Multi-underwater-robot cooperative target tracking control and decision-making method and system

Abstract

The invention provides a multi-underwater-robot cooperative target tracking control and decision-making method and system. The method comprises the following steps of: detecting and identifying an underwater dynamic target and predicting behaviors; a step of high-precision cooperative tracking under a weak communication condition; and performing target tracking control under a weak perception condition. According to the invention, aiming at a deep sea weak perception weak communication environment, an underwater dynamic target detection and identification technology based on multi-underwater robot perception information is researched, and accurate perception and accurate prediction of ocean dynamic target state information are realized. A multi-underwater-robot cooperative tracking technology under a weak communication condition is researched, and the problem of multi-underwater-robot cooperative target tracking due to communication delay and communication bandwidth limitation is solved.

Description

Technical Field

[0001] The invention belongs to the field of tracking control, and in particular relates to a multi-underwater robot collaborative target tracking control and decision-making method and system. Background Art

[0002] In deep-sea exploration and operations, collaborative target tracking by multiple underwater robots (AUVs) is crucial for marine life observation, target identification, and security. However, the deep-sea environment presents challenges such as weak communication and perception, resulting in low efficiency and accuracy for traditional single-AUV target tracking methods. Existing technologies for collaborative target tracking by multiple AUVs rely primarily on high-bandwidth communication and rich sensory information, making them difficult to adapt to the weak information conditions of the deep sea.

[0003] Existing technologies have defects in the field of collaborative target tracking of multiple underwater robots. Traditional target tracking methods rely too much on complete target state information and cannot effectively handle target loss or rapid maneuvers under weak perception conditions. In particular, there is a lack of a rapid response mechanism when switching targets, resulting in tracking lags or even failures. Existing target prediction models are insufficient in predicting the long-term motion trajectories of complex maneuvering targets and are unable to provide reliable tracking information. In addition, existing collaborative algorithms usually assume that the communication environment is real-time and reliable. They neither consider the impact of communication delays and bandwidth limitations in deep-sea environments nor design effective state estimation mechanisms to compensate for information loss, which seriously restricts the collaborative accuracy and robustness of multi-underwater robot systems. These defects together make it difficult for existing technologies to meet the needs of high-precision and high-stability collaborative tracking in complex marine environments. Summary of the Invention

[0004] In light of this, the present invention aims to propose a multi-UUV collaborative target tracking control and decision-making method. This method, based on the perception information of multiple UUVs in the deep sea, aims to accurately perceive the state of dynamic targets in the ocean. The method also addresses the challenges of multi-UUV collaborative tracking under weak communication conditions, such as communication delays and bandwidth limitations.

[0005] The present invention also proposes a multi-underwater robot collaborative target tracking control and decision-making system to implement a multi-underwater robot collaborative target tracking control and decision-making method.

[0006] To achieve the above object, the technical solution created by the present invention is implemented as follows: A multi-underwater robot collaborative target tracking control and decision-making method, the method comprising: Steps for underwater dynamic target detection and recognition; Steps for target tracking control under weak perception conditions.

[0007] Steps for multi-robot collaborative tracking under weak communication conditions.

[0008] Furthermore, the underwater dynamic target detection and recognition specifically utilizes existing ocean dynamic target sample data and trains a deep neural network model to learn the multi-layer structural features and semantic patterns of the target; at the same time, for the perception information of multiple underwater robots, through state consistency matching and time alignment methods, the state association of the same target in the sensor data of different underwater robots is established, and multi-source heterogeneous observation information is integrated.

[0009] Furthermore, training the deep neural network model specifically involves collecting data or using existing marine life or underwater target detection datasets; annotating the collected sample data, assigning a category label to each target instance, and annotating the target's bounding box or key points; Atomic clock devices are used to align the timestamps of sensor data from different underwater robots and synchronize the clocks of multiple robots. For asynchronous sensors, the time delay is adjusted by linear interpolation, as shown in the following formula:

[0010] in, express The status of the underwater robot at all times; Using Kalman filtering to fuse multi-source observations:

[0011] in, Indicates in The estimated state value at time t, Indicates in The prior state estimate at time , is the global Kalman gain, Indicates the number of sensors, For the The observation matrix of sensors, Indicates the The inverse matrix of the observation noise covariance matrix of the sensors, Indicates the The sensor in the Observed value at time.

