Underactuated AUV trajectory planning method and system based on neural dynamics structure

Abstract

The present invention relates to a method and system for trajectory planning of an underactuated AUV based on a neural dynamics structure, which is applied to the field of automatic control technology. The method comprises: encoding obstacle and target data into a pre-built map, wherein each node is represented by eight directional neurons; eliminating the connection weights between neurons beyond the steering capability of the AUV, thereby preventing the propagation of neuronal activity in inappropriate directions; in addition, considering the significant impact of water flow, especially lateral water flow, on AUV navigation, the connection weights between neurons in the area affected by the lateral flow are penalized and reduced. The present invention can perform global collision-free trajectory planning for the underactuated AUV under the condition of satisfying the kinematic constraints of the underactuated AUV, avoiding the propagation of neural activity in inappropriate directions, and planning a more direct and efficient path.

Description

Technical Field

[0001] The present invention relates to the technical field of trajectory path planning for an underactuated underwater robot with directional requirements at the starting and target positions, and more particularly to a method and system for trajectory planning of an underactuated AUV based on a neural dynamics structure. Background Art

[0002] Underactuated Autonomous Underwater Vehicles (AUVs) are increasingly being used to inspect underwater infrastructure. Trajectory planning for an underactuated AUV typically involves developing a collision-free trajectory to a target location with less stringent requirements on direction.

[0003] However, in some application scenarios, the directionality requirements of the starting and ending points are critical. For example, due to the field of view limitations of multi-beam sonar and optical cameras, the AUV may need to maintain precise direction within a narrow angle range when approaching certain bridge pier areas to observe underwater structures.

[0004] In order to solve the above problems, it is urgent to provide an under-actuated AUV trajectory planning method and system with directional constraints. Summary of the invention

[0005] In view of the technical problem of under-actuated AUV trajectory planning with directional constraints in the prior art, the present invention provides an under-actuated AUV trajectory planning method and system based on a neural dynamics structure.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a trajectory planning method for underactuated AUV based on a neural dynamics structure, the method comprising:

[0007] Encode obstacle and target data into a data map, wherein the grid cells in the data map include free space grid cells, obstacle grid cells and target grid cells, wherein the free space grid cells are represented by 0, the obstacle grid cells are represented by -E, and the target grid cells are represented by E; each of the grid cells includes eight neurons, which are numbered from 1 to 8 in a counterclockwise order, and each neuron corresponds to a direction pointing to an adjacent grid cell; the neuron in the target grid cell that best matches the desired target direction is designated as the target neuron; each neuron is connected only to eight neurons in the adjacent grid cells it is pointing to, and these connections form the neural dynamic structure;

[0008] Based on the directional requirements of the starting point and the ending point, the neural dynamics structure is modeled as follows:

[0009]

[0010]

[0011] In formula (1), represents the neural activity of neuron m at grid position (i, j) at the next time step, is the neural activity of neuron k at grid position (p,q) at the current time step; M i,j is the map data at the grid position (i, j); is the connection weight between neuron k at position (p,q) and neuron m at position (i,j), determined by equations (2) and (3);

[0012] In formula (2), is the direction difference between neuron k and neuron m, when Less than or equal to ψ lim When the heading weight term υ k is the reciprocal of the cosine of the direction angle, that is Otherwise k is infinite, which ensures that the connection weights between neurons that exceed the steering capability of the AUV are zero, which can suppress the inappropriate spread of neural activity;

[0013] In formula (3), γ is the adjustment parameter, Represents the distance between grid positions (i,j) and (p,q), sc k is the weight term of the angle between the AUV and the water flow direction, which is equal to 1 / cos(s v ), is the lateral part of the water flow relative to the direction of the neuron, e is a natural constant with a value of about 2.718;

[0014] Formula (4) describes the activation function g(x) used in formula (1). The coefficient 0<β≤1 ensures that the output of the activation function is falls within the interval [0,1].

[0015] Furthermore, after planning the global trajectory path based on the neural dynamics structure, a discrete trajectory that meets the AUV's steering constraints is generated, and then the dynamic windowing method DWA is used to plan the velocity for the under-actuated AUV with the nearest neighbor global path point as its target pose.

[0016] Furthermore, the connection weights between neurons in the area affected by the lateral water flow are penalized and reduced.

[0017] On the other hand, the present invention provides an underactuated AUV trajectory planning system based on a neural dynamics structure, the system comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the system is triggered to execute the above method.

