A bio-inspired neural network affine transformation path planning method for underwater robot swarms

Abstract

The present invention provides a bio-inspired neural network affine transformation method for underwater robot swarm path planning, relating to the technical field of swarm underwater robot path planning. The method comprises: planning a path for a swarm of autonomous underwater vehicles (AUVs) based on a bio-inspired neural network; planning a formation for the swarm AUVs based on the planned path using an artificial potential field method, and dynamically adjusting the formation using affine transformation; simplifying the planned path using the Douglas Peucker DP algorithm, and curve fitting the simplified path using third-order B-splines. This application is applicable to swarm AUVs, and the planned path has the advantages of safe obstacle avoidance, short time consumption, and short distance.

Description

Technical Field

[0001] The present invention relates to the technical field of cluster underwater robot path planning, and in particular to a bio-inspired neural network affine transformation underwater robot cluster path planning method. Background Art

[0002] Autonomous underwater vehicles (AUVs) play a vital role in seabed resource exploration and underwater search and rescue missions. Faced with increasingly complex, time-sensitive, and large-scale missions, a single AUV is often insufficient. Collaborative missions involving multiple AUVs often improve efficiency, making planning the navigation paths and formations of AUV fleets particularly important.

[0003] Currently, there are often two problems when planning the paths of swarm AUVs: 1) Path planning only considers the impact of terrain on navigation and ignores the obstruction of ocean currents on AUV navigation, or only considers the impact of ocean currents on the navigation of a single AUV; 2) Swarm AUVs expand the collision boundary during path planning. In narrow terrain, a single AUV can traverse the narrow terrain and reach the destination in a shorter path, but swarm AUVs need to circumnavigate the island as a whole to avoid collisions under the conditions of a given formation. Summary of the Invention

[0004] Aiming at the problem in the prior art that cluster AUVs need to go around the island as a whole to avoid collision when crossing narrow terrain, the present invention proposes a bio-inspired neural network affine transformation underwater robot cluster path planning method.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] A bio-inspired neural network affine transformation underwater robot cluster path planning method, the method comprising:

[0007] Path planning for swarm autonomous underwater vehicles (AUVs) based on biologically inspired neural networks;

[0008] Based on the planned path, the artificial potential field method is used to plan the formation of the cluster AUVs, and then the affine transformation is used to dynamically adjust the formation;

[0009] The Douglas Peucker DP algorithm is used to simplify the planned path, and the simplified path is curve fitted using a third-order B-spline.

[0010] Furthermore, the biologically inspired neural network specifically includes: firstly mapping the topographic map and ocean currents one by one into a grid map, each grid corresponding to a neuron, and then giving it an initial neuron activity, and the activity at the end point isE , the obstacle activity is - E ,in E is any maximum value; the neuron propagates activity outward with the propagation weight. After the activity propagation iteration is stable, the AUV moves from the starting point along the direction of the neuron activity propagation to the end point;

[0011] Among them, when calculating the propagation weight of neuron activity, the grid occupied by the current AUV formation is first determined, and then the propagation weight of each occupied neuron is calculated, and then the propagation weight of each neuron is accumulated to the neuron where the navigator AUV is located, and it is used as the activity propagation weight of the navigator AUV when performing the current step movement.

[0012] Furthermore, during the process of propagation along neuronal activity, the formula is as follows:

[0013]

[0014]

[0015]

[0016]

[0017]

[0018] in, Represents neurons activity, Represents neurons The adjacent non-obstacle neurons, is the activity constraint function, Indicates As the leader, the group goes to The neuronal propagation activity of the point, express The initial activity value of the point, is a positive constant coefficient, is the propagation coefficient, represents the neuron activity propagation radius, Represents neurons and The distance between is a constant coefficient greater than 1;

[0019] As the number of iterations As , the neuron activity gradually tends to stabilize and converge; then, according to the neuron activity at each point, it starts from the starting point and reaches the end point along the direction of neuron activity propagation.

