Collaborative operation following method and device for manta ray aircraft and medium

Abstract

The invention discloses a collaborative operation following method and device for a manta ray aircraft and a medium, and relates to the technical field of attitude control, and the method mainly comprises the steps: extracting an appearance feature vector of a pilot, and carrying out the dynamic state prediction of the pilot based on a Kalman filter; performing cosine similarity calculation between the detection target and the following target according to the appearance feature vector, and constructing a matching cost matrix; performing optimal matching on the matching cost matrix by adopting a Hungary algorithm, and generating a matching result of the detection target and the following target; updating the dynamic state of the track of the corresponding following target according to the matching result of successful matching, and obtaining the position, speed and azimuth information of the pilot through analysis; and controlling the frequency and amplitude of pectoral fins of a follower through a PID (Proportion Integration Differentiation) algorithm to carry out dynamic following according to the position, speed and azimuth information of the pilot obtained through analysis and in combination with a category label. The method can analyze the accurate azimuth information of the following target in real time, and is suitable for various complex scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of attitude control, and in particular to a collaborative operation following method, device and medium for a manta ray vehicle. Background Art

[0002] The marine environment contains abundant renewable energy, biological resources and mineral reserves, but its complex underwater environment, such as variable water flow speed, direction and unpredictable interference factors, places extremely high demands on the autonomous operation capability of the vehicle. As a new type of underwater equipment, the manta ray-like vehicle has both efficient propulsion and low disturbance characteristics by imitating the sliding and fluttering motion mechanism of manta rays, and can adapt to the observation and exploration tasks in complex marine environments. However, in the scenario of multi-vessel collaborative operation, the traditional following method relies on the sensor ranging and navigation control of the propeller propulsion system, which is difficult to match the flexible motion characteristics of the bionic vehicle, resulting in problems such as distance inaccuracy, heading deviation or target loss during the following process. Existing target tracking algorithms (such as single sensor fusion based on filters) are not robust enough in dynamic underwater environments, cannot analyze the precise motion information of the target in real time, and lack control adaptability to the bionic propulsion device. Therefore, there is an urgent need for a following algorithm that integrates high-precision target detection, dynamic tracking and bionic motion control to achieve stable collaborative operation of manta ray-like vehicles in complex marine environments. Summary of the invention

[0003] To address the deficiencies of the prior art, the present invention proposes a collaborative operation following method for manta ray vehicles, based on a logical framework combining DeepSORT target detection and target tracking, comprising the following steps: S1: Use the YOLOv8 algorithm to detect the real-time inter-frame image of the navigator, identify the bounding box and category label of each detection target on the navigator, and generate the detection result; S2: Extract the appearance feature vector of the navigator based on the detection results, and predict the dynamic state of the navigator based on the Kalman filter; S3: Calculate the cosine similarity between the detected target and the followed target based on the appearance feature vector, and construct a matching cost matrix; S4: Use the Hungarian algorithm to optimally match the matching cost matrix and generate the matching results of the detected target and the followed target; S5: updating the dynamic state of the corresponding target trajectory according to the matching result, and obtaining the position, speed and orientation information of the navigator through analysis; S6: The position, speed and orientation information of the navigator are obtained according to the analysis, combined with the category label, and the frequency and amplitude of the pectoral fin of the follower are controlled by the PID algorithm to perform dynamic following.

[0004] Furthermore, in the step S1, the YOLOv8 algorithm is specifically: Divide the real-time inter-frame image of the navigator into grid cells of a preset size; Extracting features of each detection target from the inter-frame image through a deep convolutional neural network and outputting a category label, wherein the features of the detection target include morphological features of the pectoral fin, the caudal fin, and the dorsal fin; Predict bounding box parameters for each grid cell and eliminate overlapping bounding boxes through non-maximum suppression; The final bounding box and the corresponding category label are output as the detection result.

[0005] Furthermore, in the step S2, the appearance feature vector is extracted according to the detection result through the ReID network.

