Underwater Multi-Robot Obstacle Avoidance Device and Method under Communication Connectivity Preservation Constraint

Abstract

The present invention discloses an underwater multi-robot obstacle avoidance device and method under the constraint of maintaining communication connectivity, belonging to the field of underwater multi-robot swarm system control. The device includes an underwater robot main body, an underwater acoustic wireless communication device, and a binocular vision obstacle avoidance device. The method first deploys multiple underwater robots to the cruising area. The underwater robots obtain underwater environment images in real time through binocular cameras, calculate the obstacle parallax angle through the images, and determine whether it is greater than the obstacle avoidance threshold. Subsequently, the robots use the underwater acoustic wireless communication module to determine the communication distance with other robots in a broadcast form, and design a communication gravitational function through the communication distance to ensure the communication connectivity of the underwater multi-robots. After the underwater robots complete the cruising task, they float to the water surface. The present invention enables the underwater multi-robot swarm system to work in waters with obstacles, ensures the communication connectivity of the underwater multi-robot swarm system, and improves the stability of the motion control of the underwater robots.

Description

Technical Field

[0001] The present invention relates to the field of underwater multi-robot swarm system control, in particular to an underwater multi-robot obstacle avoidance device and method under the constraint of maintaining communication connectivity. Background Art

[0002] The 21st century is the century of the ocean. With the development of the economy and the progress of science and technology, the intelligence of ocean exploration equipment is the future development trend. The underwater multi-robot swarm system has the advantage of group cooperation in large-scale ocean exploration, ocean search and rescue, and tracking and encirclement. However, with the improvement of efficiency, communication problems and control problems of the underwater multi-robot swarm system also arise. The underwater environment is different from the ground and has its particularity. First of all, compared with the ground, underwater communication means are limited and are easily affected by water temperature, light, noise, etc. Unstable communication connections will cause the cruise mission of the underwater multi-robot swarm system to fail. Secondly, the underwater environment is complex and changeable, full of many uncertain factors. For example, the existence of underwater reefs, corals, and fish schools makes it challenging for the underwater multi-robot swarm system to move without collision. These factors bring great difficulties to the cruise mission of the underwater multi-robot swarm system.

[0003] In the prior art, the Chinese invention patent with the publication number CN115185287A discloses an intelligent multi-underwater robot dynamic obstacle avoidance and encirclement control system. After a certain underwater robot discovers a target, it sends the encirclement target position, moving direction, and speed to other robots in real time through a communication device, and encircles the target through the pre-calculated encirclement range. This solution can achieve dynamic encirclement of the target through the cooperation of multiple underwater robots. However, when broadcasting target information to other robots, this solution does not consider the communication connectivity between itself and other underwater robots. Moreover, the marine environment is complex and changeable. Without considering the communication connectivity constraint, it will lead to communication connection interruption and then cause the failure of the underwater multi-robot swarm system task.

[0004] Furthermore, the Chinese invention patent with the publication number CN103529844A discloses an underwater robot obstacle avoidance method based on a forward-looking sonar. This method introduces forward-looking sonar image data into the robot's obstacle avoidance strategy, reducing the robot's collision avoidance blind area. This method can provide long-distance and high-resolution underwater images for underwater robots. However, when the underwater robot is in a close-range complex obstacle scene, due to the influence of underwater noise and multipath effects, the sonar resolution is low, resulting in a decline in the robot's obstacle avoidance performance.

[0005] In view of the above deficiencies, it is particularly important to design a communication connectivity maintenance constraint to improve the communication quality of multiple underwater robots and design an underwater multi-robot obstacle avoidance method and device under this constraint. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an underwater multi-robot obstacle avoidance device and method under the constraint of maintaining communication connectivity, which can achieve stable obstacle avoidance control of a robotic swarm system under the constraint of maintaining communication connectivity, and improve the communication and motion stability of the system.