[0020] Furthermore, the artificial potential field method is used to plan the formation of the cluster AUVs, specifically including: the artificial potential field method guides the formation of AUVs by constructing an attractive potential field and a repulsive potential field; for the formation follower AUV, the attractive force comes from the guidance of the formation leader AUV, and the repulsive force comes from the need to avoid environmental obstacles; The total force on the formation follower AUVs at time T Expressed as:

[0021] (6)

[0022] In formula (6), the attraction is:

[0023] (7)

[0024] The repulsive force is:

[0025] (8)

[0026] in, is the current position of the follower AUV; is the position of the navigator, obtained by the bio-inspired neural network planning; is the nearest obstacle position, obtained from the prior known map; is the distance between the current follower AUV and the environmental obstacles, is the safety distance threshold.

[0027] Furthermore, the formation is dynamically adjusted using affine transformation, specifically including: the nominal position of a single follower AUV in the formation is , the actual position after affine transformation for:

[0028]

[0029] Affine transformation matrix Decomposed into rotation matrix With the scaling matrix A combination of:

[0030]

[0031] Its specific matrix form is as follows:

[0032]

[0033]

[0034] in, and They are formed into teams Axis and Scaling factor in the axis direction, is the overall rotation angle of the formation; if the environmental obstacles are concentrated on axis direction, the reduced formation is in Scale on the axis ,Keep Axis scale unchanged or adjusted appropriately, and vice versa.

[0035] Furthermore, the dynamic adjustment of the formation using affine transformation also includes: based on the resultant force analysis of the artificial potential field method, the total repulsive force exerted by environmental obstacles on the formation is obtained by summing the repulsive forces of all AUVs:

[0036]

[0037] in, Indicates the artificial potential field force, Represents the resultant force exist and The component force on the axis, so the scaling factor is calculated as follows:

[0038]

[0039]

[0040] in, as well as Respectively represent the preset minimum scaling ratio to prevent AUV formations from colliding during scaling. and The scaling adjustment coefficient controls the sensitivity of scaling to the repulsive force of environmental obstacles; and Respectively reflect the degree of proximity or obstruction of the current formation to environmental obstacles in the x and y axis directions.

[0041] Furthermore, the use of affine transformation to dynamically adjust the formation also includes: using the direction angle Indicates the main direction of the distribution of environmental obstacles:

[0042]

[0043] In this way, the overall rotation angle of the formation is:

[0044]

[0045] in, represents the yaw angle of the leader AUV, is the rotation sensitivity coefficient; angle correction is used to automatically adjust the posture of the formation when entering a complex obstacle environment.

[0046] Furthermore, the Douglas Peucker DP algorithm is used to simplify the planned path, specifically including:

[0047] Initialize first: enter the planned path list and threshold ;

[0048] Find the point in the path farthest from the end line: Initialize the maximum distance ,index ;

[0049] Traverse all intermediate points in the path: calculate the vertical distance from the point to the line connecting the beginning and the end ;if , then update and ;

[0050] if , divide the path into two parts: a. From the starting point to The subpath at b. The subpath to the end point;

[0051] Recursively call the current algorithm to process these two parts; merge the two paths;

[0052] Otherwise, check whether the straight line between the first and last points collides with an obstacle: a. If there is no collision, return directly to the starting and end points; b. If there is a collision, keep the original path;

[0053] Finally, the final simplified path is returned.

[0054] Furthermore, the simplified path is subjected to curve fitting using a third-order B-spline, specifically comprising: the third-order B-spline curve equation is shown in formula (18):

[0055]

[0056] in, Represents the set of AUV planning path points points within, express degree polynomial basis functions, where represents a non-decreasing node vector, and the basis function equation is as follows:

[0057]

[0058]

[0059] Compared with the prior art, the present invention has the following advantages:

[0060] This application encodes ocean current and obstacle information into a grid map, with each grid corresponding to a neuron. Clustered AUVs can then complete point-to-point optimal path planning based on neuron activity. When the path planning neural network propagates neuron activity, the propagation weight is affected not only by the state of the AUV itself, but also by the influence of the surrounding clustered AUVs. By comprehensively considering the navigation costs of each AUV in the cluster during navigation, path planning is performed, making the planned path more suitable for clustered AUVs.