[0006] Furthermore, in the step S2, the dynamic state prediction of the navigator based on the Kalman filter is specifically: using the Kalman filter to predict the state vector of the navigator, and estimating the motion state of the navigator by minimizing the covariance, the state vector including position and speed.

[0007] Furthermore, in the step S5, the position, speed and orientation information of the navigator is obtained by analysis as follows: the position and speed of the detection target are predicted according to the dynamic state through the Kalman filter; according to the pixel coordinate changes of the detection target, the pixel coordinates are mapped into three-dimensional space coordinates through the camera intrinsic parameter matrix, and the heading and orientation information is calculated in combination with the speed prediction result of the Kalman filter.

[0008] Furthermore, in the step S3, the matching cost matrix is ​​constructed by the following formula: In the formula, To detect the target Follow the target The matching cost matrix is To detect the target The appearance feature vector of To follow the target The appearance feature vector of To detect the target and follow the target The cosine similarity between .

[0009] Furthermore, in step S6, the PID algorithm adopts feedforward-feedback compound control, and the formula is expressed as follows: In the formula, is the distance control output, is the distance error at the previous moment, For speed control output, To analyze and obtain the speed of the navigator, To dynamically adjust the parameters according to the environment, is the distance error at the current moment, and T is the sampling period.

[0010] Furthermore, in step S6, the adjustment logic of the pectoral fin frequency and amplitude is: When the distance error exceeds the set threshold, the amplitude and frequency of the pectoral fins are increased to accelerate; when the distance error is lower than the set threshold, the amplitude and frequency of the pectoral fins are reduced or the amplitude is reset to zero; when the heading deviation exceeds the preset angle, steering is achieved by adjusting the amplitude of the unilateral pectoral fin.

[0011] The present invention also includes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the collaborative operation following method for manta ray vehicles.

[0012] Also included is a device for processing data, comprising: a memory having a computer program stored thereon; A processor is used to execute the computer program in the memory to implement the steps of the collaborative operation following method for manta ray vehicles.

[0013] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The collaborative operation following method for manta ray vehicles proposed in the present invention is based on the DeepSORT framework. By extracting multi-view appearance feature vectors and combining the Kalman filter to dynamically predict the target state, the interference of water flow disturbance on tracking is effectively overcome. The matching cost matrix is ​​constructed by cosine similarity and the Hungarian algorithm is used to optimize the association matching, so that the accuracy of target tracking is improved and the phenomenon of target loss or mismatch is significantly reduced. (2) Through the feedforward-feedback composite PID control algorithm, the real-time speed of the leader is used as the feedforward input, and the frequency and amplitude of the follower's pectoral fin are dynamically adjusted in combination with the distance error feedback to achieve fast response and precise following. (3) Through the Kalman filter combined with the minimum noise covariance matrix, dynamic adjustment according to the water flow disturbance and three-dimensional spatial coordinate mapping technology, it can analyze the precise position information of the target in real time and adapt to complex scenarios with sudden changes in flow velocity or irregular direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A step diagram of a collaborative operation following method for manta ray vehicles. DETAILED DESCRIPTION

[0015] The following are specific embodiments of the present invention and the accompanying drawings to further describe the technical solution of the present invention, but the present invention is not limited to these embodiments.

[0016] In the complex and ever-changing marine environment, the collaborative operation of bionic vehicles faces core challenges such as poor target tracking stability and low motion control accuracy. Traditional methods rely on the fusion of rigid propulsion systems and single sensors, which are difficult to adapt to the flexible gliding motion characteristics of manta ray-like vehicles, especially in scenes with strong water flow interference or target occlusion, which are prone to tracking loss or control lag. To solve the above technical bottlenecks, such as Figure 1 As shown, the present invention proposes a collaborative operation following method for manta ray vehicles, based on a logical framework combining DeepSORT target detection and target tracking, comprising the steps of: S1: Use the YOLOv8 algorithm to detect the real-time inter-frame image of the navigator, identify the bounding box and category label of each detection target on the navigator, and generate the detection result; S2: Extract the appearance feature vector of the navigator based on the detection results, and predict the dynamic state of the navigator based on the Kalman filter; S3: Calculate the cosine similarity between the detected target and the followed target based on the appearance feature vector, and construct a matching cost matrix; S4: Use the Hungarian algorithm to optimally match the matching cost matrix and generate the matching results of the detected target and the followed target; S5: updating the dynamic state of the corresponding target trajectory according to the matching result, and obtaining the position, speed and orientation information of the navigator through analysis; S6: Based on the real-time position, speed and orientation information obtained through analysis and combined with the category label, the frequency and amplitude of the follower's pectoral fins are controlled through the PID algorithm to perform dynamic following.