[0012] Furthermore, the target tracking control under the weak perception conditions is specifically as follows: adaptive tracking is achieved through multi-threshold distance partition control. Specifically, three-level control strategies are divided according to the 15m / 35m distance thresholds: angle segmented smooth steering is adopted for close distances, linear acceleration and target speed compensation are superimposed for medium distances, and full-speed mode is switched to long distances and target speed prediction is integrated; a three-layer correction mechanism is designed for angle control, including 10° single-step limiting and ±5° fine-tuning anti-shake, which effectively balances tracking accuracy and motion stability.

[0013] Furthermore, the target tracking control under the weak perception condition is specifically as follows: the underwater robot's tracking control system obtains the target position information in real time through the sonar device, calculates the relative distance and angle deviation based on the current state of the underwater robot; dynamically adjusts the control strategy according to the distance threshold, and defines the system state vector:

[0014]

[0015] in represents the Euclidean distance between the underwater robot and the target, represents the azimuth deviation, represents the heading angle of the underwater robot, Indicates the current speed of the underwater robot, represents the position coordinates of the underwater robot, Indicates the location coordinates of the target; Assume that the expected speed of the underwater robot is , close range ( <15m): Only control the steering angle and speed Set to 0. Avoid sharp turns through segmented angle correction to ensure stable approach to the target; Medium distance (15m< 35m): Based on the steering control, speed control is introduced. The speed increases linearly with the distance, and the target movement speed is superimposed to match the dynamic target, that is: ; Long distance ( ≥ 35m): Approaching the target at full speed, , while continuously correcting the heading angle. Target speed is estimated through historical distance differences to achieve predictive tracking; The following strategies are used to handle angle deviation: Calculate the target relative azimuth and normalize it to the range; Limit the single-step correction amount (≤10°) to avoid sudden turns; When approaching the target direction (deviation <3°, i.e., PI / 60), the angle is actively fine-tuned (±5°) to suppress oscillation.

[0016] Furthermore, the high-precision collaborative tracking under weak communication conditions is specifically as follows: in a weak communication environment, the communication load is reduced through a collaborative tracking strategy; when the status of other robots cannot be obtained in real time, the motion status of other robots is estimated based on local observations and historical communication data to ensure the accuracy of collaborative tracking.

[0017] Furthermore, the collaborative tracking strategy is specifically: Step S1: Turn on forward-looking sonar detection to determine whether a tracking target is sensed. If so, proceed to step S2. Step S2: The current underwater robot that detects the tracking target starts the tracking mode and determines whether the underwater acoustic communication sending beat has been reached. If so, the tracking information is sent, otherwise, step S3 is performed; Step S3: Determine whether detection information from other underwater robots has been received. If yes, proceed to step S4; otherwise, proceed to step S5. Step S4: Determine whether the currently tracked target is the same target. If so, proceed to step S5; if not, return to step S1 to determine whether the tracked target is sensed. Step S5: Continue to track the detected target.

[0018] Furthermore, the steps for multi-robot collaborative tracking under weak communication conditions are as follows: based on the target's predicted trajectory and each robot's current position, a target allocation decision is made, assigning different robots to track different targets or different locations of the target, thus forming a reasonable collaborative tracking layout; Control the robot's motion based on local limited perception information and prediction results, so that it always stays within the effective range of target tracking; When the target undergoes rapid movement changes or the tracking target switches, the collaborative tracking module responds quickly based on the prediction and perception results, adjusts the division of labor and control strategy, and ensures that the robot group can continuously and stably track the new target to avoid target loss or tracking interruption.

[0019] Furthermore, in the deep-sea weak communication and weak perception environment, when the target undergoes rapid motion changes or the tracking target switches, the underwater robot's dynamic model is first used in combination with its own sensor data to make real-time predictions of its own and other robots' motion states. Its mathematical model can be expressed as:

[0020] in, For robots exist The state prediction value at the moment, is a nonlinear dynamic model, is the control input, is the process noise; Subsequently, local observation data of other robots are obtained through sonar and underwater acoustic communication equipment and integrated with the prediction results; the extended Kalman filter algorithm is used to process the nonlinear observation model:

[0021] in, is the observation model, is the observation noise. The estimated value is corrected in real time through the state update equation:

[0022] in, is the Kalman gain matrix; To solve the communication delay problem, a timestamp mechanism is introduced to receive delayed state data. Perform status backtracking:

[0023] This ensures the timeliness of the estimation results and the accuracy of collaborative control; the state estimator achieves reliable estimation of the states of multiple robots under weak perception conditions by fusing model predictions and local observations, providing key information support for collaborative control.