[0061] The path planned in this application comprehensively considers the impact of terrain and ocean currents on the AUV, and has higher navigation efficiency than the path planning method that aims at the shortest distance.

[0062] In addition, in order to prevent AUVs from navigating around an island as a whole to avoid collision, the present invention uses an artificial potential field and an affine transformation algorithm to enable clustered AUVs to automatically transform their formation in narrow terrain.

[0063] Finally, the present invention also uses B-spline curve and improved Douglas Peucker algorithm to perform path smoothing optimization based on the dynamic characteristics of AUV.

[0064] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 2. It is a flow chart of the bio-inspired neural network affine transformation underwater robot cluster path planning method according to the present invention;

[0066] Figure 2 Schematic diagram of neuron activity propagation in a bio-inspired neural network applicable to swarm AUVs according to the present invention;

[0067] Figure 3 is a biologically inspired neural network structure suitable for swarm AUVs according to the present invention;

[0068] Figure 4 is a schematic diagram of the affine transformation principle of the artificial potential field according to the present invention;

[0069] Figure 5 Schematic diagram of the optimization degree of the path by the improved DP algorithm under different thresholds according to the present invention;

[0070] FIG6( a ) is a schematic diagram of a trajectory generated using path planning points as control points according to the present invention;

[0071] FIG6( b ) is a schematic diagram of the trajectory generated by the obstacle expansion and midpoint interpolation strategy according to the present invention;

[0072] Figure 7 2. It is a schematic diagram comparing the bio-inspired planning path, the B-spline smoothing path, and the DP algorithm combined with the B-spline smoothing path according to the present invention;

[0073] Figure 8 This is a schematic diagram of the existing biologically inspired neural network planning results;

[0074] Figure 9 Schematic diagram of the planning results of the path planning framework according to the present invention. DETAILED DESCRIPTION

[0075] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention, so as to more clearly understand the purposes, features and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not limitations on the scope of the present invention, but are only intended to illustrate the essential spirit of the technical solutions of the present invention. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative work should fall within the scope of protection of the present invention.

[0076] Unless the context requires otherwise, throughout the specification and claims, the word "comprise" and variations such as "include" and "have" should be construed in an open, inclusive sense, that is, should be interpreted to mean "including, but not limited to."

[0077] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any manner in one or more embodiments.

[0078] As used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It should be noted that the term "or" is generally employed in its sense including "and / or" unless the context clearly dictates otherwise.

[0079] In the following description, in order to clearly show the structure and working mode of the present invention, many directional words will be used for description, but words such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and should not be understood as restrictive terms.

[0080] The following describes the implementation details of the embodiments of the present invention in detail with reference to the accompanying drawings. The following content is only provided to facilitate understanding of the implementation details and is not necessary for implementing this solution.

[0081] The flow chart of the bio-inspired neural network affine transformation underwater robot cluster path planning method of this application is as follows Figure 1 As shown, the method includes the following steps:

[0082] S1 Mapping: Map terrain and currents into raster maps.

[0083] S2 Obstacle Detection: Check whether the current grid is an obstacle. If it is an obstacle, go to S3; otherwise, go to S4.

[0084] S3 sets the activity to -1.

[0085] S4 sets the activity to 0.

[0086] S5 End point determination: Check whether the current grid is the end point. If it is the end point, go to S6; otherwise, go to S7.

[0087] S6 sets the activity to 1.

[0088] S7 keeps activity at 0.

[0089] S8 Neuron Activity Propagation: Propagate neuron activity based on AUV formation information.

[0090] S9 obtains the leader path based on activity: a series of path points are obtained as input of the path.

[0091] S10 Third-order B-spline interpolation: Performs third-order B-spline interpolation on the input path points to generate a smooth and continuous path, obtaining the interpolated path.

[0092] S11: Checks whether the interpolated path collides with an obstacle. If so, proceeds to S12; if not, proceeds directly to S15. Collision detection can be performed using geometric calculations or an environmental model to determine whether the interpolated path intersects with an obstacle.