[0017] In the present invention, the input real-time inter-frame image of the pilot needs to be divided into S×S grid units (the S value is dynamically adjusted according to the image resolution). Each grid unit is responsible for detecting the target object whose center point is located in the unit to ensure the adaptability of the detection granularity and image details. Subsequently, the deep convolutional neural network Darknet53 or its deeper variants are used as the feature extraction backbone to perform multi-scale feature learning on the image: the shallow detail features and deep semantic information are fused through the feature pyramid network (FPN) and path aggregation (PAN) structure, focusing on extracting key biomimetic features such as the pectoral fin swing morphology, tail fin propulsion trajectory and dorsal fin contour of the pilot, providing high-resolution feature representation for subsequent target positioning.

[0018] For each grid cell, the algorithm predicts a set of bounding box parameters ,in represents the normalized coordinates of the center point of the bounding box relative to the grid cell, are the width and height of the bounding box, is the confidence level, and the calculation formula is: in, is the probability of an object appearing in the bounding box, and is the intersection over union ratio between the predicted bounding box and the true bounding box. This formula reflects the probability of the target existing and the positioning accuracy of the predicted box. Finally, the overlapping bounding boxes are screened by the non-maximum suppression (NMS) algorithm: the predicted box with the highest confidence is retained, while the redundant boxes whose intersection over union ratio (IoU) exceeds the set threshold are suppressed to ensure that the output detection results have both high recall rate and low false positive rate. This step ultimately generates a structured detection result containing bounding box coordinates, category labels, and confidence, providing accurate input for the subsequent target tracking module.

[0019] Based on the results of real-time inter-frame image detection, the present invention here uses a target tracking method based on DeepSORT as the main framework. First, the ReID (re-identification) network is used to extract multi-view appearance features of the detected navigator. Among them, the ReID network uses the bounding box image in the detection result as input, and extracts high-dimensional feature vectors through a convolutional neural network (such as ResNet50 or a customized lightweight architecture). Its training data covers the pectoral fin motion sequence images of manta ray-like vehicles under multi-view and multi-illumination conditions to ensure that the features are highly robust to posture changes and local occlusions. During the feature extraction process, the ReID network focuses on capturing dynamic biological features such as the pectoral fin swing phase, the tail fin propulsion cycle, and the dorsal fin contour texture, and generates an appearance descriptor with unique discrimination, providing a reliable basis for subsequent target matching.

[0020] Subsequently, the Kalman filter is used to predict and update the dynamic state of the pilot. In the present invention, the state vector of the Kalman filter is defined as , contains the position of the target in the two-dimensional plane and speed The prediction process follows the following formula: in, is the predicted state at time t, is the predicted state covariance matrix, is the state transfer matrix, which is derived based on the kinematic model of the manta ray vehicle. The state vector includes position, velocity and time step. is the control input matrix, which is designed according to the influence of the control variable on the state (for example, if the control input is acceleration , converting acceleration into velocity change and position changes , is the control vector, the control instruction from the vehicle (such as the steering gear rotation angle), T is the matrix transpose operation, The noise amplitude is set according to the dynamic characteristics of the vehicle (such as maximum acceleration jitter) and dynamically adjusted according to the real-time water flow disturbance intensity. When a sudden change in flow velocity or multi-directional eddy current is detected, the noise amplitude is increased. The noise weight is used to enhance the adaptability of the filter to nonlinear motion. Among them, the parameters such as the state transfer matrix, control input matrix, and process noise covariance matrix are all designed in advance based on theoretical models and experimental calibration.