[0024] A multi-underwater robot collaborative target tracking control and decision-making system, the system using the multi-underwater robot collaborative target tracking control and decision-making method as described above, the system comprising: Steps for underwater dynamic target detection and recognition; Steps for target tracking control under weak perception conditions; Steps for multi-robot collaborative tracking under weak communication conditions.

[0025] Compared with the prior art, the present invention has the following advantages: The present invention studies the target tracking control technology of underwater robots that can cope with the maneuvering of ocean targets based on limited perception information, so as to achieve stable and accurate tracking of ocean moving targets by underwater robots.

[0026] The present invention enables multiple underwater robots to collaboratively track ocean moving targets intelligently, stably and efficiently when performing deep-sea tracking tasks.

[0027] The present invention's accurate identification of marine dynamic targets and stable, precise coordinated target tracking control are key to multi-underwater robot target tracking. Accurate perception of marine dynamic targets in deep-sea, weak-information environments is extremely difficult. A model learning technique for underwater dynamic targets, utilizing deep neural networks to continuously learn the target's multi-layered structure and semantic feature patterns, is proposed, ultimately achieving accurate identification of marine dynamic targets. During marine dynamic target tracking and observation, the targets tracked by multiple underwater robots change as the target formation maintains and changes. In weak communication and weak perception environments, it is extremely difficult for underwater robots to rapidly track and switch targets using limited information. In weak communication environments, underwater acoustic communication delays and limited communication bandwidth impact real-time state information interaction, making stable, precise coordinated tracking control extremely difficult. A target tracking controller for underwater robots is proposed, which allows for rapid convergence of tracking errors taking into account communication delays, ultimately achieving rapid response and high-precision control of multi-target tracking by multiple underwater robots. The present invention significantly improves target switching response speed and tracking stability through a hierarchical distance-angle control strategy and a weak-perception compensation mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings, which constitute part of the present invention, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of dynamic target tracking of multiple underwater robots of the present invention.

[0029] Figure 2 This is a collaborative tracking decision logic allocation diagram of the present invention.

[0030] Figure 3 This is the technical roadmap for the detection, recognition and behavior prediction of underwater dynamic targets of the present invention.

[0031] Figure 4 This is the technical roadmap for high-precision collaborative control of multiple underwater robots under weak communication conditions of the present invention.

[0032] Figure 5 This is the technical roadmap for target tracking and control of underwater robots under weak perception and weak communication conditions of the present invention.

[0033] Figure 6 These are simulation verification diagrams of the tracking effect of a single underwater robot according to the present invention, where (a) is a schematic diagram of the tracking of a single underwater robot, and (b) is a schematic diagram of the tracking bow direction change of a single underwater robot.

[0034] Figure 7 This is a simulation diagram of multiple underwater robots collaboratively tracking a target according to the present invention. DETAILED DESCRIPTION

[0035] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0036] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0037] Implementation Method 1 This embodiment provides a method for collaborative target tracking, control, and decision-making for multiple underwater robots. This method combines core technologies such as online learning of deep neural networks, sensory information fusion, maneuvering target behavior prediction, high-precision tracking control, disturbance estimation compensation, and robust control of communication delays to enhance the target perception capabilities and collaborative tracking control accuracy of multiple underwater robots in weak information environments. The method includes: Steps for underwater dynamic target detection, recognition and behavior prediction; Steps for high-precision collaborative tracking under weak communication conditions; Steps for target tracking control under weak perception conditions.

[0038] Combine Figure 1 , introduces the process of multi-underwater robot collaborative target tracking method: Step 1: Start the navigation mission and synchronously start the sonar equipment, underwater acoustic communication device, nonlinear estimation module and collaborative formation control module of each underwater robot.

[0039] Step 2: Each underwater robot uses sonar equipment to collect ocean dynamic target information, extracts the target's multi-layer structure and semantic features through deep neural networks, and integrates heterogeneous data through perception information fusion technology. Step 3: Based on the joint long short-term memory network, the historical trajectory and multi-source perception data are integrated to predict the future motion state of the target.

[0040] Step 4: When communication delays and bandwidth are limited, event-triggered distributed model predictive control is used to prioritize the transmission of critical state information. The underwater robot state estimator predicts the states of neighboring robots in real time, compensating for information loss and maintaining collaborative accuracy.

[0041] Step 5: Combine the prediction results with the current position of each robot, assign tracking tasks through a distributed path planning algorithm, and form a collaborative formation layout.