[0093] S12 Obstacle Expansion Processing: If the obstacle is expanded to increase the safety distance, the process proceeds to S13; if the obstacle is not expanded, the process proceeds to S14. To ensure safety, the obstacle is expanded to a certain radius (such as the AUV radius or safety distance), and the path is replanned.

[0094] S13 intermediate point interpolation processing: inserting intermediate points near the expanded obstacle area to adjust the path.

[0095] S14 re-performs third-order B-spline interpolation: performs third-order B-spline interpolation on a new path point set including intermediate points to avoid obstacles, and then enters S15.

[0096] S15 outputs the interpolation navigation path: the final smooth and collision-free interpolation path is used as the output result.

[0097] S16 calculates artificial potential field: obtains AUV formation based on artificial potential field algorithm.

[0098] S17 Formation contraction: Formation contraction is performed based on potential field and affine transformation.

[0099] S18 outputs the follower path.

[0100] The following is a detailed description of the above-mentioned bio-inspired neural network affine transformation underwater robot cluster path planning method.

[0101] like Figure 2 The figure shows the neuron activity propagation diagram of the bio-inspired neural network of the present invention. The activity propagates outward from the end point, is blocked and inhibited by obstacles, and the activity of the leader neuron is affected by the surrounding followers. For easier understanding, please refer to Figure 3 , Figure 3 A bio-inspired neural network structure suitable for cluster AUVs is shown. The present invention first proposes a cluster AUV path planning method based on a bio-inspired neural network. Compared with the existing bio-inspired neural network path planning method under ocean current disturbance conditions, the innovation of the present invention lies in: when the path planning neural network performs neuron activity propagation, the propagation weight is not only affected by the state of the AUV itself, but also by the influence of each surrounding cluster AUV. By comprehensively considering the navigation cost of each AUV in the cluster during navigation, path planning is performed, making the planned path more suitable for cluster AUVs.

[0102] like Figure 3 As shown, first map the topography and ocean currents to the grid map one by one, each grid corresponds to a neuron, and then assign it an initial neuron activity. The activity at the end point is E , the obstacle activity is - E ,in Eis any maximum value. On this basis, the neuron will combine the terrain and the current speed, direction and other information at each point to propagate the activity outward with a certain propagation weight. After the activity propagation iteration is stable, the AUV moves all the way from the starting point along the direction of the neuron activity propagation to the end point. In the present invention, unlike the existing work, when calculating the propagation weight of neuron activity, it is necessary to first determine the grid occupied by the current formation, and then calculate the propagation weight of each occupied neuron by analogy with the traditional method, and then accumulate the propagation weight of each neuron to the neuron where the navigator AUV is located, and use it as the propagation weight of its activity when performing the current step. Such an operation is similar to the field of image processing, where the convolution kernel slides on the picture to obtain a value to represent the characteristics of this area. Specifically, the formula for the neuron activity propagation process is as follows:

[0103] (1)

[0104] (2)

[0105] 、 (3)

[0106] (4)

[0107] (5)

[0108] in, Represents neurons activity, Represents neurons The adjacent non-obstacle neurons, is the activity constraint function, Indicates As the leader, the group goes to The specific calculation formula is shown in (4). express The initial activity value of the point. In formula (2) is a positive constant coefficient, (3) is the propagation coefficient, Represents the neuron activity propagation radius Represents neurons and The distance between (5) is a constant coefficient much larger than 1.

[0109] As the number of iterations As the activity of the neurons increases, the neuron activity gradually converges to a stable state. Based on the activity of each node, we can then follow the propagation direction of the activity from the starting point to the destination. Thanks to the design of the neuron activity propagation, this path is guaranteed to ensure the safety of the pilot and the overall efficiency of the swarm under the influence of ocean currents. However, since the route is only planned for a single pilot AUV, the safety of the swarm AUVs cannot be guaranteed. In narrow terrain, there is a high risk of collision between the safety of a single AUV and the safety of the swarm AUVs. Therefore, an algorithm for changing the formation of the swarm is required to ensure safety.

