claim 2, characterized in that, The underwater rust removal robot dynamically adjusts the confidence weights of each sensor in the extended Kalman filter based on the data consistency index and the data volatility index, combined with an evolutionary optimization algorithm, and outputs real-time global perception fusion data, including: The underwater rust removal robot inputs the data consistency index and the data volatility index into the evolutionary optimization algorithm, which iteratively updates the observation variance and confidence weight of each sensor. The underwater rust removal robot inputs the updated observation variance and the confidence weight into the extended Kalman filter, performs measurement updates, weights and fuses the measurements of the multi-source sensors at the current moment, and outputs the real-time pose and surrounding environment perception information of the underwater rust removal robot as global perception fusion data.

4. The remote control method for an underwater rust removal robot according to claim 3, characterized in that, The underwater rust removal robot inputs the updated observation variance and the confidence weights into the extended Kalman filter to perform measurement updates. It then performs weighted fusion of the measurements from the multi-source sensors at the current moment, outputting the real-time pose and surrounding environment perception information of the underwater rust removal robot. This output data, as global perception fusion data, includes: The underwater rust removal robot inputs the updated observation variance and confidence weights of each sensor into the extended Kalman filter, which then performs the measurement update. The extended Kalman filter performs a weighted correction on the residuals between the measured values ​​and predicted measured values ​​of each sensor at the current time based on the confidence weights and observation variances of each sensor, thereby obtaining the posterior state estimate and the corresponding error covariance matrix at the current time, wherein the error covariance matrix characterizes the uncertainty of the posterior state estimate. The underwater rust removal robot extracts real-time pose data from the posterior state estimation, extracts surrounding environment perception data from the real-time data collected by the multi-source sensors, and integrates the real-time pose data and the surrounding environment perception data into global perception fusion data.

5. The remote control method for an underwater rust removal robot according to claim 1, characterized in that, The underwater rust removal robot, combining the high-level task instructions and the fusion of global perception data, generates a rust removal trajectory through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, and completes the underwater rust removal operation, including: The underwater rust removal robot extracts the work area information from the high-level task instructions and combines it with the real-time pose data in the global perception fusion data. Using the hull 3D model as geometric constraints, it uses an improved genetic algorithm to generate a global macro path covering the target rust removal area. As the underwater rust removal robot travels along the global macro path, it uses the surrounding environment perception data from the global perception fusion data as input and runs the artificial potential field method to calculate in real time the repulsive force generated by local obstacles and the attractive force generated by the target rust removal area. It generates a local obstacle avoidance path to dynamically correct the deviation of the global macro path in the local area, so that the underwater rust removal robot can complete the underwater rust removal operation while maintaining a constant optimal working distance between the rust removal nozzle and the hull wall.

6. The remote control method for an underwater rust removal robot according to claim 5, characterized in that, The step of combining the real-time pose data from the global perception fusion data, using the 3D model of the hull as geometric constraints, and employing an improved genetic algorithm to generate a global macropath covering the target rust removal area includes: The underwater rust removal robot extracts the current position from the real-time pose data of the global perception fusion data as the path start point, extracts the working area from the high-level task instructions as the path end point set, and uses the surface of the ship's three-dimensional model as the path search space. The underwater rust removal robot constructs a fitness function with the total path length, energy consumption, and rust removal coverage as optimization objectives. The surface geometric constraints of the three-dimensional model of the hull are introduced into the fitness function. The selection, crossover, and mutation operations of the improved genetic algorithm are used to iteratively optimize the algorithm. Finally, the candidate path with the best fitness is output as the global macro path covering the target rust removal area.

7. The remote control method for an underwater rust removal robot according to claim 1, characterized in that, As the underwater rust removal robot moves towards the target work area, it collects water environment parameters in real time and inputs them into a preset reinforcement learning decision model, dynamically matching the optimal communication mode under the current water environment, including: The underwater rust removal robot collects water environment parameters in real time as it moves toward the target work area; The underwater rust removal robot inputs the water environment parameters into a preset reinforcement learning decision model, and the reinforcement learning decision model simultaneously obtains the communication quality index of the current link and the priority parameters of the current task. The reinforcement learning decision model takes the water environment parameters, the communication quality index, and the priority parameters as input states, calculates the cumulative expected value of each candidate communication mode, and outputs the optimal communication mode under the current water environment. The underwater rust removal robot switches communication links according to the optimal communication mode to achieve dynamic matching of communication modes.

8. The remote control method for an underwater rust removal robot according to claim 7, characterized in that, The reinforcement learning decision model takes the water environment parameters, the communication quality index, and the priority parameters as input states, calculates the cumulative expected value of each candidate communication mode, and outputs the optimal communication mode under the current water environment, including: The reinforcement learning decision model combines the water environment parameters, the communication quality indicators, and the priority parameters into a state vector. The reinforcement learning decision model calculates the cumulative expected value of each candidate communication mode based on the state vector through a preset reward function. The reward function uses the communication success rate as the main positive reward term and the mode switching frequency as the penalty term to suppress frequent switching of communication modes. The reinforcement learning decision model selects the candidate communication mode with the highest cumulative expected value as the optimal communication mode output in the current water environment.

9. A remote control system for an underwater rust removal robot, characterized in that, The system includes an underwater rust removal robot, a surface relay platform, and a remote control terminal. The underwater rust removal robot includes: The task instruction receiving module is used to receive high-level task instructions, including a three-dimensional model of the ship hull and the work area, issued by the remote control terminal via the surface relay platform. The communication mode matching module is used to collect water environment parameters in real time and input them into a preset reinforcement learning decision model during the underwater rust removal robot's journey to the target work area, so as to dynamically match the optimal communication mode in the current water environment. The data output module is used to dynamically correct the confidence weights of each sensor in the extended Kalman filter based on the real-time sensing data collected by the multi-source sensors on the machine, combined with the evolutionary optimization algorithm, and output the full-domain sensing fusion data in real time. The rust removal trajectory generation module is used to combine the high-level task instructions and the global perception fusion data to generate a rust removal motion trajectory through a hybrid path algorithm that combines global planning with local dynamic obstacle avoidance, so as to complete the underwater rust removal operation. The intervention request module is used to send an intervention request to the remote control terminal through the matched optimal communication mode when the uncertain complex working conditions are identified based on the full-domain perception fusion data. After receiving the high-level adjustment instruction generated by the remote control terminal based on virtual gestures, the module replans the operation path and continues to execute the rust removal operation.

Images