[0042] Step 6: A nonlinear model predictive controller generates optimal motion commands. An extended Kalman filter estimates environmental disturbances such as ocean currents and turbulence in real time and dynamically compensates for them in the controller input. If the target changes suddenly or switches, a rapid response mechanism is triggered to reallocate tracking tasks and adjust the control strategy.

[0043] Step 7: Continuously monitor the target tracking status. If the estimated data converges, end the task; otherwise, return to step 2 and iteratively optimize the perception, prediction, and control processes.

[0044] Furthermore, the underwater dynamic target detection, recognition and behavior prediction are specifically as follows: Start the navigation mission and start the sonar equipment, underwater acoustic communication equipment, control module, etc.

[0045] Each underwater robot uses sonar equipment to autonomously detect and identify dynamic targets in the ocean.

[0046] The target information collected by multiple underwater robots is integrated through information fusion algorithm to enhance the integrity and accuracy of target state estimation.

[0047] Furthermore, a deep neural network model is trained to learn the multi-layered structural features and semantic patterns of targets. By introducing an unsupervised online learning strategy, the accuracy and generalization of target detection and recognition are continuously improved. During operation, the system combines perception confidence with a historical consistency strategy to dynamically select high-quality samples and update neural network parameters in real time, achieving adaptive evolution and long-term stable operation of the model in complex marine environments. Furthermore, a multi-underwater robot perception information fusion mechanism is designed. Through state consistency matching and spatiotemporal alignment methods, the state association of the same target in the sensor data of different underwater robots is established, fusing multi-source heterogeneous observation information to enhance the overall target perception accuracy, stability, and robustness of the system.

[0048] Furthermore, a joint prediction model for target behavior is constructed, and a long-short-term memory network structure is introduced to integrate historical trajectory, speed status and environmental data to complete multimodal joint prediction of the target's future state and potential behavior, providing high-quality dynamic target perception and prediction information support for the system to achieve collaborative tracking control in weak perception and weak communication environments.

[0049] Furthermore, the high-precision collaborative tracking under weak communication conditions is specifically as follows: in a weak communication environment, the communication load is reduced through a collaborative tracking strategy; when the status of other robots cannot be obtained in real time, the motion status of other robots is estimated based on local observations and historical communication data to ensure the accuracy of collaborative tracking.

[0050] Furthermore, the impact of weak communication environments on the coordinated tracking of multiple underwater robots is analyzed, and a communication delay and bandwidth limitation model is established. A controller for coordinated tracking of multiple underwater robots is designed to reduce the impact of communication delay on the stability and accuracy of coordinated tracking control. Considering the limited communication bandwidth of deep-sea systems, a controller for coordinated tracking of multiple underwater robots is designed to utilize limited communication resources. Considering the difficulty of obtaining real-time state information of other underwater robots, an underwater robot state estimator is constructed to estimate the state information of other underwater robots in real time. A coordinated tracking control method for multiple underwater robots, combined with an underwater robot state estimator for the conditions of communication delay and limited communication bandwidth, is developed to achieve high-precision coordinated control of multiple underwater robots under weak communication conditions.

[0051] Furthermore, the underwater dynamic target detection and recognition specifically utilizes existing ocean dynamic target sample data and trains a deep neural network model to learn the multi-layer structural features and semantic patterns of the target; at the same time, for the perception information of multiple underwater robots, through state consistency matching and time alignment methods, the state association of the same target in the sensor data of different underwater robots is established, and multi-source heterogeneous observation information is integrated.

[0052] Furthermore, training deep neural network models involves collecting data in real ocean environments using devices such as sonar or leveraging existing datasets for detecting marine life or underwater targets. Sample data of dynamic marine targets, such as marine mammals, underwater robots, and frogmen, is collected and labeled, with each target instance assigned a category label (e.g., "whale," "underwater robot," etc.). The target's bounding box or key points are also annotated so that the model can learn the target's shape, position, and posture. Atomic clock devices are used to align the timestamps of sensor data from different underwater robots and synchronize the clocks of multiple robots. For asynchronous sensors, the time delay is adjusted by linear interpolation, as shown in the following formula:

[0053] in, express The status of the underwater robot at all times; Using Kalman filtering to fuse multi-source observations:

[0054] in, Indicates in The estimated state value at time t, Indicates in The prior state estimate at time , is the global Kalman gain, Indicates the number of sensors, For the The observation matrix of sensors, Indicates the The inverse matrix of the observation noise covariance matrix of the sensors, Indicates the The sensor in the Observed value at time.

