An intelligent autonomous remote control underwater robot ARV and a control system thereof

Abstract

The present disclosure provides an intelligent autonomous remote control underwater robot ARV and its control system, and relates to the technical field of machine learning, which comprises: a body environment perception and risk assessment system that collects environment risk information and current operation execution information and constructs a current risk vector; an adaptive boundary generation and optimization system that stores at least two risk boundary vectors and calculates a similarity index between the current risk vector; a control mode decision and switching system that determines a target control mode according to the comparison result of the similarity index and a preset threshold value and executes; an intelligent performance evaluation and feedback optimization system that collects key performance indicators to determine a task score and judges whether to update the risk boundary vector according to the task score, and when updating, the adaptive boundary generation and optimization system generates a candidate boundary vector based on the task score and replaces the original risk boundary vector after verification. The scheme solves the problem that the ARV control mode switching in the prior art relies on manual operation and is difficult to adapt to environmental changes.

Description

Technical Field

[0001] This disclosure relates to the field of machine learning technology, and in particular to an intelligent autonomous remotely operated underwater vehicle (ARV) and its control system. Background Technology

[0002] Autonomous Remotely Operated Vehicles (ARVs) are increasingly used in marine exploration, facility maintenance, and underwater surveys. Traditional ARVs typically have two control modes: fully autonomous and manually controlled. However, each mode has its limitations: fully autonomous mode struggles to cope with complex dynamic environments, while manually controlled mode is constrained by communication latency and operator subjectivity. Existing control mode switching methods mainly rely on operator judgment or simple logic based on single sensor thresholds, resulting in issues such as response lag, subjective judgment, and abrupt switching. Especially in complex scenarios involving multi-task collaboration, unknown underwater disturbances, and energy sensitivity, existing switching methods struggle to flexibly adjust to dynamic environmental changes and cannot achieve adaptive optimization of ARV control strategies. Summary of the Invention

[0003] This disclosure provides an intelligent autonomous remotely operated underwater vehicle (ARV) and its control system to at least solve the above-mentioned technical problems existing in the prior art.

[0004] According to a first aspect of this disclosure, a control system for an intelligent autonomous remotely operated underwater vehicle (ARV) is provided, the system comprising:

[0005] An embodied environmental perception and risk assessment system is used to collect environmental risk information and current operation status information of the ARV, and to construct a current risk vector based on the environmental risk information and current operation status information;

[0006] An adaptive boundary generation and optimization system is used to store at least two risk boundary vectors and calculate the similarity index between the current risk vector and each of the risk boundary vectors.

[0007] A control mode decision and switching system is used to determine and execute the target control mode of the ARV from multiple preset control modes based on the comparison results of the similarity index and the preset threshold.

[0008] The intelligent performance evaluation and feedback optimization system is used to collect key performance indicators of the ARV when it performs tasks in the target control mode, determine the task score based on the key performance indicators, and determine whether to update the risk boundary vector based on the task score.

[0009] If updated, the adaptive boundary generation and optimization system is also used to generate candidate boundary vectors based on the task score, and replace the original risk boundary vectors after verification. The updated risk boundary vectors are used to determine the target control mode for the next time.

[0010] In one possible implementation, the embodied environmental perception and risk assessment system includes:

[0011] The multimodal sensing data acquisition module is used to collect environmental risk information of the ARV, including environmental parameters and operating status parameters;

[0012] The job status monitoring module is used to collect information on the current job execution status of the ARV;

[0013] The intelligent fuzzy evaluation module is used to filter and fuse the environmental risk information and the current operation execution information, and map the fused data to multiple preset risk levels to construct a multi-dimensional risk fuzzy set.

[0014] The defuzzification risk output module is used to perform defuzzification calculations on the multidimensional risk fuzzy set and construct the current risk vector.

[0015] In one possible implementation, the adaptive boundary generation and optimization system includes:

[0016] An initial boundary setting module is used to preset the risk boundary vector, which includes at least a first risk boundary vector and a second risk boundary vector.

[0017] The embodied perception similarity matching module is used to calculate the similarity between the current risk vector and the first risk boundary vector and the second risk boundary vector, respectively, and generate a first similarity index and a second similarity index.

[0018] In one embodiment, the first risk boundary vector and the second risk boundary vector are respectively a low-risk boundary vector and a high-risk boundary vector, and the control mode decision and switching system includes:

[0019] The intelligent decision-making reasoning module is used to determine that the target control mode of the ARV is the fully autonomous mode when the first similarity index is greater than or equal to the first preset threshold and the second similarity index is less than the second preset threshold.

[0020] When the first similarity index is less than the first preset threshold and the second similarity index is less than the second preset threshold, the target control mode of the ARV is determined to be semi-autonomous mode.

[0021] When the second similarity index is greater than or equal to the second preset threshold, the target control mode of the ARV is determined to be the manual remote control mode.

[0022] In one possible implementation, the intelligent performance evaluation and feedback optimization system includes:

[0023] The real-time performance monitoring module is used to collect key performance indicators of the ARV when performing tasks, including at least one of the following: trajectory deviation, task completion time, energy consumption and failure rate.