[0007] To solve the above technical problems, the technical solution adopted by the present invention is:

[0008] An underwater multi-robot obstacle avoidance device under the constraint of maintaining communication connectivity, each underwater robot includes an underwater robot main body, an underwater acoustic wireless communication device, and a binocular vision obstacle avoidance device;

[0009] The underwater robot main body includes a robot carrier constituting the robot main body frame, a power system including six thruster modules, four buoyancy materials symmetrically fixed to the front and rear sides of the robot main body frame in pairs, a control cabin fixedly installed in the center of the robot main body frame, a battery cabin containing a power supply system, and two underwater searchlights fixedly installed at the bottom of the front end of the robot main body frame;

[0010] The underwater acoustic wireless communication device includes an underwater acoustic transducer, a modem, a communication module, a single-chip microprocessor, and a lithium battery power supply system; the underwater acoustic transducer is used to send and receive communication high-frequency ultrasonic waves; the modem is used to convert analog signals into digital signals; the communication module is used for data exchange between the single-chip microprocessor and the modem; the single-chip microprocessor is used to process the data information sent by the modem; the lithium battery power supply system is used to supply power to the underwater acoustic wireless communication device;

[0011] The binocular vision obstacle avoidance device is composed of a first monocular camera, a second monocular camera, a third monocular camera, and a fourth monocular camera. The first monocular camera, the second monocular camera, the third monocular camera, and the fourth monocular camera are respectively fixedly installed at the front left, front right, directly above, and directly below the front part of the robot main body frame, and are used to capture real-time optical images of the underwater environment.

[0012] A further improvement of the technical solution of the present invention is that: the robot carrier includes a first carrier, a second carrier constituting the outer frame of the robot main body frame, and a third carrier constituting the bottom of the robot main body frame; the first carrier and the second carrier are arranged vertically and parallel to each other; the third carrier is perpendicular to the first carrier and the second carrier, and is fixedly connected to the first carrier and the second carrier;

[0013] The six thruster modules included in the power system specifically refer to two ascent / descent thrusters fixedly installed on the left and right sides of the control cabin, and four forward / backward thrusters fixed on the first and second carriers below the buoyancy material at an angle of 45° to the horizontal direction; the buoyancy material is located on the front and back sides of the ascent / descent thrusters;

[0014] The interior of the control cabin includes a motor drive module, a single-chip microprocessor unit, and a microcomputer image processing unit; the battery cabin is fixedly installed in the middle of the third carrier.

[0015] An underwater multi-robot obstacle avoidance method under communication connectivity maintenance constraints includes the following steps:

[0016] Step 1: Collect the environmental images of the underwater cruise mission area and the images of the underwater robots, preprocess the obstacle avoidance areas in the collected images and generate corresponding data sets, perform offline training on the data sets through a deep convolutional neural network, and deploy the trained model to the microcomputer image processing unit in each underwater robot control cabin;

[0017] Step 2: Deploy each underwater robot to the cruise mission area respectively. The underwater robot captures the environmental image information in real time through the binocular camera carried by itself. The microcomputer image processing unit judges whether there is an obstacle avoidance area in this image through the deployed neural network model. If there is an obstacle avoidance area, go to Step 3; if not, go to Step 4;

[0018] Step 3: Obtain the coordinates of the central pixel point of the obstacle avoidance area on the image, and calculate the parallax angle value of the obstacle avoidance area through these pixel point coordinates;

[0019] Step 4: Based on the communication radius of the underwater acoustic wireless communication device, obtain the neighbor set of each underwater robot respectively, and then calculate the communication gravitational field value between each underwater robot and other underwater robots in its neighbor set;

[0020] Step 5: Construct the reward function of this underwater robot through the parallax angle information obtained in Step 3 and the communication gravitational field obtained in Step 4, construct the value function based on the reward function, and fit the value function through a deep reinforcement learning neural network;

[0021] Step 6: Repeat the above Steps 2 to 5 until the optimal value function is obtained. At this time, the deep reinforcement learning neural network has converged, and deploy it to each underwater robot to obtain the optimal control strategy.