[0021] Here, the present invention combines ReID feature extraction with Kalman prediction, so that the system can maintain tracking continuity under the condition of short-term occlusion of the target or interference from appearance similarity. At the same time, through dynamic noise covariance optimization, the state estimation accuracy in complex underwater environments is significantly improved.

[0022] After that, the appearance feature vector of each target detected in the current frame is used to quantify the matching degree between the detected target and the followed target in the appearance feature space. Here, we use cosine similarity calculation for quantitative evaluation. The value range of cosine similarity is [-1,1]. The closer the value is to 1, the higher the consistency of the features between the two. Otherwise, the difference is significant. The formula is as follows: in, To detect the target The appearance feature vector of To follow the target The appearance feature vector of To detect the target and follow the target The cosine similarity between .

[0023] Then, the cosine similarity is converted into a matching cost matrix , the conversion formula is: This design amplifies the matching cost of low-similarity targets through squaring operations, thereby giving priority to high-similarity targets in the subsequent optimization process, effectively reducing mismatches caused by short-term changes in appearance (such as phase differences in pectoral fin swings or partial occlusion). At the same time, combined with the dynamic state (position and speed) predicted by the Kalman filter, the matching cost matrix is ​​constrained to spatial consistency: if the distance between the predicted position of the detected target and the tracked target exceeds a preset threshold, A penalty coefficient is applied to further exclude matching pairs with unreasonable spatial positions. The resulting matching cost matrix combines the similarity of appearance features and the consistency of motion states to ensure the robustness and accuracy of target association.

[0024] Based on the constructed matching cost matrix, the present invention then uses the Hungarian algorithm to perform optimal association matching between the detection target and the follow target, which is specifically implemented as follows: First, the matching cost matrix As input, the matrix rows represent the detected targets in the current frame, the columns represent the followed targets (including the historical data of the followed targets), and the matrix elements The smaller the value, the more likely the target is to be detected. Follow the target The higher the matching possibility. The Hungarian algorithm solves the matching combination with the lowest global cost by constructing an augmenting path and iteratively updating the labeling matrix. Its core is to transform the bipartite graph matching problem into a linear programming problem of minimum weight matching to ensure that the optimal solution is obtained in polynomial time. For successfully matched detection-tracking target pairs, their association results are marked as valid matches; for unmatched detection targets, they are determined to be newly added tracking targets and assigned unique IDs; for unmatched follow targets, whether to terminate tracking is determined based on their life cycle counts. In addition, in order to improve the robustness of the algorithm, a similarity threshold is set. Distance threshold to space Double check: If the cosine similarity between the detected target and the followed target is lower than or its location is more than ,Even if the Hungarian algorithm outputs a matching result, it is still considered an invalid match and triggers the reinitialization logic. The final output association result includes the ID of the successfully matched tracking target, the updated state vector, and the initialization information of the newly added target, providing high-confidence input for subsequent trajectory updates and motion information analysis.

[0025] For the successfully matched detection-following target pairs, the Kalman filter is also required to perform state correction. First, the corresponding bounding box center coordinates in the S1 detection result are As an observed value , combined with the previously predicted state vector ,in To predict the position, the Kalman gain To make a status update: in, is the updated state at time t, is the updated state covariance matrix (representing the updated state estimate uncertainty), is the observation matrix (used to map the state vector to the observation space, that is, to describe how to extract or calculate the observation value from the state vector), is the identity matrix. The observation matrix Map the state vector to the observation space and output the corrected position and speed , as the real-time motion status of the navigator.

[0026] Then, based on the corrected position information, the two-dimensional pixel coordinates are converted to Mapping to 3D space coordinates , the specific formula is: Combined with the speed of the Kalman filter output , further calculate the heading angle : Finally, the fusion position ,speed and heading angle , generate real-time motion information of the navigator, including three-dimensional space coordinate position, motion speed and orientation information.