[0024] A multi-dimensional performance scoring module is used to calculate the task score based on the key performance indicators using a weighted loss function.

[0025] An adaptive reward feedback module is used to map the task score into a reinforcement learning reward signal and send it to the adaptive boundary generation and optimization system when the task score meets the update conditions.

[0026] In one embodiment, the adaptive boundary generation and optimization system further includes an adaptive policy evolution module, used for:

[0027] Obtain the task success rate S, energy efficiency E, and mode switching frequency N of the ARV, and construct the following reward function. :

[0028]

[0029] in, , , These are the weighting factors;

[0030] Based on the reward function Calculate its gradient with respect to the current risk vector. The risk boundary vector is updated according to the following iterative formula to generate the candidate boundary vector:

[0031]

[0032] in, For candidate boundary vectors, For the risk boundary vector, This is the learning rate.

[0033] In one possible implementation, the adaptive boundary generation and optimization system further includes a boundary verification and replacement module, used for:

[0034] The candidate boundary vector is subjected to first-level verification, which includes: simulating and calculating a first simulation score when using the candidate boundary vector and a second simulation score when using the original risk boundary vector in a simulation environment;

[0035] The score gain is determined based on the first and second simulated scores;

[0036] When the scoring gain is greater than a preset gain threshold, the candidate boundary vector is determined to have passed the first-level verification.

[0037] The candidate boundary vectors that pass the first-level verification are subjected to a second-level verification, which includes: performing the task at least once in a real environment and calculating the average task score.

[0038] When the average task score is greater than the minimum score threshold, the candidate boundary vector is determined to have passed the second-level verification.

[0039] When the candidate boundary vector passes both the first-level verification and the second-level verification, the boundary verification and replacement module replaces the candidate boundary vector with the current risk boundary vector; otherwise, the replacement is terminated and the original risk boundary vector is maintained.

[0040] In one possible implementation, the multidimensional performance scoring module is specifically used for:

[0041] The task score is calculated using the following formula:

[0042]

[0043] in These are the normalized values ​​for trajectory deviation, task completion time, energy consumption, and failure rate, respectively. , , and They are weighting factors and ;

[0044] The adaptive reward feedback module is specifically used for:

[0045] The task score is mapped to a reinforcement learning reward signal j according to the following formula:

[0046]

[0047] in, Scaling factor The historical average score.

[0048] In one possible implementation, the control mode decision and switching system further includes a smooth switching control module, used for:

[0049] After determining the target control mode, a pre-check mechanism is used to determine whether the ARV meets the switching conditions;

[0050] If the conditions are met, the control parameters between the current control mode and the target control mode are continuously interpolated and transitioned, and the status of the ARV is monitored in real time during the transition process.

[0051] When an abnormal state is detected, the switching process is aborted and the control parameters are rolled back to the original control mode.

[0052] According to a second aspect of this disclosure, an intelligent autonomous remotely operated underwater vehicle (ARV) is provided, including the intelligent autonomous remotely operated underwater vehicle (ARV) control system described in any of the preceding claims.

[0053] This disclosure discloses an intelligent autonomous remotely operated underwater vehicle (ARV) and its control system. The system employs an embodied environment perception and risk assessment system to collect environmental risk information and current operation status information, constructing a current risk vector. An adaptive boundary generation and optimization system stores at least two risk boundary vectors and calculates the similarity index between the current risk vector and each risk boundary vector. A control mode decision and switching system determines and executes a target control mode from multiple preset control modes based on the comparison results of the similarity index and preset thresholds. An intelligent performance evaluation and feedback optimization system collects key performance indicators to determine a task score and decides whether to update the risk boundary vector based on the task score. During updating, the adaptive boundary generation and optimization system generates candidate boundary vectors based on the task score and replaces the original risk boundary vector after verification. The updated risk boundary vector is used to determine the next target control mode. To address the technical problems of existing ARV control mode switching, which relies on manual judgment or fixed thresholds, suffers from lag, abrupt switching, and difficulty in adapting to dynamic environmental changes, this solution combines environmental risk perception with operational status monitoring to construct a quantified risk vector. It then replaces fixed thresholds with a dynamically adjustable risk boundary vector for similarity matching decisions, achieving intelligent and flexible switching of control modes. This avoids the subjectivity of manual judgment and the rigidity of fixed thresholds. By introducing performance evaluation and reinforcement learning feedback mechanisms, the system can continuously optimize the risk boundary vector based on task execution results, forming a closed-loop self-evolution capability. This significantly improves the ARV's adaptability, switching smoothness, and long-term operational reliability in complex and dynamic underwater environments.

[0054] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0055] The above and other objects, features, and advantages of this disclosure will become readily apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. Several embodiments of this disclosure are illustrated in the drawings by way of example and not limitation, in which:

[0056] In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0057] Figure 1 A schematic diagram of the control system structure of the intelligent autonomous remotely operated underwater vehicle (ARV) according to an embodiment of this disclosure is shown.