[0022] A further improvement of the technical solution of the present invention lies in: in step 3, if there is an obstacle avoidance area in the image, the pixel coordinates of its center point are recorded as (X, Y), where X and Y are the horizontal and vertical pixel coordinates of the center point of the obstacle avoidance area respectively; by obtaining the pixel coordinates of the center of the obstacle avoidance area of the image, the horizontal parallax angle and the vertical parallax angle of this obstacle avoidance area are calculated:

[0023]

[0024]

[0025] where, θ H and θ V are the horizontal parallax angle and the vertical parallax angle of this obstacle area respectively, θ T is the obstacle avoidance parallax angle threshold, and θ A , θ B , θ C and θ D are the horizontal and vertical parallax included angle values calculated by the first monocular camera, the second monocular camera, the third monocular camera, and the fourth monocular camera according to the pixel coordinates of the center of the image obstacle avoidance area respectively.

[0026] A further improvement of the technical solution of the present invention lies in: in step 4, the connection function between the underwater robot U m and the underwater robot U n is:

[0027]

[0028] where, m, n ∈ {1,...M}, X m = [x m , y m , z m T and X n = [x n , y n , z n T respectively represent the positions of the underwater robots U m and U n in the world coordinate system; L(X m , r v ) = {X ∈ R 2 : ||X - X m || ≤ r v} represents a circular area with a communication radius of r m centered on the underwater robot U v ; through the connection function, the neighbor set of the underwater robot U m is:

[0029] P m = {X​​n ·f mn (X m )} (4)

[0030] Underwater robot U m The communication gravitational field generated within its neighbor set is as follows:

[0031]

[0032] where d mn =||X m -X n || represents the distance between underwater robot U m and underwater robot U n , and r s is the maximum stable communication distance.

[0033] A further improvement of the technical solution of the present invention lies in: in step 5, based on the obstacle parallax angle information and the communication gravitational field constraint obtained in the above technical solution steps, a single-step reward function can be constructed to calculate the reward value of the current strategy. The greater the reward, the better the control strategy; the reward function is as follows:

[0034]

[0035] where R m (X m ,τ m ) is the single-step reward of underwater robot U m , τ m is the control input of underwater robot U m at this time, and K1 and K2 are weight coefficients.

[0036] A further improvement of the technical solution of the present invention lies in: in step 6, the value function is updated based on the single-step reward function in step 5. The definition of the value function is as follows:

[0037] Q(X mk ,τ mk )=R m (X mk ,τ mk )+γ×maxQ(X mk+1 ,τ mk+1 ) (7)

[0038] where X mk and τ mk are the position and control input of underwater robot U m at time step k, 0<γ≤1 is the discount factor; the value function is iteratively updated by fitting through a deep reinforcement learning neural network, and steps 2 to 5 are repeated until the convergence requirement of the neural network is met; at this time, the optimal control strategy is obtained through the neural network

[0039] Due to the adoption of the above technical solution, the technical progress achieved by the present invention is as follows:

[0040] 1. The binocular disparity obstacle avoidance method based on binocular vision of the present invention can enable the underwater multi-robot swarm system to avoid collisions between underwater robots and underwater obstacles as well as other underwater robots when working in complex obstacle waters, getting rid of the low close-range resolution of traditional acoustic obstacle avoidance sensors and the influence of acoustic wave multipath effects, and improving the control stability of underwater robots.

[0041] 2. The communication connectivity maintenance constraint scheme proposed by the present invention improves the communication stability during the collaborative work of underwater multi-robots, solves the problem of disconnection of the underwater robot swarm system, and combines with the deep reinforcement learning neural network to improve the communication connectivity and control stability of the underwater multi-robot swarm system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts;

[0043] Figure 1 Stereo schematic diagram of an underwater robot carrying a binocular vision obstacle avoidance device in an embodiment of the present invention;

[0044] Figure 2 Side view schematic diagram of the mechanical structure of an underwater robot in an embodiment of the present invention;

[0045] Figure 3 Schematic diagram of the principle of an underwater acoustic wireless communication device in an embodiment of the present invention;

[0046] Figure 4 Schematic diagram of binocular vision disparity detection in an embodiment of the present invention;

[0047] Figure 5 Schematic diagram of the communication connectivity area division of an underwater robot in an embodiment of the present invention;

[0048] Figure 6 Flowchart of a multi-robot obstacle avoidance method under communication connectivity maintenance constraint in an embodiment of the present invention;