[0110] Based on the above, the safety of cluster AUVs cannot be guaranteed. In narrow terrain, it is easy to have the problem of single AUV safety and cluster AUV collision risk, such as Figure 4 As shown, the present invention proposes a cluster AUV formation planning method based on artificial potential field and affine transformation. When the AUV formation passes through narrow terrain, it will be affected by the potential field repulsion of obstacles, thereby squeezing the formation, and the formation is scaled by affine transformation to avoid obstacles.

[0111] During the path planning and formation control process for swarm AUVs, the formation must be dynamically adjusted to accommodate the distribution characteristics of obstacles in complex underwater environments. Traditional formation methods typically rely on fixed formation structures and struggle to cope with real-time changes in dynamic obstacle environments. To address this, the present invention proposes combining the affine transformation artificial potential field method with the aforementioned leader path planning method. These two complementary approaches enable swarm AUVs to maintain an orderly formation while possessing efficient obstacle avoidance capabilities and environmental adaptability.

[0112] The artificial potential field method realizes the formation guidance of AUVs by constructing attractive potential fields and repulsive potential fields. For the formation follower AUV, the attractive force mainly comes from the guidance of the formation leader, while the repulsive force comes from the need to avoid environmental obstacles. The total force on an AUV at time T Expressed as:

[0113] (6)

[0114] In formula (6), the attraction is:

[0115] (7)

[0116] The repulsive force is:

[0117] (8)

[0118] in, is the current position of the follower AUV; is the position of the navigator, which can be obtained by the above-mentioned biologically inspired neural network planning; is the nearest obstacle position, which can be obtained from the prior known map; is the distance between the current follower AUV and the obstacle, is the safety distance threshold.

[0119] To ensure the overall structural consistency of the entire formation during obstacle avoidance, the present invention uses affine transformation to dynamically adjust the formation's shape. Affine transformation is a linear transformation between two-dimensional coordinates that preserves the flatness and parallelism of two-dimensional graphics. Affine transformations include rotation and scaling operations. Its core concept is to dynamically control the formation's direction and scale based on environmental information, enabling it to adapt to narrow areas or complex obstacle environments.

[0120] The nominal position of a single follower in an ideal formation is , the actual position after affine transformation Expressed as:

[0121]

[0122] Affine transformation matrix Can be decomposed into a rotation matrix With the scaling matrix A combination of:

[0123]

[0124] Its specific matrix form is as follows:

[0125]

[0126]

[0127] in, and They are formed into teams Axis and Scaling factor in the axis direction, is the overall rotation angle of the formation. The scaling factor and the overall rotation angle are reasonably affected by the artificial potential field. In complex environments, the distribution of obstacles in different directions is heterogeneous, so it is necessary to Axis and The scaling in the axis direction is controlled independently. Specifically, if the obstacles are mainly concentrated in axis direction, the formation should be reduced in Scale on the axis ,Keep Axis scale The method improves the flexibility and passability of the formation in avoiding obstacles in different directions. Based on the resultant force analysis of the artificial potential field method, the total repulsive force exerted by the obstacle on the formation can be obtained by summing the repulsive forces of all AUVs:

[0128]

[0129] in, Indicates the artificial potential field force, Respectively represent the resultant force exist Axis and The component force on the axis. Therefore, the scaling factor is calculated as follows:

[0130]

[0131]

[0132] in, as well as Respectively represent the preset minimum zoom ratio to prevent AUV formations from colliding during zooming. and It is the scaling adjustment coefficient, which controls the sensitivity of scaling to the repulsive force of obstacles. and Respectively reflect the current formation's proximity to or obstruction of obstacles in the x- and y-axis directions.

[0133] In addition, its direction angle Indicates the main direction of obstacle distribution:

[0134]

[0135] In this way, the overall rotation angle of the formation is:

[0136]

[0137] in, represents the yaw angle of the leader AUV, is the rotation sensitivity coefficient. This angle correction strategy enables the formation to automatically adjust its posture when entering a complex obstacle environment, thereby improving its passability.