[0058] Figure 2 A schematic diagram of the structure of the embodied environmental perception and risk assessment system according to an embodiment of this disclosure is shown;

[0059] Figure 3 A schematic diagram of the composition structure of an intelligent autonomous remotely controlled underwater robot according to an embodiment of the present disclosure is shown. Detailed Implementation

[0060] To make the objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0061] This disclosure provides an intelligent autonomous remotely controlled underwater vehicle (ARV) control system, such as Figure 1 As shown, the system includes:

[0062] The embodied environmental perception and risk assessment system is used to collect environmental risk information and current operation status information of the ARV, and to construct a current risk vector based on the environmental risk information and current operation status information.

[0063] In this example, the embodied environmental perception and risk assessment system is responsible for comprehensively perceiving the external environmental state of the ARV and its own operational status. Environmental risk information refers to data collected by various sensors reflecting the physical and optical characteristics of the underwater environment, such as water flow velocity and turbidity. Current operational status information refers to data reflecting the ARV's own workload and status, such as thruster load. This system fuses the aforementioned environmental risk information and current operational status information to construct a current risk vector that quantitatively characterizes the current environmental risk state, thus unifying multi-source heterogeneous information into a vector form.

[0064] An adaptive boundary generation and optimization system is used to store at least two risk boundary vectors and calculate the similarity index between the current risk vector and each of the risk boundary vectors.

[0065] In this example, the risk boundary vectors are pre-defined benchmark vectors used to delineate the switching thresholds between different control modes. There are at least two such vectors, each corresponding to a switching threshold between different control modes. The adaptive boundary generation and optimization system stores and updates these risk boundary vectors in real time. Upon receiving the current risk vector, it calculates a similarity index between the current risk vector and each risk boundary vector to characterize the degree of proximity between the current environmental risk state and the preset switching threshold.

[0066] The control mode decision and switching system is used to determine and execute the target control mode of the ARV from a variety of preset control modes based on the comparison results of the similarity index and the preset threshold.

[0067] In this example, the control mode decision and switching system receives a similarity index from the adaptive boundary generation and optimization system and compares it with an internal preset threshold. Based on the comparison result, the system can determine the most suitable ARV control mode, i.e., the target control mode, for the current environment and operating state. The preset control modes include at least fully autonomous, semi-autonomous, and manually controlled modes to adapt to operating scenarios of varying complexity. After determining the target control mode, the control mode decision and switching system performs a mode switching operation, adjusting the ARV's control authority or control logic to the target control mode.

[0068] The intelligent performance evaluation and feedback optimization system is used to collect key performance indicators of the ARV when it performs tasks in the target control mode, determine the task score based on the key performance indicators, and determine whether to update the risk boundary vector based on the task score.

[0069] In this example, as the ARV executes the task according to the target control mode, the intelligent performance evaluation and feedback optimization system continuously collects key performance indicators that reflect the task execution effect, such as trajectory tracking accuracy and task time. By quantifying and scoring these multi-dimensional key performance indicators, a task score representing the overall quality of the task execution is obtained. Based on this task score, it is further determined whether the currently stored risk boundary vector needs adjustment and optimization.

[0070] If updated, the adaptive boundary generation and optimization system is also used to generate candidate boundary vectors based on the task score, and replace the original risk boundary vectors after verification. The updated risk boundary vectors are used to determine the target control mode for the next time.

[0071] In this example, when the intelligent performance evaluation and feedback optimization system determines that the risk boundary vector needs to be updated, the adaptive boundary generation and optimization system generates one or more candidate boundary vectors that may be superior to the currently stored boundary vector based on the received task scores. These candidate boundary vectors need to be verified; only after their performance is confirmed to be superior to the original risk boundary vector will they be officially replaced. The updated risk boundary vector will be stored in the adaptive boundary generation and optimization system and used in subsequent tasks to calculate the similarity index with the current risk vector, thereby affecting the determination of the ARV's next target control mode.

[0072] This disclosure provides an intelligent autonomous remotely operated underwater vehicle (ARV) control system. The system employs an embodied environment perception and risk assessment system to collect environmental risk information and current operation status information, constructing a current risk vector. An adaptive boundary generation and optimization system stores at least two risk boundary vectors and calculates the similarity index between the current risk vector and each risk boundary vector. A control mode decision and switching system determines and executes a target control mode from multiple preset control modes based on the comparison results of the similarity index and a preset threshold. An intelligent performance evaluation and feedback optimization system collects key performance indicators to determine a task score and decides whether to update the risk boundary vector based on the task score. During updating, the adaptive boundary generation and optimization system generates candidate boundary vectors based on the task score and replaces the original risk boundary vector after verification. The updated risk boundary vector is used to determine the next target control mode. To address the technical problems of existing ARV control mode switching, which relies on manual judgment or fixed thresholds, suffers from lag, abrupt switching, and difficulty in adapting to dynamic environmental changes, this solution combines environmental risk perception with operational status monitoring to construct a quantified risk vector. It then replaces fixed thresholds with a dynamically adjustable risk boundary vector for similarity matching decisions, achieving intelligent and flexible switching of control modes. This avoids the subjectivity of manual judgment and the rigidity of fixed thresholds. By introducing performance evaluation and reinforcement learning feedback mechanisms, the system can continuously optimize the risk boundary vector based on task execution results, forming a closed-loop self-evolution capability. This significantly improves the ARV's adaptability, switching smoothness, and long-term operational reliability in complex and dynamic underwater environments.