[0055] Furthermore, the target tracking control under the weak perception conditions is specifically as follows: adaptive tracking is achieved through multi-threshold distance partition control. Specifically, three-level control strategies are divided according to the 15m / 35m distance thresholds: angle segmented smooth steering is adopted for close distances, linear acceleration and target speed compensation are superimposed for medium distances, and full-speed mode is switched to long distances and target speed prediction is integrated; a three-layer correction mechanism is designed for angle control, including 10° single-step limiting and ±5° fine-tuning anti-shake, which effectively balances tracking accuracy and motion stability.

[0056] Furthermore, the target tracking control under the weak perception condition is specifically as follows: the existing underwater robot tracking control system obtains the target position information in real time through the sonar equipment, calculates the relative distance and angle deviation based on the current state of the underwater robot; dynamically adjusts the control strategy according to the distance threshold (15m, 35m), and defines the system state vector:

[0057]

[0058] in represents the Euclidean distance between the underwater robot and the target, represents the azimuth deviation, represents the heading angle of the underwater robot, Indicates the current speed of the underwater robot, represents the position coordinates of the underwater robot, Indicates the location coordinates of the target; Assume that the expected speed of the underwater robot is , close range ( <15m): Only control the steering angle and speed Set to 0. Avoid sharp turns by performing segmented angle correction (PI / 18 as the step size) to ensure a stable approach to the target; Medium distance (15m< 35m): Based on the steering control, speed control is introduced. The speed increases linearly with the distance, and the target movement speed is superimposed to match the dynamic target, that is: ; Long distance ( ≥ 35m): Approaching the target at full speed, , while continuously correcting the heading angle. Target speed is estimated through historical distance differences to achieve predictive tracking; The following strategies are used to handle angle deviation: Calculate the target relative azimuth and normalize it to the range; Limit the single-step correction amount (≤10°) to avoid sudden turns; When approaching the target direction (deviation <3°, i.e., PI / 60), the angle is actively fine-tuned (±5°) to suppress oscillation.

[0059] Furthermore, the high-precision collaborative tracking under weak communication conditions is specifically as follows: in a weak communication environment, the communication load is reduced through a collaborative tracking strategy; when the status of other robots cannot be obtained in real time, the motion status of other robots is estimated based on local observations and historical communication data to ensure the accuracy of collaborative tracking.

[0060] Furthermore, the collaborative tracking strategy is specifically as follows: Figure 2 As shown, Step S1: Turn on forward-looking sonar detection to determine whether a tracking target is sensed. If so, proceed to step S2. Step S2: The current underwater robot that detects the tracking target starts the tracking mode and determines whether the underwater acoustic communication sending beat has been reached. If so, the tracking information is sent, otherwise, step S3 is performed; Step S3: Determine whether detection information from other underwater robots has been received. If yes, proceed to step S4; otherwise, proceed to step S5. Step S4: Determine whether the currently tracked target is the same target. If so, proceed to step S5; if not, return to step S1 to determine whether the tracked target is sensed. Step S5: Continue to track the detected target.

[0061] Among them, the specific tracking task allocation decision is as follows: Figure 2 As shown in the following pseudo code: Collaborative tracking allocation algorithm Start formation tracking task() While the task is not completed: If the forward-looking sonar detects a target: Enable target tracking mode() If the underwater acoustic communication sending beat is reached: SendTrackingInfo() If other AUV detection information is received: If it is the same target as the current tracking target: Fusion target information () else: Continue to track detected targets () else: Continue searching for the target() EndTask() Furthermore, the target tracking control under the weak perception condition is specifically as follows: according to the target predicted trajectory and the current position of each robot, a target allocation decision is executed, and different robots are assigned to track different targets or different orientation areas of the target to form a reasonable collaborative tracking layout; Control the robot's motion based on local limited perception information and prediction results, so that it always stays within the effective range of target tracking; When the target undergoes rapid movement changes or the tracking target switches, the collaborative tracking module responds quickly based on the prediction and perception results, adjusts the division of labor and control strategy, and ensures that the robot group can continuously and stably track the new target to avoid target loss or tracking interruption.

