Underwater robot layered online path planning method based on neurodynamics

Abstract

The invention discloses an underwater robot layered online path planning method based on neurodynamics, which belongs to the field of underwater robot detection, and comprises the following steps: constructing a traditional underwater robot strong dynamic marine environment path planning model; global path planning is carried out based on a neurodynamics mode; a dynamic grid mechanism is introduced into the neural mechanical network; a global-local hierarchical algorithm is introduced into the path planning method for path planning; a fusion local path evaluation mechanism is constructed on the basis of a global-local hierarchical algorithm, the obstacle avoidance performance of the robot in a strong dynamic environment can be guaranteed, the calculation complexity of the algorithm is reduced, and time-saving, energy-saving, efficient and safe path planning is realized.

Description

Technical Field

[0001] The invention relates to the field of underwater robot detection, and in particular to a layered online path planning method for underwater robots based on neurodynamics. Background Art

[0002] At present, autonomous underwater robots have been widely used in various underwater missions. The quality of the navigation path of autonomous underwater robots is directly related to the efficiency of task execution. However, due to the communication difficulties and limited energy supply in the marine environment, autonomous underwater robots cannot obtain sufficient cloud computing supporting services and external energy supply. Autonomous underwater robots can only rely on their limited hardware computing and energy carrying capabilities to independently generate paths to perform tasks in a highly dynamic and complex obstacle environment. Therefore, there are two key issues that need to be urgently addressed. 1) How to formulate a lightweight underwater robot path planning method and enhance the online applicability of the method; 2) How to design an efficient and accurate obstacle avoidance method so that the autonomous underwater robot can quickly and accurately plan the optimal path to bypass complex obstacles, avoiding excessive computational burden due to high algorithm complexity, affecting real-time performance.

[0003] In recent years, in order to reduce the energy consumption of underwater robot navigation, some graph search algorithms (Dijkstra algorithm, A* algorithm and D* algorithm, etc.) have been proposed for path planning of underwater robots. However, when faced with complex graph topology, the path search efficiency of the above graph search algorithms is seriously insufficient, resulting in low path planning efficiency. At the same time, many swarm intelligence optimization algorithms benefit from superior global optimization capabilities, such as double-layer hybrid algorithms, quantum behavior particle swarm optimization and adaptive genetic algorithms, which are proposed to generate feasible paths. However, swarm intelligence optimization algorithms require prior global scene information, which is often inaccurate in dynamic scenes. Moreover, in highly dynamic scenes, the above algorithms need to be frequently restarted to track fast dynamic scenes, resulting in huge computational workload. Therefore, how to design a safe and lightweight method to achieve path planning in dynamic ocean scenarios is a key scientific issue. In addition, in a global environment, the information of sudden dynamic obstacles is difficult to obtain accurately. Therefore, how to use known environmental information to quickly respond to sudden obstacles in a highly dynamic ocean environment is an important part of balancing the safety and efficiency of navigation. Therefore, designing a hierarchical safe path planning algorithm that can quickly respond to dynamic obstacles is another key scientific issue that needs to be urgently addressed to achieve high-quality underwater robot path planning. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for hierarchical online path planning of underwater robots based on neurodynamics. It aims to solve the problem that existing path planning methods are difficult to complete accurate path planning efficiently and lightly in complex environments, and can ensure the obstacle avoidance performance of robots in strong dynamic environments, while reducing the algorithm calculation complexity, and realizing time-saving, energy-saving, efficient and safe path planning.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a layered online path planning method for underwater robots based on neurodynamics, characterized in that: the specific steps are as follows:

[0006] Step 1: Construct a traditional underwater robot path planning model for a strong dynamic marine environment;

[0007] Step 2: Use a neural dynamics-based approach to perform global path planning;

[0008] Step 3: Introduce a dynamic grid mechanism into the neuromechanical network;

[0009] Step 4: Introduce a global-local hierarchical algorithm into the path planning method for path planning;

[0010] Step 5: Construct a fusion local path evaluation mechanism based on the global-local hierarchical algorithm.