[0073] In one example, the embodied environmental perception and risk assessment system includes:

[0074] The multimodal sensing data acquisition module is used to collect environmental risk information of the ARV, including environmental parameters and operating status parameters.

[0075] In this example, combined Figure 2As shown, the multimodal sensing data acquisition module integrates various types of sensors to collect environmental and operational status parameters of the ARV during underwater operation. Environmental parameters refer to data on the external physical fields and optical characteristics of the ARV, such as water flow velocity and direction obtained through a flow velocity sensor, water turbidity obtained through a turbidity sensor, ambient light intensity obtained through a light sensor, current water depth obtained through a depth sensor, and underwater terrain and obstacle information obtained through an underwater acoustic imager. Operational status parameters refer to data on the ARV's own motion state, such as attitude angle, angular velocity, and acceleration obtained through a pose angular velocity meter.

[0076] The job status monitoring module is used to collect information on the current job execution status of the ARV.

[0077] In this example, the job status monitoring module collects information on the current job execution status, primarily reflecting the ARV's workload and task progress. This includes data such as the real-time load power of the thrusters, the current and voltage information of the power supply system, the current task completion percentage, and the robot's posture stability indicators. After collecting the current job execution status information, the job status monitoring module performs standardized preprocessing, such as normalization, to enable data fusion and analysis with data from other modules.

[0078] The intelligent fuzzy evaluation module is used to filter and fuse the environmental risk information and the current operation execution information, and map the fused data to multiple preset risk levels to construct a multi-dimensional risk fuzzy set.

[0079] In this example, the intelligent fuzzy assessment module filters and fuses the received environmental risk information and current operation status information. For example, it uses a Kalman filter to eliminate noise interference in the sensor data and then makes predictions and estimates. Secondly, the module introduces an asymmetric fuzzification mechanism. Through a preset fuzzy rule base, it maps the filtered and fused precise values ​​to multiple preset risk levels, such as low-risk, medium-risk, and high-risk domains, thereby constructing a multi-dimensional risk fuzzy set that forms a buffer zone in the boundary transition area.

[0080] The defuzzification risk output module is used to perform defuzzification calculations on the multidimensional risk fuzzy set and construct the current risk vector.

[0081] In this example, the defuzzification risk output module uses a defuzzification calculation method, such as the centroid method, to process the multidimensional risk fuzzy set and output a normalized environmental risk index. And construct the current risk vector. Current risk vector It not only includes the current blurred normalized environmental risk index It also includes the extracted set of key risk characteristic variables, the structure of which is as follows:

[0082]

[0083] in, Indicates the intensity of the flow velocity disturbance. Indicates the light fluctuation coefficient. This indicates the power to drive load fluctuations. Indicates the standard deviation of attitude angles. Indicates the remaining percentage of tasks completed. This serves as a risk index. The current risk vector, composed of the aforementioned variables, can fully characterize the current environmental and task load status, providing interpretable and highly robust quantitative input for subsequent mode switching.

[0084] In one example, the adaptive boundary generation and optimization system includes:

[0085] The initial boundary setting module is used to preset the risk boundary vector, which includes at least a first risk boundary vector and a second risk boundary vector.

[0086] In this example, the initial boundary setting module analyzes expert experience and historical mission data to pre-set at least two risk boundary vectors. The first risk boundary vector is configured to define the transition threshold between fully autonomous control mode and semi-autonomous control mode, while the second risk boundary vector is configured to define the transition threshold between semi-autonomous control mode and manual remote control mode. Depending on the specific application scenario, these two boundary vectors can be quantified into multi-dimensional numerical forms; for example, the first risk boundary vector can be set to... The second risk boundary vector can be set as follows: The value of each dimension corresponds to the critical value of different types of risk characteristics.

[0087] The embodied perception similarity matching module is used to calculate the similarity between the current risk vector and the first risk boundary vector and the second risk boundary vector, respectively, and generate a first similarity index and a second similarity index.

[0088] In this example, the embodied perception similarity matching module receives the current risk vector from the embodied environment perception and risk assessment system. And respectively compared with the first risk boundary vector stored in the initial boundary setting module. Second risk boundary vector Perform similarity calculation.

[0089] This disclosure employs a cosine similarity algorithm, which measures the degree of similarity between vectors by calculating the cosine value of the angle between them, and generates a first similarity index. Second similarity index The first similarity index The second similarity index reflects the degree of proximity between the current risk status and the fully autonomous-semi-autonomous transition boundary. This reflects the degree of proximity between the current risk status and the semi-autonomous-manual remote control switching boundary. Both values ​​range from 0 to 1, with larger values ​​indicating closer proximity to the corresponding boundary.