[0027] For unmatched detection targets, the system initializes a new tracking target and records its initial state vector; for continuously unmatched historical tracking targets, the tracking is terminated according to the preset life cycle threshold to avoid the accumulation of invalid trajectories.

[0028] Finally, based on the analyzed real-time position, speed and orientation information of the leader, the frequency and amplitude of the pectoral fin of the follower vehicle are dynamically adjusted through the feedforward-feedback composite PID control algorithm to achieve high-precision collaborative following. As a feedforward input, combined with the distance error between the follower and the leader at the current moment ( To set the following distance), generate distance control output : in, is the distance error at the previous moment, the proportional term Fast response distance deviation, integral term Eliminate steady-state error, differential term Suppress overshoot. PID parameters The gain parameter is dynamically adjusted according to the environment. Its value is adaptively optimized according to the water flow disturbance intensity and target movement speed in the underwater environment: In the strong water flow disturbance scene, increase To improve the anti-interference ability; in the stable following stage, enhance To eliminate residual errors; when the target turns quickly, improve To reduce the risk of vibration.

[0029] Finally, the target velocity of the follower is The comprehensive feedforward and feedback calculations yield: Then, according to the speed control output and heading deviation , dynamically adjust the pectoral fin parameters according to the following rules.

[0030] Distance adjustment: When the actual distance ( is the tolerance threshold), linearly increase the pectoral fin swing amplitude and frequency , the relationship is: in, It is a proportional coefficient calibrated based on the fluid dynamics model of the vehicle to ensure the optimal balance between thrust and energy consumption.

[0031] When the actual distance When gradually reducing and , if the distance continues to decrease, then To enter gliding mode and reduce energy consumption.

[0032] Course adjustment: Unilateral thrust adjustment: If the heading deviation (e.g. 5°), generates steering torque through asymmetric pectoral fin movement: Left deviation correction: increase the amplitude of the right pectoral fin , generating right thrust, pushing the vehicle to turn left; right deviation correction: increasing the amplitude of the right pectoral fin , generating left thrust, pushing the spacecraft to turn right.

[0033] After the turn is completed, the symmetrical swing of the bilateral pectoral fins is restored to maintain straight-line navigation. In this way, through the deep integration of intelligent control algorithms and bionic actuators, the manta ray-like vehicle can achieve efficient and stable following in complex underwater environments, providing an innovative solution for marine resource exploration and collaborative operations.

[0034] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the collaborative operation following method for manta ray vehicles.

[0035] Also included is a device for processing data, comprising: a memory having a computer program stored thereon; A processor is used to execute the computer program in the memory to implement the steps of the collaborative operation following method for manta ray vehicles.

[0036] In summary, the collaborative operation following method for manta ray vehicles proposed in the present invention is based on the DeepSORT framework, uses the ReID network to extract multi-view appearance feature vectors, and combines the Kalman filter to dynamically predict the target state, effectively overcoming the interference of water flow disturbance on tracking, and constructs the matching cost matrix through cosine similarity and optimizes the association matching through the Hungarian algorithm, so as to improve the accuracy of target tracking and significantly reduce the phenomenon of target loss or mismatch.

[0037] Through the feedforward-feedback composite PID control algorithm, the real-time speed of the leader is used as the feedforward input, and the frequency and amplitude of the follower's pectoral fin are dynamically adjusted in combination with the distance error feedback to achieve fast response and precise following.

[0038] By combining the Kalman filter with the minimum noise covariance matrix, dynamically adjusting according to the water flow disturbance and combining it with three-dimensional spatial coordinate mapping technology, it is possible to analyze the precise position information of the target in real time and adapt to complex scenarios with sudden changes in flow velocity or irregular direction.