[0011] The further improvement of the technical solution of the present invention is that: the specific steps of step 2 are as follows: by simulating the dynamic behavior of the nervous system, the neuron activity value gradient is established according to the path point value and the propagation mode of the neuron activity value, and the neuron with the largest gradient descent is selected as the final generated global optimal path point set. The specific implementation method of neurodynamics is as follows:

[0012]

[0013] where x i (t+1) represents the neural activity value of the current neuron i at iteration t+1. Correspondingly, x j (t) represents the neural activity value of neuron j at the previous iteration t, j∈J={1,2,...,j} represents the neighboring neuron of i, and the connection weight between i and j is w ij , where |P i -P j | is the distance between units i and j, represents the external factor of the dynamic grid, which is defined by the state of unit i, n is the number of grid changes, E is an infinite positive value, and the conversion function g(x) ensures that x j (t)∈[0,1].

[0014] The further improvement of the technical solution of the present invention is that the dynamic grid mechanism in step 3 is specifically a dynamic grid scaling method based on an event trigger mechanism, which dynamically changes the scene and the size of the unit grid according to different driving states of the underwater robot. The specific implementation method is:

[0015]

[0016] in Represents the external factor of the dynamic grid, which is defined by the state of unit i, n is the number of grid changes, and E is an infinite positive value.

[0017] A further improvement of the technical solution of the present invention is that the global-local hierarchical algorithm in step 4 is intended to combine the macro-guidance of global planning and the real-time adjustment capability of local planning to achieve efficient path planning of AUV in complex environments. The path generation method of the global-local hierarchical algorithm is as follows:

[0018] d(p n ,p n+1 )=||p n -p n+1 ||<d th

[0019] x=[X,Y,Z,θ yaw ,θ pitch ,v,ω 1 ,ω 2 ] T

[0020] v∈[max(0,va max ·Δt),min(v max ,v+a max ·Δt)]

[0021] ω 1 ∈[max(-ω max ,ω 1 -α 1,max ·Δt),min(ω max ,ω 1 +α 1,max ·Δt)]

[0022] ω 2 ∈[max(-ω max ,ω 2 -α 2,max ·Δt),min(ω max ,ω 2 +α 2,max ·Δt)]

[0023] where p n ∈P={p 1,p 2 ,...,p n} represents the current path coordinate point set, d th is the distance threshold for starting the local dynamic window algorithm, x is the current motion state matrix of the AUV, which consists of the position coordinates (X, Y, Z), the yaw angle θ yaw , pitch angle θ pitch , linear velocity v, yaw angular velocity ω 1 , pitch angular velocity ω 2 Composition, where v, ω 1 ,ω 2 In the local dynamic window algorithm, the unit time Δt and the maximum linear acceleration a are satisfied. max , maximum angular acceleration α ,max kinematic constraints.

[0024] A further improvement of the technical solution of the present invention is that the fusion local path evaluation mechanism in step 5 is divided into a target evaluation strategy and a safety evaluation strategy.

[0025] The further improvement of the technical solution of the present invention is that the specific evaluation method of the fusion local path evaluation mechanism is as follows:

[0026] G(v,ω 1 ,ω 2 )=a·G heading +b·G dist +c·G velocity +d·G θ +e·G route_heading

[0027] where G(v,ω 1 ,ω 2 ) is the evaluation function of local path selection, a, b, c, d, e are weight parameters, and the heading evaluation function G is set to evaluate the degree of fit between the path guidance direction and the expected heading. heading :

[0028] G heading =CalcHeadingEval(xt,goal)

[0029] In order to ensure navigation safety and reduce the risk of collision during driving, the obstacle distance evaluation function G is set dist Evaluate the safe distance between the current navigation path and the identified obstacles:

[0030]

[0031] In order to ensure that the local path does not deviate from the global path, the global path trajectory evaluation function G is set route_heading :

[0032]

[0033] In order to measure the angle between the current heading and the obstacle, a safety angle evaluation function G is set θ :

[0034]

[0035] In order to ensure that the robot has a higher working efficiency in actual work, the robot needs to have a higher driving speed as much as possible under the premise of ensuring safety. Therefore, the speed evaluation function G is added velocity :

[0036]

[0037] By integrating the local path evaluation mechanism, the path planning result is ensured to meet the robot's kinematic constraints, effectively avoiding falling into the local optimal solution due to the greedy pursuit of a slightly faster navigation speed or a conservative safety distance.