[0090] In one example, the first risk boundary vector and the second risk boundary vector are a low-risk boundary vector and a high-risk boundary vector, respectively. The control mode decision and switching system includes:

[0091] The intelligent decision-making reasoning module is used to determine that the target control mode of the ARV is fully autonomous mode when the first similarity index is greater than or equal to the first preset threshold and the second similarity index is less than the second preset threshold; to determine that the target control mode of the ARV is semi-autonomous mode when the first similarity index is less than the first preset threshold and the second similarity index is less than the second preset threshold; and to determine that the target control mode of the ARV is manual remote control mode when the second similarity index is greater than or equal to the second preset threshold.

[0092] In this example, the intelligent decision-making reasoning module determines the most suitable target control mode for the ARV based on the comparison results between the similarity index and the preset threshold.

[0093] By setting the first preset threshold Second preset threshold As a trigger threshold for decision-making, Dangdang and When the current environmental risk has reached or exceeded the switching boundary between fully autonomous and semi-autonomous, but has not yet reached the high-risk area that requires human intervention, the ARV has the ability to independently cope with the current environment, and the target control mode is determined to be the fully autonomous mode.

[0094] when and This indicates that the current environmental risks have neither reached the boundary between fully autonomous and semi-autonomous operation, nor have they touched the high-risk boundary requiring human intervention. Therefore, the target control mode is determined to be a semi-autonomous mode, that is, a collaborative working mode between the robot and the operator, which leverages the robot's autonomous capabilities while retaining the interface for human intervention.

[0095] when When the situation reaches a certain point, it indicates that the current environmental risk has reached or exceeded the boundary between semi-autonomous and remotely controlled operation, entering a high-risk zone. At this point, the complexity of the environment or unexpected situations may exceed the ARV's autonomous response capabilities. Continuing to maintain autonomous or semi-autonomous mode could lead to safety accidents or mission failure. Therefore, the target control mode is determined to be remotely controlled, completely transferring control to the operator for remote operation, ensuring the robot's safe operation in extreme environments.

[0096] In one example, the intelligent performance evaluation and feedback optimization system includes:

[0097] The real-time performance monitoring module is used to collect key performance indicators of the ARV when performing tasks, including at least one of the following: trajectory deviation, task completion time, energy consumption, and failure rate.

[0098] In this example, the real-time performance monitoring module collects key performance indicators that quantify the task execution effect from the ARV execution unit during the ARV's execution of the task according to the target control mode. Among them, trajectory deviation refers to the degree of deviation between the ARV's actual flight trajectory and the preset expected trajectory; task completion time refers to the total time consumed from the start to the end of the task; energy consumption refers to the total amount of energy consumed by the ARV during task execution; and failure rate refers to the frequency of various abnormal events or equipment failures that occur during task execution.

[0099] A multidimensional performance scoring module is used to calculate the task score based on the key performance indicators using a weighted loss function.

[0100] In this example, the multidimensional performance scoring module uses a weighted loss function to sum multiple key performance indicators, such as trajectory deviation, task completion time, energy consumption, and failure rate, based on different weight coefficients, to obtain the task score. The weight coefficients of the weighted loss function can be dynamically configured according to the priority of different task types. For example, in emergency search and rescue missions, a higher weight can be given to task completion time, while in long-duration scientific expeditions, a higher weight can be given to energy consumption.

[0101] In a specific example, the multidimensional performance scoring module calculates the task score according to the following formula:

[0102]

[0103] in These are the normalized values ​​for trajectory deviation, task completion time, energy consumption, and failure rate, respectively. , , and They are weighting factors and .

[0104] An adaptive reward feedback module is used to map the task score into a reinforcement learning reward signal and send it to the adaptive boundary generation and optimization system when the task score meets the update conditions.

[0105] In this example, the adaptive reward feedback module determines whether the current task score meets preset update conditions, such as the task score reaching a certain threshold or showing a significant change compared to the historical average. When the update conditions are met, the task score is converted into a reinforcement learning reward signal through a mapping function. This reinforcement learning reward signal is sent to the adaptive boundary generation and optimization system via a secure communication mechanism for the next round of risk boundary vector optimization.

[0106] In a specific example, the adaptive reward feedback module maps the task score to a reinforcement learning reward signal j according to the following formula:

[0107]

[0108] in, Scaling factor The historical average score. Higher than When j is positive, it generates a positive stimulus, indicating that the current control strategy is better than the historical average; when Below When j is negative, a negative penalty is generated, indicating that the current control strategy needs improvement; when equal When j is zero, it means that the performance of this task is in line with the historical average.

[0109] In one example, the adaptive boundary generation and optimization system further includes an adaptive policy evolution module, used for:

[0110] Obtain the task success rate S, energy efficiency E, and mode switching frequency N of the ARV, and construct the following reward function. :

[0111]

[0112] in, , , These are weighting factors, used to balance the importance of different optimization objectives; task success rate. Characterizes the effectiveness of ARV in completing tasks under the guidance of the current risk boundary vector, and its energy efficiency. Characterizing the system's energy efficiency and mode switching frequency Characterizing the stability of the control strategy, more frequent switching usually means that the system state is more unstable.

[0113] Based on the reward function The gradient of the risk vector with respect to the current risk vector is calculated using the gradient descent method. This indicates the adjustment direction that will result in the fastest increase in the reward function value under the current parameter conditions.