[0049] Among them, 1. Buoyancy material, 2. Underwater searchlight, 3-1. First carrier, 3-2. Second carrier, 4. Ascending / descending thruster, 5-1. First monocular camera, 5-2. Second monocular camera, 5-3. Third monocular camera, 5-4. Fourth monocular camera, 6. Third carrier, 7. Control cabin, 8. Battery cabin, 9. Forward / backward thruster, 10. Threaded bolt. Detailed implementation mode

[0050] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0051] The present invention will be further described in detail below with reference to the drawings and embodiments:

[0052] See Figure 1 and Figure 2 、 Figure 3 and Figure 4 and, which show an underwater robot obstacle avoidance device under the constraint of maintaining communication connectivity in an embodiment of the present invention. Each underwater robot includes an underwater robot main body, an underwater acoustic wireless communication device, and a binocular vision obstacle avoidance device;

[0053] The underwater robot main body includes a robot carrier, a power system, a buoyancy material 1, a control cabin 7, a battery cabin 8, and an underwater searchlight 2;

[0054] The robot carrier is a robot main body frame composed of a first carrier 3-1, a second carrier 3-2, and a third carrier 6; the first carrier 3-1 and the second carrier 3-2 are vertically and parallelly arranged, constituting the outer frame of the underwater robot main body frame; the third carrier 6 constitutes the bottom of the underwater robot main body frame, is perpendicular to the first carrier 3-1 and the second carrier 3-2, and is fixedly connected to the first carrier 3-1 and the second carrier 3-2;

[0055] The power system includes a total of six thruster modules. Specifically, two ascending / descending thrusters 4 are fixed on the left and right sides of the control cabin 7, and four forward / backward thrusters 9 are fixed under the buoyancy material 1 and on the first carrier 3-1 and the second carrier 3-2 at an angle of 45° to the horizontal direction.

[0056] There are a total of four pieces of the buoyancy material 1, and they are symmetrically fixed in pairs on the front and back sides of the robot main body frame.

[0057] The control cabin 7 is fixed in the center of the robot main body frame and contains a motor drive module, a single-chip microprocessor unit, and a microcomputer image processing unit. The motor drive module is used to drive the power system to work. The single-chip microprocessor is used to receive the signals sent by all sensors and send control instructions to drive the robot to move. The microcomputer image processing unit is used to process the environmental image information collected in real time by the binocular vision system.

[0058] The battery cabin 8 is fixed in the middle of the upper side of the third carrier 6 and is used to assemble the power supply system of the underwater robot.

[0059] There are a total of two underwater searchlights 2, which are fixed on the inner sides of the bottoms of the first carrier 3-1 and the second carrier 3-2 and are connected to the inside of the battery cabin 8 through threading bolts 10, and can provide lighting for the underwater robot in deep water areas.

[0060] As Figure 3 shown, it shows the schematic diagram of the underwater acoustic wireless communication device in the embodiment of the present invention. The underwater acoustic wireless communication device includes an underwater acoustic transducer, a modem, a communication module, a single-chip microprocessor, and a lithium battery power supply system;

[0061] The underwater acoustic transducer is connected to the modem and is used to send and receive high-frequency ultrasonic signals;

[0062] The modem converts the analog signal transmitted by the underwater acoustic transducer into a digital signal;

[0063] The communication module is respectively connected to the modem and the single-chip microprocessor. The modem sends the digital signal to the single-chip microprocessor through the communication module according to the communication protocol;

[0064] The single-chip microprocessor analyzes the data information sent by the communication module according to the communication protocol;

[0065] The lithium battery power supply system powers the entire underwater acoustic wireless communication device.

[0066] As Figure 4As shown in the figure, it shows a binocular vision parallax detection schematic diagram in an embodiment of the present invention. The binocular vision obstacle avoidance device is composed of four monocular cameras. The first monocular camera 5-1 and the second monocular camera 5-2 are respectively fixed on the outer sides of the middles of the first carrier 3-1 and the second carrier 3-2. The third monocular camera 5-3 is fixed on the upper side in front of the control cabin 7, and the fourth monocular camera 5-4 is fixed on the lower side in front of the third carrier 6. The first monocular camera 5-1 and the second monocular camera 5-2 are horizontally symmetrically distributed to form a horizontal binocular camera, and the third monocular camera 5-3 and the fourth monocular camera 5-4 are vertically symmetrically distributed to form a vertical binocular camera. The four monocular cameras are used to capture underwater environmental optical images in real time, and calculate the horizontal parallax angle and the vertical parallax angle for the obstacle avoidance area respectively.