[0018] Compared with the prior art, the present invention proposes a novel neural dynamics model, which is a finite directional Hopfield neural network (FDHNN). Unlike the typical neural network path planning method, the model encodes obstacle and target data into a pre-built map, where each grid cell (grid cell on the map) is represented by eight directional neurons, and each neuron is only connected to eight neurons in the neighboring grid cells it points to. The design of FDHNN prevents the propagation of neuronal activity in inappropriate directions by eliminating the connection weights between neurons that are beyond the steering capability of the AUV. In addition, the model takes into account the significant impact of water flow, especially lateral water flow, on AUV navigation, and reduces the impact by penalizing and reducing the connection weights between neurons in the area affected by the lateral flow. FDHNN can perform global collision-free trajectory planning for the under-actuated AUV under the condition of satisfying the kinematic constraints of the under-actuated AUV, avoiding the propagation of neural activity in inappropriate directions, and planning a more direct and efficient path. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a neural network structure provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of a neural network model provided by an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of the appropriate diffusion of neural activity provided by an embodiment of the present invention;

[0022] Figure 4 A trajectory planning result diagram provided by an embodiment of the present invention;

[0023] Figure 5 A global path diagram of FDHNN planning provided by an embodiment of the present invention;

[0024] Figure 6 A global path diagram for DDQN planning in the prior art;

[0025] Figure 7 A planning result diagram of the combination of FDHNN and DWA provided in an embodiment of the present invention;

[0026] Figure 8 This is the planning result diagram of the combination of DDQN and DWA in the prior art;

[0027] Fig. 9 A bridge structure detection path planning diagram in a water flow environment provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] 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 described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0029] 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, ie, should be interpreted as "including, but not limited to."

[0030] References throughout the specification to "one embodiment" or "an embodiment" indicate 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 the 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.

[0031] 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.

[0032] 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 the words "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", "down", etc. should be understood as convenient terms and should not be understood as restrictive terms.

[0033] The implementation details of this embodiment are described in detail below with reference to the accompanying drawings. The following content is only provided for easy understanding of the implementation details and is not necessary for implementing this solution.

[0034] Trajectory planning for an underactuated autonomous underwater vehicle (AUV) typically involves developing a collision-free trajectory to reach a target location, since in some applications, the directionality requirements of the start and end points are critical. The AUV may need to maintain precise direction within a narrow range of angles to observe underwater structures.

[0035] In some embodiments, the trajectory path planning of an underactuated AUV with directional constraints is modeled as follows:

[0036]

[0037] Among them, η s represents the initial position and heading of the AUV, η g represents the target position with soft heading constraint, and U(·) defines the neighborhood space around the target. is the kinematic model of the underactuated AUV, and its detailed definition is given in (2). Where v is the velocity vector containing the forward velocity u and the turning rate r. c and v c are the ocean current velocities in the horizontal and vertical directions on the AUV body coordinate system, respectively. O represents the obstacle set, so is to ensure that the AUV does not collide with obstacles. Therefore, the planning goal is to find the control input: forward speed u and turning rate r to minimize the cost function c(η i ,η,η g ), which usually represents the distance cost to the target.

[0038] In the global planning stage, the time step T is chosen so that the AUV navigates one grid cell per step. The inequality constraints in Equation (1) ensure that the control inputs (e.g., speed, steering, etc.) are within the feasible range that the AUV can execute. and are the maximum and minimum acceleration limits of the AUV’s forward speed, and and are the maximum and minimum acceleration limits of the AUV’s turning rate, respectively. lim is the maximum distance the AUV can travel between two time steps, and ψ lim is the maximum allowable change in heading angle during this time interval.

[0039] Understandably, in the field of trajectory path planning, the environment map is usually discretized into a grid. The neural network model assigns a neuron to each grid cell and plans the path by following the gradient of the neuron activity. However, the direct path generated by these one-to-one neuron mappings may contain sharp turns and cannot meet the constraints of the underactuated characteristics of the AUV.

[0040] In order to overcome this limitation, this embodiment proposes a novel neural dynamics model for the above-mentioned underactuated AUV trajectory planning model with directional constraints. The model is a finite directional Hopfield neural network (FDHNN), which introduces a finite directional neuron architecture into the neural network path planning algorithm. Figure 1 As shown in , the environment containing obstacles and targets is first encoded into a data map. In the data map, free space is represented by 0, obstacle cells are represented by -E, and target cells are represented by E. Different from the typical neural network path planning method, each grid cell contains eight neurons, numbered 1 to 8 in counterclockwise order, and each neuron corresponds to a direction pointing to the neighboring cell. The neuron in the target grid cell that best matches the desired target direction is designated as the target neuron, as shown in Figure 1 The starting neuron is determined in a similar way. In addition, each neuron is connected to only eight neurons in the neighboring grid cells it points to. For example, Figure 1 In the middle grid cell highlighted by the yellow border, neuron 5 is connected only to eight neurons in the obstacle grid, while neuron 8 is connected only to eight neurons in the target grid. Neurons in adjacent grid cells may also be connected to neurons in the central yellow-bordered grid cell. These connections form a neural structure similar to the Hopfield network, which is characterized by neurons that are interconnected but not fully interconnected.