[0062] Furthermore, in the deep-sea environment with weak communication and perception, underwater robots face the dual challenges of insufficient information acquisition and limited system coordination when performing target tracking and control tasks. To address this issue of limited perception information, the system designed a high-precision controller that enables each underwater robot to autonomously complete target identification and tracking tasks based on local perception data, maintaining stable and continuous tracking of the target despite sparse perception information and delayed updates. Given the frequent changes in motion patterns, sudden maneuvers, or position shifts of dynamic targets in the ocean, the system thoroughly analyzes key issues such as tracking interruption and path reconstruction that may occur during target switching. A fast-response target switching tracking control strategy is proposed, enabling the underwater robot to rapidly adjust its tracking behavior after identifying a target change, ensuring mission continuity and stability. Furthermore, the system fully considers the influence of external disturbances such as turbulence, ocean currents, and buoyancy changes in the deep-sea environment. A targeted ocean disturbance modeling system is constructed, and a disturbance estimator and disturbance compensator are designed based on the time-varying characteristics of the disturbance. This system actively compensates for environmental disturbances by incorporating them into the control feedback loop, thereby improving the underwater robot's tracking accuracy and overall system stability for dynamic targets in complex ocean conditions.

[0063] In the deep-sea environment with weak communication and weak perception, when the target undergoes rapid motion changes or the tracking target switches, a state estimator based on the fusion of model prediction and local observation is designed. The estimator first uses the dynamic model of the underwater robot combined with its own sensor data to predict the motion state of itself and other robots in real time. Its mathematical model can be expressed as:

[0064] in, For robots exist The predicted state value at the moment, is a nonlinear dynamic model, is the control input, is the process noise; Subsequently, local observation data of other robots are obtained through sonar and underwater acoustic communication equipment and integrated with the prediction results; the extended Kalman filter algorithm is used to process the nonlinear observation model:

[0065] in, is the observation model, is the observation noise. The estimated value is corrected in real time through the state update equation:

[0066] in, is the Kalman gain matrix; To solve the communication delay problem, a timestamp mechanism is introduced to receive delayed state data. Perform status backtracking:

[0067] This ensures the timeliness of the estimation results and the accuracy of collaborative control; the state estimator achieves reliable estimation of the states of multiple robots under weak perception conditions by fusing model predictions and local observations, providing key information support for collaborative control.

[0068] Combine Figure 3 The technical route for underwater dynamic target detection, recognition, and behavior prediction is shown in the figure. This involves extracting the target's multi-layered structure and semantic feature patterns using a large amount of ocean target sample data. Deep neural networks are then used to enhance feature generalization. A multi-level autonomous screening mechanism is then used to select high-quality samples to continuously update network parameters and improve recognition performance. The ocean target perception information collected by multiple underwater robots is further integrated with their own state information to form the perception input for dynamic ocean targets. Continuous detection and recognition of dynamic ocean targets is achieved through an unsupervised online learning mechanism. Subsequently, this perception information and the state information of multiple underwater robots are input into a joint long-short-term memory network to predict the target's future behavior.

[0069] Combine Figure 4 By combining the collaborative tracking controller with communication delay and the collaborative tracking controller considering bandwidth limitation with the underwater robot state estimator to compensate for information loss, a complete high-precision collaborative tracking technology for multiple underwater robots under weak communication conditions is formed, realizing stable collaborative tracking operations in communication-restricted environments.

[0070] Combine Figure 5The system obtains ocean dynamic target information in a deep-sea weak-perception environment and constructs a target model under weak-perception conditions based on limited perception information. Subsequently, the underwater robot uses a high-precision controller with limited perception information, combined with a complex ocean environment interference model and an interference estimator and compensator, to achieve stable tracking of dynamic targets. When an ocean dynamic target changes, the system triggers a fast-response tracking controller after target switching, ensuring continuous and precise tracking and control of the target under weak-perception and weak-communication conditions.

[0071] Combine Figure 6 Simulations verifying the single-AUV tracking algorithm demonstrate the effectiveness of the tiered strategy, demonstrating that the system dynamically switches control modes based on target distance (steering priority at close range, speed overlay at mid-range, and full-speed tracking at long range). Tests demonstrate that within the critical range of 15m-35m, the system smoothly transitions control commands, avoiding sudden changes in speed or steering and ensuring tracking stability. Comparative experiments confirm that angle correction limits (≤10°) significantly reduce steering oscillations, while fine-tuning compensation (±5°) keeps heading error within 3° during the final approach phase.