[0039] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0040] In addition, in the present invention, descriptions such as "first", "second", "one", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0041] In the present invention, unless otherwise clearly specified and limited, the terms "connection", "fixation", etc. should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0042] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A collaborative operation following method for a manta ray vehicle, characterized in that: The logical framework based on DeepSORT target detection and target tracking includes the following steps: S1: Use the YOLOv8 algorithm to detect the real-time inter-frame image of the navigator, identify the bounding box and category label of each detection target on the navigator, and generate the detection result; S2: Extract the appearance feature vector of the navigator based on the detection results, and predict the dynamic state of the navigator based on the Kalman filter; S3: Calculate the cosine similarity between the detected target and the followed target based on the appearance feature vector, and construct a matching cost matrix; S4: Use the Hungarian algorithm to optimally match the matching cost matrix and generate the matching results of the detected target and the followed target; S5: updating the dynamic state of the corresponding target trajectory according to the matching result, and obtaining the position, speed and orientation information of the navigator through analysis; S6: The position, speed and orientation information of the navigator are obtained according to the analysis, combined with the category label, and the frequency and amplitude of the pectoral fin of the follower are controlled by the PID algorithm to perform dynamic following.

2. A collaborative operation following method for a manta ray vehicle as claimed in claim 1, characterized in that: In the step S1, the YOLOv8 algorithm is specifically: Divide the real-time inter-frame image of the navigator into grid cells of a preset size; Extracting features of each detection target from the inter-frame image through a deep convolutional neural network and outputting a category label, wherein the features of the detection target include morphological features of the pectoral fin, the caudal fin, and the dorsal fin; Predict bounding box parameters for each grid cell and eliminate overlapping bounding boxes through non-maximum suppression; The final bounding box and the corresponding category label are output as the detection result.

3. A collaborative operation following method for a manta ray vehicle as claimed in claim 1, characterized in that: In the step S2, the appearance feature vector is extracted according to the detection result through the ReID network.

4. A collaborative operation following method for a manta ray vehicle as claimed in claim 1, characterized in that: In the step S2, the dynamic state prediction of the navigator based on the Kalman filter is specifically: using the Kalman filter to predict the state vector of the navigator, and estimating the motion state of the navigator by minimizing the covariance, the state vector including position and speed.

5. A collaborative operation following method for a manta ray vehicle as claimed in claim 4, characterized in that: In the step S5, the position, speed and orientation information of the navigator are obtained by analysis as follows: the position and speed of the detection target are predicted according to the dynamic state through the Kalman filter; according to the pixel coordinate changes of the detection target, the pixel coordinates are mapped into three-dimensional space coordinates through the camera intrinsic parameter matrix, and the heading and orientation information is calculated in combination with the speed prediction result of the Kalman filter.

6. A collaborative operation following method for a manta ray vehicle as claimed in claim 1, characterized in that: In the step S3, the matching cost matrix is ​​constructed by the following formula: In the formula, To detect the target Follow the target The matching cost matrix is To detect the target The appearance feature vector of To follow the target The appearance feature vector of To detect the target and follow the target The cosine similarity between .

7. A collaborative operation following method for a manta ray vehicle as claimed in claim 1, characterized in that: In step S6, the PID algorithm adopts feedforward-feedback compound control, and the formula is expressed as follows: In the formula, is the distance control output, is the distance error at the previous moment, For speed control output, To analyze and obtain the speed of the navigator, To dynamically adjust the parameters according to the environment, is the distance error at the current moment, and T is the sampling period.

8. A collaborative operation following method for a manta ray vehicle as claimed in claim 7, characterized in that: In step S6, the adjustment logic of the pectoral fin frequency and amplitude is: When the distance error exceeds the set threshold, the amplitude and frequency of the pectoral fins are increased to accelerate; when the distance error is lower than the set threshold, the amplitude and frequency of the pectoral fins are reduced or the amplitude is reset to zero; when the heading deviation exceeds the preset angle, steering is achieved by adjusting the amplitude of the unilateral pectoral fin.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of a collaborative operation following method for manta ray vehicles as described in any one of claims 1 to 8 are implemented.

10. A device for processing data, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of a collaborative operation following method for manta ray vehicles as described in any one of claims 1 to 8.

Images