[0041] In some embodiments, the neuron activity update model of the neural dynamics structure is as follows:

[0042]

[0043] The neuron activity in the FDHNN model gradually evolves over time and tends to be stable after a certain number of iterations. The model of the neural network can be described by formulas (3)-(6). In formula (3), represents the neural activity of neuron m at grid position (i, j) at the next time step, is the neural activity of neuron k at grid position (p,q) at the current time step. Each neuron is connected only to eight neurons in the neighboring grid that it points to. i,j is the map data at grid position (i, j).

[0044] is the connection weight between neuron k at position (p, q) and neuron m at position (i, j), which can be determined by formulas (4) and (5). is the direction difference between neuron k and neuron m, such as Figure 2As shown, when Less than or equal to ψ lim When the heading weight term υ k is the reciprocal of the cosine of the direction angle, that is Otherwise k This ensures that the connection weights between neurons that exceed the AUV’s steering capability are zero, thereby suppressing the inappropriate spread of neural activity.

[0045] In formula (5), γ is the adjustment parameter, Represents the distance between grid positions (i,j) and (p,q), sc k is the weight term of the angle between the AUV and the water flow direction, which is equal to 1 / cos(s v ), is the lateral part of the water flow relative to the direction of the neuron, and e is a natural constant with a value of approximately 2.718.

[0046] Formula (6) describes the activation function g(x) used in formula (3), which is similar to the piecewise function proposed by Glasius. The coefficient 0<β≤1 ensures that the output of the activation function is such that falls within the interval [0,1].

[0047] Figure 3 Figure 5 shows the selective diffusion of neural activity in the proposed algorithm. Grid 5 contains eight neurons of different colors, each pointing to an adjacent grid. Neurons are connected only to grids in their direction; for example, the purple neuron in grid 5 is connected only to neurons in grid 2. The grid size allows the AUV to turn 45 degrees, limiting its direction change to only the directions of the three purple neurons in grid 2. The weights of connections that exceed the AUV's turning ability are set to zero, and this rule applies to all neurons.

[0048] In some embodiments, further, after the proposed FDHNN technology plans the global trajectory path, a discrete trajectory that meets the AUV steering constraints is generated, and then a dynamic window approach (DWA) is used to plan the velocity for the underactuated AUV with the nearest neighbor global path point as its target pose.

[0049] Example 1

[0050] The initial simulation is set up in a 6m x 6m enclosed space with obstacles. The AUV starts at coordinates (2,2) with a yaw angle of 135 degrees and is tasked with reaching the target point (5,5) with a target direction of 90 degrees. In the absence of underactuation and orientation constraints, the most direct path is a straight line, which is a solution that many algorithms (such as A*, Dijkstra, and BNN) can easily calculate. However, this straight path is not feasible for the AUV, which is limited by the steering angle. Figure 4 The trajectory drawn when FDHNN is used in conjunction with the dynamic window method DWA is shown. Based on this technology, a feasible path is successfully generated and meets the direction requirements. Among them, the global path generated by FDHNN is as follows Figure 5 shown.

[0051] We compare it with the deep reinforcement learning algorithm DDQN (Double Deep Q-Learning). The path planned by DDQN is as follows Figure 6 As shown in the figure, the DDQN algorithm successfully identified a feasible path after about 300 iterations; the final path after 500 iterations was 0.59 meters longer than the path planned by the FDHNN algorithm.

[0052] Further comparisons of planning effectiveness were conducted in more complex environments. Figure 7 and Figure 8 An example case is shown where DDQN is stuck in a suboptimal path. The navigation distances of FDHNN and DDQN are 28.7 meters and 31.3 meters, respectively, and the corresponding navigation times are 73.2 seconds and 75.0 seconds, respectively. It can be seen that FDHNN combined with DWA can find a more direct path for the underactuated AUV that meets the direction constraints.

[0053] For example, a planning study using FDHNN combined with DWA in a water flow environment was carried out. Take an engineering case as an example: Fig. 9 As shown in the figure, in the bridge structure inspection, the AUV initially detects the underwater structure on the right at the coordinates (12,30), and then needs to detect the underwater structure on the left at the coordinates (47,53). In this case, the task is to plan a suitable trajectory for the AUV to reach the next inspection location. Considering the water flow of the river, the flow rate reaches a maximum of 0.8m / s. The trajectory planned by the algorithm proposed in this scheme has a good effect: at first, the AUV must go upstream, face the water flow, and offset the influence of the water flow. As the journey progresses, the AUV takes advantage of the favorable water flow and eventually reaches an area with weak water flow. In this area, the FDHNN algorithm strategically plans a discrete path based on the influence of the lateral water flow. If the lateral water flow component is not considered, the path of the AUV may intersect with the lateral flow area with strong water flow, causing the AUV to lose control due to its inability to withstand the strong flow. Therefore, the algorithm proposed in this scheme shows high efficiency in waters with different flow dynamics.