[0072] Combine Figure 7 As shown in collaborative tracking simulation results, three AUVs (AUV1-3) ​​successfully achieved stable tracking and task allocation for multiple dynamic targets. Within the test range of 0-1000 meters on the Y axis, AUVs 1-3 autonomously allocated tracking targets using a distributed collaborative algorithm, forming a complementary observation network. AUV1, shown in the red path, served as the master AUV. Besides observing and tracking the target, it also tracked the target according to pre-set path points. AUVs 2 and 3 served as slave AUVs, receiving information from the master AUV1 and sailing in formation. Each AUV's trajectory exhibited intelligent target tracking and path optimization. When Target 2 suddenly maneuvered into AUV1's sensing range, the system automatically switched tracking authority (AUV1 took over Target 2) through an event-triggered mechanism. Simultaneously, AUV 2 quickly took over and tracked Target 1's position. The entire process was efficient and did not interrupt tracking of other targets. Simulation data demonstrates that the system maintains a high target tracking rate through dynamic priority adjustment and load balancing, validating the robustness and real-time performance of the collaborative allocation algorithm in weak communication environments.

[0073] The core algorithm of the target tracking method includes the following three aspects: 1. Core Logic of Target Tracking The system obtains target position information in real time through sonar equipment and calculates relative distance and angle deviation based on the current state of the underwater robot. The control strategy is dynamically adjusted according to the distance threshold (15m, 35m): Close range (<15m): Only steering angle is controlled, with speed set to 0. Segmented angle correction (PI / 18 steps) is used to avoid abrupt turns and ensure stable approach to the target. Medium range (15m-35m): Speed ​​control is introduced in addition to steering control. Speed ​​increases linearly with distance, and the target's motion speed is superimposed to match dynamic targets. Long range (≥35m): Approach the target at full speed (0.5m / s) while continuously correcting the heading angle. Target speed is estimated based on historical distance differences, enabling predictive tracking.

[0074] 2. Angle correction and anti-shake mechanism The following strategies are used to handle angle deviation: Calculate the target's relative azimuth and normalize it to the range. Limit the single-step correction (≤10°) to avoid sudden turns. When approaching the target (deviation <3°, i.e., PI / 60), actively fine-tune the angle (±5°) to suppress oscillation.

[0075] 3. Prediction and compensation under weak perception conditions Weak Perception Impact: If the target is temporarily lost, the system maintains its last state (speed = 0, heading unchanged). A state estimator (such as the extended Kalman filter) fills the perception gap and combines communication data from other AUVs (the underwater acoustic communication clock in the figure) to correct the target position. Prediction Method: Target velocity is estimated based on historical range differences and 0.1s intervals for feedforward control.

[0076] Implementation Method 2 This embodiment provides a multi-underwater robot collaborative target tracking control and decision-making system, which uses the multi-underwater robot collaborative target tracking control and decision-making method described in Embodiment 1. The system includes: Module for underwater dynamic target detection, recognition and behavior prediction; A module for high-precision collaborative tracking under weak communication conditions; Module for target tracking control under weak perception conditions.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for cooperative target tracking, control, and decision-making of multiple underwater robots, characterized by: The method comprises: Steps for underwater dynamic target detection and recognition; Steps for target tracking control under weak perception conditions. Steps for multi-robot collaborative tracking under weak communication conditions.

2. The method according to claim 1, wherein: The underwater dynamic target detection and recognition method specifically utilizes existing ocean dynamic target sample data and trains a deep neural network model to learn the multi-layer structural features and semantic patterns of the target; at the same time, for the perception information of multiple underwater robots, through state consistency matching and time alignment methods, the state association of the same target in the sensor data of different underwater robots is established, and multi-source heterogeneous observation information is integrated.

3. The method according to claim 2, wherein: Training a deep neural network model involves collecting data or using existing marine life or underwater target detection datasets; annotating the collected sample data, assigning a category label to each target instance, and annotating the target's bounding box or key points; Atomic clock devices are used to align the timestamps of sensor data from different underwater robots and synchronize the clocks of multiple robots. For asynchronous sensors, the time delay is adjusted by linear interpolation, as shown in the following formula: in, express The status of the underwater robot at all times; Using Kalman filtering to fuse multi-source observations: in, Indicates in The estimated state value at time t, Indicates in The prior state estimate at time , is the global Kalman gain, Indicates the number of sensors, For the The observation matrix of sensors, Indicates the The inverse matrix of the observation noise covariance matrix of the sensors, Indicates the The sensor in the Observed value at time.

4. The method according to claim 1, wherein: The target tracking control under the weak perception condition is specifically: adaptive tracking is achieved through multi-threshold distance partition control. Specifically, three-level control strategies are divided according to the 15m / 35m distance threshold: angle segmented smooth steering is used for close distance, linear acceleration and target speed compensation are superimposed for medium distance, and full speed mode is switched to integrate target speed prediction for long distance; A three-layer correction mechanism has been designed for angle control, including 10° single-step limiting and ±5° fine-tuning for anti-shake, which effectively balances tracking accuracy and motion stability.