[0114] The risk boundary vector is updated according to the following iterative formula to generate the candidate boundary vector:

[0115]

[0116] in, For candidate boundary vectors, The risk boundary vector; The learning rate is used to control the update step size in each iteration.

[0117] By accumulating experiential data through real physical interactions between the ARV and the underwater environment, and progressively training the curriculum based on a reinforcement learning framework according to the complexity of the environment, the system continuously senses embodied feedback signals such as water flow disturbances and energy consumption during the interaction process. The system iterates over the first and second risk boundary vectors to obtain updated candidate boundary vectors.

[0118]

[0119]

[0120] In one example, the adaptive boundary generation and optimization system further includes a boundary verification and replacement module, used for:

[0121] The candidate boundary vector undergoes a first-level verification, which includes: simulating the process in a simulation environment and calculating a first simulated score when using the candidate boundary vector and a second simulated score when using the original risk boundary vector. A score gain is determined based on the first and second simulated scores; when the score gain is greater than a preset gain threshold, the candidate boundary vector is deemed to have passed the first-level verification.

[0122] In this example, the boundary verification and replacement module uses a two-level verification mechanism to verify candidate boundary vectors, ensuring that only boundary vectors that actually improve performance can be officially enabled.

[0123] The boundary verification and replacement module first performs the first-level verification: in the physical model-based simulation system, it runs the control strategy using the candidate boundary vector and the control strategy using the original risk boundary vector respectively, calculates and records the corresponding first simulation score. Second simulation score .

[0124] Calculate the score gain And compare the gain with a preset gain threshold. Comparison. Only when... Only when the candidate boundary vector passes the first level of verification is it determined that the candidate boundary does have a performance improvement compared to the original boundary in the simulation environment.

[0125] The candidate boundary vector that has passed the first-level verification is subjected to a second-level verification, which includes: executing the task continuously at least once in a real environment and calculating the average task score; when the average task score is greater than the minimum score threshold, the candidate boundary vector is determined to have passed the second-level verification.

[0126] In this example, for candidate boundary vectors that pass the first-level simulation verification, the boundary verification and replacement module performs a second-level verification: that is, it conducts a field test in a real underwater environment.

[0127] This module deploys candidate boundary vectors into the ARV control system, guiding the ARV to continuously execute M tasks in a real-world scenario, where M is a preset threshold for the number of verification attempts. During each task execution, the system collects key performance indicators through an intelligent performance evaluation and feedback optimization system and calculates the task score for that task. After completing M tasks, the module calculates the average of these M task scores to obtain the average task score. Then, the average task score is compared with a preset minimum score threshold. When comparing, At that time, the candidate boundary vector is determined to have passed the second-level verification.

[0128] When the candidate boundary vector passes both the first-level verification and the second-level verification, the boundary verification and replacement module replaces the candidate boundary vector with the current risk boundary vector; otherwise, the replacement is terminated and the original risk boundary vector is maintained.

[0129] In this example, the boundary verification and replacement module makes a judgment. ∧ This indicates that when the candidate boundary vector passes the first-level and second-level verifications, the candidate boundary vector is used. , Replace the original risk boundary vector respectively , If a candidate boundary vector fails verification at any level, the module executes a rollback process to retain the original parameters of the candidate threshold, terminating the replacement operation and maintaining the original risk boundary vector unchanged.

[0130] In one example, the control mode decision and switching system further includes a smooth switching control module, used for:

[0131] After determining the target control mode, a pre-check mechanism is used to determine whether the ARV meets the switching conditions.

[0132] In this example, after receiving the control mode switching command, the control mode decision and switching system uses a pre-check mechanism executed by the smooth switching control module to determine whether the ARV meets the conditions for switching.

[0133] The pre-check mechanism includes two core checks: the first is the communication signal strength check, which determines whether the current communication signal strength is greater than a preset communication threshold. (For example, 70%), to ensure the reliability of the remote control link in high-risk modes requiring manual intervention; the second is the switching cooling interval verification, that is, to determine whether the time interval since the last switching exceeds the preset minimum cooling interval. (For example, 15 seconds) to avoid system oscillation and instability caused by frequent switching.

[0134] If any of the above conditions are not met, the system will reject the handover request and record the reason for rejection to ensure that the handover operation is only performed if the handover conditions are met.

[0135] If the conditions are met, the control parameters between the current control mode and the target control mode are continuously interpolated and transitioned, and the status of the ARV is monitored in real time during the transition process.

[0136] In this example, after the ARV passes the pre-check, the smooth switching control module uses a parameter interpolation mechanism to continuously interpolate and transition control parameters such as PID (Proportional-Integral-Derivative) parameters, control frequency, and sensor filter bandwidth between the current control mode and the target control mode.

[0137] During the parameter interpolation transition, the operating status of the ARV is monitored in real time. Specific monitoring indicators include: whether the attitude disturbance amplitude exceeds the preset upper limit δ (e.g., 15 degrees / second), whether the control response delay exceeds the set threshold τ (e.g., 500 milliseconds), and whether there are consistency verification failures among multiple redundant sensors.

[0138] When an abnormal state is detected, the switching process is aborted and the control parameters are rolled back to the original control mode.