[0067] As Figure 6 shown, an underwater multi-robot obstacle avoidance method under communication connectivity maintenance constraints specifically includes the following steps:

[0068] Step 1: Combine the characteristic information of underwater obstacles and underwater robots, clean the collected underwater cruise mission area environment images and underwater robot images, and process the required obstacle avoidance areas to generate a dataset to be trained. Subsequently, build a convolutional neural network model to perform offline training on the dataset. When the neural network loss function converges to a set threshold, the training is completed, and the trained model is deployed to the microcomputer image processing unit in the underwater robot control cabin 7.

[0069] Step 2: Deploy all underwater robots to the cruise mission area respectively. Before starting the cruise mission, for any robot, it can communicate normally with all other underwater robots except itself. The underwater robot captures the image information of the surrounding environment in real time through the binocular camera carried by itself, and transmits the image to the microcomputer image processing unit in the control cabin 7 in real time. Predict through the neural network model trained in Step 1. If there is an area that needs to avoid obstacles in the image, go to Step 3, otherwise go to Step 4;

[0070] Step 3: Obtain the coordinates (X, Y) of the central pixel point of the obstacle avoidance area on the image, where X and Y are the pixel abscissa and ordinate of the center point of the obstacle avoidance area respectively; further, through the coordinates of the central pixel point of this obstacle avoidance area, calculate the horizontal parallax angle and the vertical parallax angle of this obstacle avoidance area:

[0071]

[0072]

[0073] where θ H and θ V are respectively the horizontal parallax angle and the vertical parallax angle of this obstacle area, and θ Tis the obstacle avoidance parallax angle threshold, θ A , θ B , θ C and θ D are the horizontal and vertical parallax angle values calculated by the first monocular camera 5-1, the second monocular camera 5-2, the third monocular camera 5-3, and the fourth monocular camera 5-4 according to the central pixel coordinates of the image obstacle avoidance area, respectively.

[0074] Step 4, taking the underwater robot U m as an example, its connection relationship with the underwater robot U n is:

[0075]

[0076] where m, n ∈ {1,... M}, M is the number of all underwater robots included in the underwater multi-robot cluster system, X m = [x m , y m , z m T and X n = [x n , y n , z n T represent the positions of the underwater robots U m and U n in the world coordinate system respectively. L(X m , r v ) = {X ∈ R 2 : ||X - X m || ≤ r v} represents a circular area with a communication radius of r m centered at the position X m of the underwater robot U v . Through this connection function relationship, the neighbor set of the underwater robot U m can be calculated as:

[0077] P m = {X n ·f mn (X m )} (4)

[0078] Furthermore, as Figure 5 shown, it shows the schematic diagram of the underwater robot communication connection area division in the embodiment of the present invention. Inside the circular area with a communication radius of r m centered at the underwater robot U v , a maximum stable communication area with a radius of r m centered at the underwater robot U s is divided. According to this division rule, the underwater robot U is calculated.​m The communication gravitational field generated within its neighbor set P m :

[0079]

[0080] where d mn = ||X m - X n || represents the distance between the underwater robot U m and the underwater robot U n . The underwater robot U m located inside the maximum stable communication area of the underwater robot U n and the underwater robot U m have very safe and reliable communication guarantees, and no communication gravity will be generated at this time. When located outside the maximum safe communication area, at this time, the communication ability between the underwater robot U n and the underwater robot U m is weak, and a mutually attractive communication gravity needs to be generated to ensure the subsequent safe, stable, and reliable communication ability;

[0081] Step 5, construct the single-step reward function of this underwater robot U m using the parallax angle information obtained in Step 3 and the communication gravitational field obtained in Step 4:

[0082]