5. The method according to claim 4, characterized in that: The target tracking control under the weak perception condition is specifically as follows: the underwater robot's tracking control system obtains the target position information in real time through the sonar device, calculates the relative distance and angle deviation based on the current state of the underwater robot; dynamically adjusts the control strategy according to the distance threshold, and defines the system state vector: in represents the Euclidean distance between the underwater robot and the target, represents the azimuth deviation, represents the heading angle of the underwater robot, Indicates the current speed of the underwater robot, represents the position coordinates of the underwater robot, Indicates the location coordinates of the target; Assume that the expected speed of the underwater robot is , close range ( < 15m): Only control the steering angle and speed Set to 0. Avoid sharp turns through segmented angle correction to ensure stable approach to the target; Medium distance (15m < 35m): Based on the steering control, speed control is introduced. The speed increases linearly with the distance, and the target movement speed is superimposed to match the dynamic target, that is: ; Long distance ( ≥ 35m): Approaching the target at full speed, , while continuously correcting the heading angle. Target speed is estimated through historical distance differences to achieve predictive tracking; The following strategies are used to handle angle deviation: Calculate the target relative azimuth and normalize it to the range; Limit the single-step correction amount (≤10°) to avoid sudden turns; When approaching the target direction (deviation <3°, i.e., PI / 60), the angle is actively fine-tuned (±5°) to suppress oscillation.

6. The method according to claim 1, wherein: The high-precision collaborative tracking under weak communication conditions is specifically as follows: in a weak communication environment, the communication load is reduced through a collaborative tracking strategy; when the status of other robots cannot be obtained in real time, the motion status of other robots is estimated based on local observations and historical communication data to ensure the accuracy of collaborative tracking.

7. The method according to claim 1, wherein: The collaborative tracking strategy is specifically: Step S1: Turn on forward-looking sonar detection to determine whether a tracking target is sensed. If so, proceed to step S2. Step S2: The current underwater robot that detects the tracking target starts the tracking mode and determines whether the underwater acoustic communication sending beat has been reached. If so, the tracking information is sent, otherwise, step S3 is performed; Step S3: Determine whether detection information from other underwater robots has been received. If yes, proceed to step S4; if not, proceed to step S5. Step S4: Determine whether the currently tracked target is the same target. If so, proceed to step S5; if not, return to step S1 to determine whether the tracked target is sensed. Step S5: Continue to track the detected target.

8. The method according to claim 1, wherein: The specific steps of multi-robot collaborative tracking under weak communication conditions are as follows: based on the target's predicted trajectory and the current position of each robot, a target allocation decision is made, and different robots are assigned to track different targets or different locations of the target, forming a reasonable collaborative tracking layout; Control the robot's motion based on local limited perception information and prediction results, so that it always stays within the effective range of target tracking; When the target undergoes rapid movement changes or the tracking target switches, the collaborative tracking module responds quickly based on the prediction and perception results, adjusts the division of labor and control strategy, and ensures that the robot group can continuously and stably track the new target to avoid target loss or tracking interruption.

9. The method according to claim 8, characterized in that: In the deep-sea environment with weak communication and perception, when the target undergoes rapid motion changes or the tracking target switches, the underwater robot's dynamic model is first used in combination with its own sensor data to make real-time predictions of its own and other robots' motion states. The mathematical model can be expressed as: in, For robots exist The predicted state value at the moment, is a nonlinear dynamic model, is the control input, is the process noise; Subsequently, local observation data of other robots are obtained through sonar and underwater acoustic communication equipment and integrated with the prediction results; the extended Kalman filter algorithm is used to process the nonlinear observation model: in, is the observation model, is the observation noise. The estimated value is corrected in real time through the state update equation: in, is the Kalman gain matrix; To solve the communication delay problem, a timestamp mechanism is introduced to receive delayed state data. Perform status backtracking: This ensures the timeliness of the estimation results and the accuracy of collaborative control; the state estimator achieves reliable estimation of the states of multiple robots under weak perception conditions by fusing model predictions and local observations, providing key information support for collaborative control.

10. A multi-underwater robot collaborative target tracking control and decision-making system, characterized by: The system uses the multi-underwater robot collaborative target tracking control and decision-making method according to any one of claims 1 to 9, and the system includes: Steps for underwater dynamic target detection and recognition; Steps for target tracking control under weak perception conditions; Steps for multi-robot collaborative tracking under weak communication conditions.

Images