[0054] The present invention also provides an under-actuated AUV trajectory path planning system based on a neural dynamics structure, the system comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the system is triggered to execute the aforementioned under-actuated AUV trajectory path planning method based on a neural dynamics structure.

[0055] In summary, the novel neural dynamics model FDHNN provided by the present invention encodes obstacle and target data into a pre-built map, in which each node (grid unit on the map) is represented by eight directional neurons. The design of FDHNN prevents the propagation of neuronal activity in inappropriate directions by eliminating the connection weights between neurons that are beyond the steering capability of the AUV. In addition, the model takes into account the significant impact of water flow, especially lateral water flow, on AUV navigation by penalizing and reducing the connection weights between neurons in the area affected by the lateral flow. FDHNN is able to perform global collision-free trajectory planning for the under-actuated AUV while satisfying the kinematic constraints, avoiding the propagation of neural activity in inappropriate directions, and planning a more direct and efficient path.

[0056] It should be noted that the system of the present invention may also include one or more of the following components: a processor and a memory.

[0057] Optionally, the processor uses various interfaces and lines to connect various parts within the entire autonomous operation equipment, and executes various functions of the autonomous operation equipment and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory, and calling data stored in the memory. Optionally, the processor can be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor can integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a neural network processor (NPU). Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing the content that needs to be displayed on the touch display; and the NPU is used to implement artificial intelligence (AI) functions.

[0058] The memory may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory includes a non-transitory computer-readable storage medium. The memory may be used to store instructions, programs, codes, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the above various method embodiments, etc.; the data storage area may store data created according to the use of the autonomous operating device, etc.

[0059] The present disclosure also provides a computer-readable storage medium storing a computer program, wherein the computer program is used to be executed by a processor to implement the positioning control method of the autonomous working equipment as described in the above embodiment.

[0060] Although the present invention has been described in detail by referring to the accompanying drawings and in combination with the 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 these modifications or substitutions shall be within the scope of the present invention or any person skilled in the art who is familiar with the present invention may easily think of changes or substitutions within the technical scope disclosed by the present invention, and shall be 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 trajectory planning method for underactuated AUV based on neural dynamics structure, characterized in that: The method includes: The obstacle and target data are encoded into a data map, wherein the grid cells in the data map include free space grid cells represented by 0, obstacle grid cells represented by -E, and target grid cells represented by E; each grid cell includes eight neurons, which are numbered from 1 to 8 in counterclockwise order, and each neuron corresponds to a direction pointing to an adjacent grid cell; the neuron in the target grid cell that best matches the desired target direction is designated as the target neuron; each neuron is only connected to the eight neurons in the adjacent grid cells it points to, and these connections form a neural dynamic structure; Based on the directional requirements of the starting and ending points, the neural dynamics structure is modeled as follows: in, is the neural activity of neuron m at grid position (i, j) at the next time step, is the neural activity of neuron k at grid position (p,q) at the current time step; M i,j is the map data at the grid position (i, j); is the connection weight between neuron k at position (p,q) and neuron m at position (i,j); is the direction difference between neuron k and neuron m, when When the heading weight term υ k is the inverse of the cosine of the direction angle, otherwise υ k is infinite, which ensures that the connection weights between neurons that exceed the steering capability of the AUV are zero, which can suppress the inappropriate spread of neural activity; γ is the adjustment parameter, is the distance between grid positions (i,j) and (p,q), sc k is the weight term of the angle between AUV and water flow direction, which is the lateral part of water flow relative to the direction of the neuron, and e is a natural constant; Formula (4) is the activation function g(x) in formula (1), and the coefficient 0<β≤1 ensures that the output of g(x) is such that falls within the interval [0,1].

2. The method according to claim 1, characterized in that The method further includes: after planning the global trajectory path based on the neural dynamics structure, generating a discrete trajectory that meets the AUV steering constraints, and then using a dynamic window method DWA to plan the speed for the under-actuated AUV with the nearest neighbor global path point as its target pose.

3. The method according to claim 1, characterized in that The method further includes: penalizing and reducing the connection weights between neurons in the area affected by the lateral water flow.

4. An underactuated AUV trajectory planning system based on a neural dynamics structure, the system comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein: When the computer program instructions are executed by the processor, the system is triggered to execute the method according to any one of claims 1 to 3.

Images