[0138] This completes the path planning for the Navigator AUV on a grid map. The path consists of a set of discrete points, each adjacent to the other. The turning angles between these points depend entirely on the resolution of the map grid. This can cause the final planned path to appear jagged. While theoretically optimal for the current sea conditions, the numerous twists and turns make it difficult to track the Navigator AUV in practice.

[0139] In some embodiments, the present invention improves the Douglas-Peucker algorithm (DP) by adding a safety detection link to make it applicable to the path simplification problem; and uses third-order B-spline to perform curve fitting on the basis of the simplified path to be more suitable for the AUV second-order system, wherein the third-order B-spline (Cubic B-spline) is a mathematical tool widely used in computer graphics, CAD / CAM, path planning and numerical analysis, etc., for generating smooth curves or surfaces; the Douglas-Peucker algorithm is an efficient algorithm for simplifying polygons or polylines, and its algorithm principle is: the core idea of ​​the algorithm is to recursively split the polyline into two segments, and then remove those points whose deviation from the straight line is less than the threshold according to the set threshold, so as to achieve the purpose of simplifying the polyline, simplifying the path by reducing the number of points while retaining the original shape as much as possible.

[0140] The improved Douglas Peucker DP algorithm proposed in the present invention is as follows, mainly focusing on the third point for secure recursive detection.

[0141] Improved DP algorithm for path post-processing:

[0142] Initialization: Input planning path list and threshold ;

[0143] 1. Find the point in the path farthest from the line connecting the beginning and the end:

[0144] Initialize maximum distance ,index ;

[0145] Traverse all intermediate points in the path:

[0146] Calculate the vertical distance from the point to the line connecting the head and tail ;

[0147] if , then update and ;

[0148] 2. If (Indicates that there are key inflection points that need to be retained):

[0149] Split the path into two parts:

[0150] a. From the starting point to The subpath at

[0151] b. From The subpath to the end point;

[0152] Recursively call the current algorithm to process these two parts;

[0153] Merge two paths (remove duplicate points);

[0154] 3. Otherwise (the path is almost a straight line):

[0155] Check if the line between the first and last points collides with an obstacle:

[0156] a. If there is no collision, return directly to the starting point and end point;

[0157] b. If there is a collision, keep the original path (do not simplify it directly);

[0158] 4. Return the final simplified path.

[0159] Improved DP algorithm for path optimization Figure 5 As shown, it is obvious that as the threshold As the path increases, the jagged feeling of the path gradually decreases, but turning points are still retained in dangerous areas to ensure safety.

[0160] In some embodiments, to ensure that the path of the AUV is second-order continuous and differentiable at each moment, the third-order B-spline curve equation is shown in formula (18):

[0161]

[0162] in, Represents the set of AUV planning path points points within, express degree polynomial basis functions, where represents the non-decreasing node vector, i.e., the discrete points obtained from the path planning in the previous article. The basis function equation can be derived from the Cox-de Boor recursive equation. The Cox-de Boor recursive equation is the core mathematical tool for calculating the B-spline basis function and is used to define the shape and properties of the B-spline curve.

[0163]

[0164]

[0165] In some embodiments, the trajectory tracking control of the AUV generally requires the position, velocity and acceleration information of the AUV at each moment, so a third-order basis function is required. However, due to the nature of the B-spline curve itself, if the discrete control points of the path planning happen to be close to and around the edge of the obstacle, then the fitted B-spline curve can easily conflict with the obstacle, as shown in Figure 6(a), where green is the discrete points of the path, blue is the fitted B-spline curve, and black is the obstacle. In this case, the B-spline trajectory conforms to the planned path points, but cannot avoid obstacles at the bends of the trajectory. Based on this problem, the present application adopts two strategies: boundary obstacle expansion and midpoint interpolation method. The core idea of ​​the boundary obstacle expansion method is to generate an "expansion area" by expanding the boundary of the obstacle outward by a certain distance, thereby providing a safe buffer zone for the AUV to avoid collision; the midpoint interpolation method is to approximate the value of the unknown point by calculating the midpoint value between two known points. First, when an obstacle collision occurs during the B-spline fitting, the locally controllable property of the B-spline curve is exploited to add an expansion obstacle at the critical position, as shown in Figure 6(b). The yellow square is the expansion obstacle. The expansion obstacle causes the top green control point to move upward, further away from the actual obstacle, so the fitted B-spline curve can bypass the obstacle, as shown by the blue curve in Figure 6(b).