[0139] In this example, when the current ARV operating status indicators are detected to be outside the safe range, the smooth switching control module takes differentiated handling measures based on the anomaly type:

[0140] If a conflict is detected as a redundant sensor, a manual intervention request is issued and the handover process is terminated. If a control response delay exceeds the limit, a delay retry mechanism is initiated, and the number of abnormalities is accumulated. If multiple retries are ineffective, further processing is carried out. If a severe attitude disturbance is detected, the handover is immediately aborted, and the control parameters are rolled back to the stable mode before the handover.

[0141] Regardless of the response strategy adopted, the system will record the switching status and anomaly type, and feed the relevant data back to the intelligent performance evaluation and feedback optimization system for subsequent strategy optimization learning. When the switching is completed successfully without triggering any anomalies, the system enters a success-marked state and records the execution log, parameter change trajectory, and summary data of monitoring indicators.

[0142] According to embodiments of this disclosure, this disclosure also provides an intelligent autonomous remotely controlled underwater robot.

[0143] Figure 3 A schematic block diagram of an example intelligent autonomous remotely operated underwater vehicle (AUV) 300 that can be used to implement embodiments of the present disclosure is shown. The intelligent AUV is intended to represent various forms of underwater work equipment, including but not limited to autonomous underwater vehicles (AUVs), remotely operated underwater vehicles (ROVs), hybrid underwater robots, and various underwater mobile platforms suitable for tasks such as ocean exploration, underwater maintenance, and seabed surveys. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0144] like Figure 3 As shown, the intelligent autonomous remotely operated underwater vehicle (ARUV) 300 includes a computing unit 301, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. The RAM 303 can also store various programs and data required for the operation of the intelligent ARUV 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0145] Multiple components of the intelligent autonomous underwater vehicle (AUV) 300 are connected to the I / O interface 305, including: an input unit 306, such as various sensors (including current sensors, turbidity sensors, depth sensors, light sensors, underwater acoustic imagers, attitude angular velocity meters, etc.), a thruster status monitoring device, a mission progress detection device, etc.; an output unit 307, such as actuators like thrusters, servos, and robotic arms, as well as communication devices for data transmission; a storage unit 308, such as non-volatile memory like solid-state drives and flash memory; and a communication unit 309, such as an underwater acoustic communication module, a radio communication module, and a fiber optic communication module. The communication unit 309 allows the intelligent AUV 300 to exchange information / data with the mother ship or control center through an underwater communication network and / or a surface relay network.

[0146] The computing unit 301 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the functions of the various systems and modules described above, such as the aforementioned embodied environment perception and risk assessment system, adaptive boundary generation and optimization system, control mode decision and switching system, and intelligent performance evaluation and feedback optimization system. For example, in some embodiments, the functions of the above-described control system can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on the robot 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by the computing unit 301, the functions of one or more modules of the control system described above can be performed. Alternatively, in other embodiments, the computing unit 301 may be configured to perform the functions of the control system described above by any other suitable means (e.g., by means of firmware).

[0147] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0148] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0149] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0150] To provide interaction with the user, the systems and technologies described herein can be implemented on a surface control station, which includes: a display device (e.g., an LCD screen, touchscreen monitor) for displaying robot status information to the user; and an input device (e.g., a keyboard, mouse, joystick, touchscreen) through which the user can send control commands or adjust control parameters to the intelligent autonomous underwater vehicle (AUV). Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including voice input, tactile input). Furthermore, the intelligent AUV itself can also exchange data bidirectionally with the surface control station through its onboard acoustic communication module and fiber optic communication module, receiving operator commands and transmitting its own status information and sensory data.

[0151] The systems and technologies described herein can be implemented in computing architectures that include underwater robot body control systems, surface monitoring and data processing systems, cloud-based data processing and analysis systems, or any combination thereof. The intelligent autonomous remotely operated underwater robot body, the surface control station, and the cloud server can be connected via digital data communication of any form or medium. Examples of communication networks include: underwater acoustic communication networks, fiber optic communication networks, wireless local area networks (WLANs), wide area networks (WANs), and the Internet.

[0152] The system can include a robot terminal and a server. The robot terminal is typically deployed at the underwater work site, while the server can be deployed at a surface control station or in the cloud. The two usually interact via a communication network. Collaboration between the robot terminal and the server is achieved through computer programs running on corresponding processors and exchanging data with each other. The server can be a cloud server, a distributed system server, or a server incorporating blockchain technology. Environmental and operational status data collected by the robot terminal can be uploaded to the server for storage and analysis. Optimization strategies or updated control parameters calculated by the server can also be distributed to the robot terminal via the communication network, enabling remote upgrades and continuous optimization of the control system.