[0083] where R m (X m , τ m ) is the single-step reward of the underwater robot U m . Its main body is composed of the obstacle parallax angle and the communication connectivity constraint relationship. When the underwater robot U m is closer to the obstacle, the obtained horizontal parallax angle θ H and vertical parallax angle θ V are larger at this time, and the obtained reward value is smaller. Conversely, the reward is larger. Similarly, when the underwater robot U n is closer to the underwater robot U m , the communication is more stable, and the obtained reward value is larger. Conversely, the reward is smaller. τ m is the control input of the underwater robot U m at this time. K1 and K2 are weight coefficients. The magnitude of the single-step reward value reflects the quality of the control strategy τ m at this time. The larger the reward, the better the control strategy. Conversely, the control strategy is worse.

[0084] Step 6, update the value function using the single-step reward in Step 5. The value function is defined as follows:

[0085] Q(X mk , τ mk ) = R m (X mk , τ mk ) + γ × max Q(X mk+1 , τ mk+1 ) (7)

[0086] Wherein, X mk and τ mk are the position and control input of the underwater robot U m at time step k, and 0 < γ ≤ 1 is the discount factor. All underwater robots perform fitting iterative updates on the value function through the deep reinforcement learning neural networks built by themselves, and repeat steps 2 to 5 until the neural networks converge. The controller is respectively deployed on the microcontroller unit in the corresponding underwater robot control cabin 7. At this time, the underwater robot can obtain the optimal control strategy