[0166] However, in extreme cases, the aforementioned expansion strategy may also be unable to produce a safe path. In this case, midpoint interpolation is necessary. This method interpolates the midpoints of the expanded obstacle with the adjacent control points, as shown by the red dots in Figure 6(b). The interpolated points and the planned path points are then used to produce a safer, but more curvy, B-spline trajectory, as shown by the red curve in Figure 6(b). The conservatism of the boundary obstacle expansion and midpoint interpolation strategies increases, resulting in a B-spline trajectory fitting collision avoidance strategy based on boundary obstacle expansion and midpoint interpolation.

[0167] In some embodiments, the planned path can be obtained according to different planning methods. Figure 7 The results of the bio-inspired planning path, the B-spline smoothing path, and the DP algorithm combined with the B-spline smoothing path are shown. Figure 7 The planned path shown in (a) is rather tortuous and has many discontinuous points, making it difficult for the rudder-controlled AUV to track directly; Figure 7 The path shown in (b) uses the midpoint of (a) as the control point and performs B-spline trajectory smoothing. Although the path is second-order continuous and differentiable, a curved path still appears. Figure 7 The path shown in (c) is By improving the DP algorithm and B-spline trajectory optimization results, the path is made smoother and more continuous, and the collision problem caused by B-spline curve fitting is successfully avoided at the port.

[0168] This application conducts a hexagonal AUV formation simulation experiment in the waters near Dong'ao Island, combined with actual ocean current data. The formation starts from (21.92916469165571°N, 113.70205240151581°E) and heads to (22.00386416627433°N, 113.78519767480674°E). Compared with the existing biologically inspired neural network, the method proposed in this application can plan a path with better overall time consumption and shorter distance. Comparison of planning results and time Figure 8 and Figure 9 , as shown in Table 1.

[0169] Table 1. Consumption time and path length of different planned paths

[0170]

[0171] This application aims to solve the problem of AUV cluster formation planning under current disturbance conditions, improves the existing methods, and proposes a path planning framework based on bio-inspired neural networks, affine transformation, and DP algorithm combined with B-spline algorithm. Compared with the existing algorithms, this framework is more suitable for cluster AUVs, and the planned path has the characteristics of safe obstacle avoidance, short time consumption and short distance.