[0153] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0154] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0155] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A control system for an intelligent autonomous remotely controlled underwater vehicle (ARV), characterized in that, The system includes: An embodied environmental perception and risk assessment system is used to collect environmental risk information and current operation status information of the ARV, and to construct a current risk vector based on the environmental risk information and current operation status information; An adaptive boundary generation and optimization system is used to store at least two risk boundary vectors and calculate the similarity index between the current risk vector and each of the risk boundary vectors. A control mode decision and switching system is used to determine and execute the target control mode of the ARV from multiple preset control modes based on the comparison results of the similarity index and the preset threshold. The intelligent performance evaluation and feedback optimization system is used to collect key performance indicators of the ARV when it performs tasks in the target control mode, determine the task score based on the key performance indicators, and determine whether to update the risk boundary vector based on the task score. If updated, the adaptive boundary generation and optimization system is also used to generate candidate boundary vectors based on the task score, and replace the original risk boundary vectors after verification. The updated risk boundary vectors are used to determine the target control mode for the next time. The adaptive boundary generation and optimization system further includes an adaptive policy evolution module, used for: Obtain the task success rate S, energy efficiency E, and mode switching frequency N of the ARV, and construct the following reward function. : in, , , These are the weighting factors; Based on the reward function Calculate its gradient with respect to the current risk vector. The risk boundary vector is updated according to the following iterative formula to generate the candidate boundary vector: in, For candidate boundary vectors, For the risk boundary vector, The learning rate; The adaptive boundary generation and optimization system further includes a boundary verification and replacement module, used for: The candidate boundary vector is subjected to first-level verification, which includes: simulating and calculating a first simulation score when using the candidate boundary vector and a second simulation score when using the original risk boundary vector in a simulation environment; The score gain is determined based on the first and second simulated scores; When the scoring gain is greater than a preset gain threshold, the candidate boundary vector is determined to have passed the first-level verification. The candidate boundary vectors that pass the first-level verification are subjected to a second-level verification, which includes: performing the task at least once in a real environment and calculating the average task score. When the average task score is greater than the minimum score threshold, the candidate boundary vector is determined to have passed the second-level verification. When the candidate boundary vector passes both the first-level verification and the second-level verification, the boundary verification and replacement module replaces the candidate boundary vector with the current risk boundary vector; otherwise, the replacement is terminated and the original risk boundary vector is maintained.

2. The system according to claim 1, characterized in that, The embodied environmental perception and risk assessment system includes: The multimodal sensing data acquisition module is used to collect environmental risk information of the ARV, including environmental parameters and operating status parameters; The job status monitoring module is used to collect information on the current job execution status of the ARV; The intelligent fuzzy evaluation module is used to filter and fuse the environmental risk information and the current operation execution information, and map the fused data to multiple preset risk levels to construct a multi-dimensional risk fuzzy set. The defuzzification risk output module is used to perform defuzzification calculations on the multidimensional risk fuzzy set and construct the current risk vector.

3. The system according to claim 1, characterized in that, Adaptive boundary generation and optimization systems include: An initial boundary setting module is used to preset the risk boundary vector, which includes at least a first risk boundary vector and a second risk boundary vector. The embodied perception similarity matching module is used to calculate the similarity between the current risk vector and the first risk boundary vector and the second risk boundary vector, respectively, and generate a first similarity index and a second similarity index.

4. The system according to claim 3, characterized in that, The first risk boundary vector and the second risk boundary vector are respectively a low-risk boundary vector and a high-risk boundary vector. The control mode decision and switching system includes: The intelligent decision-making reasoning module is used to determine that the target control mode of the ARV is the fully autonomous mode when the first similarity index is greater than or equal to the first preset threshold and the second similarity index is less than the second preset threshold. When the first similarity index is less than the first preset threshold and the second similarity index is less than the second preset threshold, the target control mode of the ARV is determined to be semi-autonomous mode. When the second similarity index is greater than or equal to the second preset threshold, the target control mode of the ARV is determined to be the manual remote control mode.

5. The system according to claim 1, characterized in that, The intelligent performance evaluation and feedback optimization system includes: The real-time performance monitoring module is used to collect key performance indicators of the ARV when performing tasks, including at least one of the following: trajectory deviation, task completion time, energy consumption and failure rate. A multi-dimensional performance scoring module is used to calculate the task score based on the key performance indicators using a weighted loss function. An adaptive reward feedback module is used to map the task score into a reinforcement learning reward signal and send it to the adaptive boundary generation and optimization system when the task score meets the update conditions.

6. The system according to claim 5, characterized in that, The multidimensional performance scoring module is specifically used for: The task score is calculated using the following formula: in These are the normalized values ​​for trajectory deviation, task completion time, energy consumption, and failure rate, respectively. , , and They are weighting factors and ; The adaptive reward feedback module is specifically used for: The task score is mapped to a reinforcement learning reward signal j according to the following formula: in, Scaling factor The historical average score.

7. The system according to claim 1, characterized in that, The control mode decision and switching system also includes a smooth switching control module, used for: After determining the target control mode, a pre-check mechanism is used to determine whether the ARV meets the switching conditions; If the conditions are met, the control parameters between the current control mode and the target control mode are continuously interpolated and transitioned, and the status of the ARV is monitored in real time during the transition process. When an abnormal state is detected, the switching process is aborted and the control parameters are rolled back to the original control mode.

8. An intelligent autonomous remotely controlled underwater robot, characterized in that, Including the intelligent autonomous remotely operated underwater vehicle (ARV) control system as described in any one of claims 1 to 7.

Images