[0087] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An underwater multi-robot obstacle avoidance method under the constraint of maintaining communication connectivity, characterized in that: It includes the following steps: Step 1: Collect the environmental images and underwater robot images in the underwater cruise mission area. Preprocess the obstacle avoidance areas in the collected images and generate corresponding datasets. Offline train the datasets through a deep convolutional neural network, and deploy the trained model to the microcomputer image processing unit in each underwater robot control cabin; Step 2: Deploy each underwater robot to the cruise mission area respectively. The underwater robot captures the environmental image information in real time through the binocular camera carried by itself. The microcomputer image processing unit determines whether there is an obstacle avoidance area in this image through the deployed neural network model. If there is an obstacle avoidance area, go to Step 3; if not, go to Step 4; Step 3: Obtain the coordinates of the central pixel point of the obstacle avoidance area on the image, and calculate the parallax angle value of the obstacle avoidance area through these pixel point coordinates; If there is an obstacle avoidance area in the image, record the pixel coordinates of its center point as (X, Y), where X and Y are the pixel abscissa and ordinate of the center point of the obstacle avoidance area respectively; by obtaining the coordinates of the central pixel point of the obstacle avoidance area of the image, calculate the horizontal parallax angle and vertical parallax angle of this obstacle avoidance area: where θ H and θ V are the horizontal parallax angle and the vertical parallax angle of this obstacle area respectively, and θ T is the obstacle avoidance parallax angle threshold, and θ A , θ B , θ C and θ D are the horizontal and vertical parallax included angle values calculated by the first monocular camera (5-1), the second monocular camera (5-2), the third monocular camera (5-3), and the fourth monocular camera (5-4) according to the coordinates of the central pixel point of the image obstacle area respectively; Step 4: Calculate the neighbor set of each underwater robot based on the communication radius of the underwater acoustic wireless communication device, and then calculate the communication gravitational field value between each underwater robot and other underwater robots in its neighbor set; Underwater robot U m The connection function with the underwater robot U n is as follows: where \(m,n\in\{1,\ldots,M\}\), \(X\) m \( = [x\) m ,y\) m ,z\) m \) T and \(X\) n \( = [x\) n ,y\) n ,z\) n \) T represent the positions of the underwater robots \(U\) m and \(U\) n in the world coordinate system respectively; \(L(X\) m ,r\) v )=\(\{X\in\mathbb{R}\) 2 :\|X - X\) m \|\leq r\) v \} represents a circular area with the underwater robot \(U\) m as the center and the communication radius \(r\) v ; through the connectivity function, the neighbor set of the underwater robot \(U\) m is as follows: P m = {X n · f mn (X m )} (4) Robot U m The communication gravitational field generated within its neighbor set is as follows: Among them, d mn =||X m -X n || represents the distance between the underwater robot U m and the underwater robot U n , r s is the maximum stable communication distance; Step 5: Construct the reward function of this underwater robot through the parallax angle information obtained in Step 3 and the communication gravitational field obtained in Step 4. Construct the value function based on the reward function, and fit the value function through a deep reinforcement learning neural network; Through the obstacle parallax angle information and communication gravitational field constraints obtained in the above steps, a single-step reward function can be constructed to calculate the reward value of the current strategy. The greater the reward, the better the control strategy; the reward function is as follows: Among them, R m (X m , τ m ) is the single-step reward of the underwater robot U m , τ m is the control input of the underwater robot U m at this time, and K1 and K2 are weight coefficients; Step 6: Repeat the above Steps 2 to 5 until the optimal value function is obtained. At this time, the deep reinforcement learning neural network has converged, and deploy it to each underwater robot to obtain the optimal control strategy; Update the value function based on the single-step reward function in Step 5. The definition of the value function is as follows: Q(X mk ,τ mk ) = R m (X mk ,τ mk ) + γ × maxQ(X mk+1 ,τ mk+1 ) (7) where X mk and τ mk are the position and control input of the underwater robot U m at time step k, 0 < γ ≤ 1 is the discount factor; the value function is iteratively updated by fitting through a deep reinforcement learning neural network, and steps 2 to 5 are repeated until the convergence requirement of the neural network is met; at this time, the optimal control strategy is obtained through the neural network 2. An obstacle avoidance device for the underwater multi-robot obstacle avoidance method under the communication connectivity maintenance constraint described in claim 1, characterized in that: Each underwater robot includes an underwater robot main body, an underwater acoustic wireless communication device, and a binocular vision obstacle avoidance device; The underwater robot main body includes a robot carrier that constitutes the robot main body frame, a power system including six thruster modules, 4 buoyancy materials (1) symmetrically fixed on the front and rear sides of the robot main body frame in pairs, a control cabin (7) fixedly installed in the center of the robot main body frame, a battery cabin (8) containing a power supply system, and 2 underwater searchlights (2) fixedly installed at the bottom of the front end of the robot main body frame; The underwater acoustic wireless communication device includes an underwater acoustic transducer, a modem, a communication module, a single-chip microprocessor, and a lithium battery power supply system; the underwater acoustic transducer is used to transmit and receive communication high-frequency ultrasonic waves; the modem is used to convert analog signals into digital signals; the communication module is used to perform data exchange between the single-chip microprocessor and the modem; the single-chip microprocessor is used to process the data information sent by the modem; the lithium battery power supply system is used to supply power to the underwater acoustic wireless communication device; The binocular vision obstacle avoidance device is composed of a first monocular camera (5-1), a second monocular camera (5-2), a third monocular camera (5-3), and a fourth monocular camera (5-4). The first monocular camera (5-1), the second monocular camera (5-2), the third monocular camera (5-3), and the fourth monocular camera (5-4) are respectively fixedly installed at the front due left, due right, directly above, and directly below of the robot main body frame, and are used to capture optical images of the underwater environment in real time.

3. The underwater multi-robot obstacle avoidance device under the constraint of maintaining communication connectivity according to claim 2, wherein: The robot carrier includes a first carrier (3-1) and a second carrier (3-2) that form the outer frame of the robot main body frame, and a third carrier (6) that forms the bottom of the robot main body frame; the first carrier (3-1) and the second carrier (3-2) are arranged vertically and parallel; the third carrier (6) is perpendicular to the first carrier (3-1) and the second carrier (3-2), and is fixedly connected to the first carrier (3-1) and the second carrier (3-2); The six thruster modules included in the power system specifically refer to two ascent / descent thrusters (4) fixedly installed on the left and right sides of the control cabin (7), and four forward / backward thrusters (9) fixed on the first carrier (3-1) and the second carrier (3-2) below the buoyancy material (1) at an angle of 45° to the horizontal direction; the buoyancy material (1) is located on the front and back sides of the ascent / descent thrusters (4); The control cabin (7) internally includes a motor drive module, a single-chip microprocessor unit, and a microcomputer image processing unit; the battery cabin (8) is fixedly installed at the middle part of the third carrier (6).

Images