[0172] Although the present invention has been described in detail with reference to the accompanying drawings and in conjunction with preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, a person of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and such modifications or substitutions shall be within the scope of the present invention or any person skilled in the art who can easily conceive of changes or substitutions within the technical scope disclosed in the present invention shall be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A bio-inspired neural network affine transformation underwater robot cluster path planning method, characterized by: The method comprises: Path planning for swarm autonomous underwater vehicles (AUVs) based on biologically inspired neural networks; Based on the planned path, the artificial potential field method is used to plan the formation of the cluster AUVs, and then the affine transformation is used to dynamically adjust the formation; The planned path is simplified by using Douglas Peucker DP algorithm, and the simplified path is curve fitted by using third-order B-spline; The artificial potential field method is used to plan the formation of cluster AUVs, specifically including: the artificial potential field method guides the formation of AUVs by constructing an attractive potential field and a repulsive potential field; for the formation follower AUV, the attractive force comes from the guidance of the formation leader AUV, and the repulsive force comes from the need to avoid environmental obstacles; The total force on the formation follower AUVs at time T Expressed as: （6） In formula (6), the attraction is: （7） The repulsive force is: （8） in, is the current position of the follower AUV; is the position of the navigator, obtained by the bio-inspired neural network planning; is the nearest obstacle position, obtained from the prior known map; is the distance between the current follower AUV and the environmental obstacles, is the safety distance threshold; The dynamic adjustment of the formation using affine transformation specifically includes: the nominal position of a single follower AUV in the formation is , the actual position after affine transformation for: Affine transformation matrix Decomposed into rotation matrix With the scaling matrix A combination of: Its specific matrix form is as follows: in, and They are formed into teams Axis and Scaling factor in the axis direction, is the overall rotation angle of the formation; if the environmental obstacles are concentrated on axis direction, the reduced formation is in Scale on the axis ,Keep Axis scale unchanged or adjusted appropriately, or vice versa; The Douglas Peucker DP algorithm is used to simplify the planned path, specifically including: Initialize first: enter the planned path list and threshold ; Find the point in the path farthest from the end line: Initialize the maximum distance ,index ; Traverse all intermediate points in the path: calculate the vertical distance from the point to the line connecting the beginning and the end ;if , then update and ; if , divide the path into two parts: a. From the starting point to The subpath at b. The subpath to the end point; Recursively call the current algorithm to process these two parts; merge the two paths; Otherwise, check whether the straight line between the first and last points collides with an obstacle: a. If there is no collision, return directly to the starting and end points; b. If there is a collision, keep the original path; Finally, the final simplified path is returned; The simplified path is subjected to curve fitting using a third-order B-spline, specifically comprising: a third-order B-spline curve equation is shown in formula (18): in, Represents the set of AUV planning path points points within, express degree polynomial basis functions, where represents a non-decreasing node vector, and the basis function equation is as follows: 。 2. The method according to claim 1, characterized in that The biologically inspired neural network specifically includes: firstly, mapping the topographic map and ocean currents one by one to a grid map, each grid corresponds to a neuron, and then giving it an initial neuron activity. The activity at the end is E , the obstacle activity is - E ,in E is any maximum value; the neuron propagates activity outward with the propagation weight. After the activity propagation iteration is stable, the AUV moves from the starting point along the direction of the neuron activity propagation to the end point; Among them, when calculating the propagation weight of neuron activity, the grid occupied by the current AUV formation is first determined, and then the propagation weight of each occupied neuron is calculated, and then the propagation weight of each neuron is accumulated to the neuron where the navigator AUV is located, and it is used as the activity propagation weight of the navigator AUV when performing the current step movement.

3. The method according to claim 2, characterized in that The process of propagating activity along neurons is as follows: in, Represents neurons activity, Represents neurons The adjacent non-obstacle neurons, is the activity constraint function, Indicates As the leader, the group goes to The neuronal propagation activity of the point, express The initial activity value of the point, is a positive constant coefficient, is the propagation coefficient, represents the neuron activity propagation radius, Represents neurons and The distance between is a constant coefficient greater than 1; As the number of iterations As , the neuron activity gradually tends to converge steadily; then, according to the neuron activity at each point, it starts from the starting point and reaches the end point along the direction of neuron activity propagation.

4. The method according to claim 3, characterized in that The dynamic adjustment of the formation using affine transformation also includes: based on the resultant force analysis of the artificial potential field method, the total repulsive force exerted by environmental obstacles on the formation is obtained by summing the repulsive forces of all AUVs: in, Indicates the artificial potential field force, Respectively represent the resultant force exist Axis and The component force on the axis, so the scaling factor is calculated as follows: in, as well as Respectively represent the preset minimum scaling ratio to prevent AUV formations from colliding during scaling. and The scaling adjustment coefficient controls the sensitivity of scaling to the repulsive force of environmental obstacles; and Respectively reflect the degree of proximity or obstruction of the current formation to environmental obstacles in the x and y axis directions.

5. The method according to claim 4, characterized in that The dynamic adjustment of the formation by using affine transformation also includes: using direction angle Indicates the main direction of the distribution of environmental obstacles: ； In this way, the overall rotation angle of the formation is: ； in, represents the yaw angle of the leader AUV, is the rotation sensitivity coefficient; angle correction is used to automatically adjust the posture of the formation when entering a complex obstacle environment.

Images