[0038] A further improvement of the technical solution of the present invention is that it also includes step 6: performing a performance evaluation on the path planning method, specifically: conducting several groups of simulation experiments with random start and target positions, and comparing the simulation results with the improved A* algorithm (A*+DWA) combined with the dynamic window algorithm, the improved biologically inspired neural network algorithm (IBINN), and the improved GBNN algorithm (IGBNN).

[0039] Due to the adoption of the above technical scheme, the technical progress achieved by the present invention is: under the premise of accurately describing the scene, a dynamic grid scaling method based on an event trigger mechanism is introduced to construct a neural dynamics dynamic grid network structure, and the scene and unit grid size are dynamically changed according to the driving state of the robot. While ensuring the path accuracy, the calculation complexity is reduced, the path output efficiency is improved, and the problems of formulating a lightweight underwater robot path planning method and enhancing online applicability are solved.

[0040] In complex underwater environments, a global-local hierarchical online path planning method is proposed. Based on comprehensive considerations of the underwater environment, such as the starting and ending points of the mission, the target point, and the distribution of obstacles, a macro-guiding global path is planned for the robot, and local planning is performed in layers to respond quickly based on the robot's real-time position, motion state, and local environmental changes. This method ensures that the robot responds to environmental changes in a timely manner and ensures that it always sails towards the target point. In addition, it avoids repeated calculations of the planned area, thereby improving the efficiency of path output, and effectively guarantees the efficient path planning of autonomous underwater robots in complex environments.

[0041] By building a fusion local path evaluation mechanism, the target evaluation strategy ensures that the robot does not deviate from the global path points when avoiding local obstacles, avoids falling into the local optimal solution, and reaches the mission target point accurately and efficiently; through the safety evaluation strategy, when faced with sudden obstacles that change the target accessibility, the cost of bypassing the obstacle is judged, and the path is locally re-planned when necessary to ensure navigation safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only 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 work.

[0043] Figure 1 It is a schematic diagram of the process of the present invention;

[0044] Figure 2 A comparison chart of the voyage time achieved by the method of the present invention and the comparative method;

[0045] Figure 3 A comparison chart of the sailing distances achieved by the method of the present invention and the comparative method;

[0046] Figure 4 A comparison chart of the average sailing speeds achieved by the method of the present invention and the comparative method;

[0047] Figure 5 It is a comparison chart of the execution time of the method of the present invention and the comparative method; DETAILED DESCRIPTION

[0048] The present invention is further described in detail below in conjunction with embodiments:

[0049] like Figure 1 As shown in FIG. 1 , a flowchart of a hierarchical online path planning method for underwater robots based on neural dynamics is shown. The specific steps are as follows:

[0050] Step 1: Construct a traditional underwater robot path planning model for a strong dynamic marine environment;

[0051] The autonomous underwater robot starts from the starting position S p Set off, autonomously avoid interference from known and sudden underwater dynamic and static obstacles, and navigate to the target position T p The underwater scene can be divided into many identical cells. These cells can be divided into two categories: obstacle cells and free cells. The positions of obstacle cells and free cells can be represented by two sets: and The path of the AUV can be represented by the set The path between two adjacent cells is defined by the straight line segment between their center coordinates. The path length can be calculated using the Euclidean distance. The speed of the autonomous underwater robot can be calculated using the following formula:

[0052] The navigation time from the divided unit i to the unit j can be expressed as follows:

[0053]

[0054] Where |ij| represents the Euclidean distance from unit i to unit j. Therefore, the total navigation time of the autonomous underwater vehicle is:

[0055]

[0056] When the speed provided by the engine of the autonomous underwater vehicle is kept constant, the energy consumption per unit time is also fixed. Therefore, the total energy consumption is equivalent to the total navigation time.

[0057] Step 2: Use a neural dynamics-based approach to perform global path planning;

[0058] This planning method is based on neurodynamics for global path planning. The method of neurodynamics for path planning is to simulate the dynamic behavior of the nervous system, establish the gradient of neuron activity value according to the value of path points and the propagation mode of neuron activity value, and select the neuron with the largest gradient descent as the final generated global optimal path point set. The specific implementation of neurodynamics is as follows:

[0059]

[0060] where x i (t+1) represents the neural activity value of the current neuron i at iteration t+1. Correspondingly, x j (t) represents the neural activity value of neuron j at the previous iteration t. j∈J={1,2,...,j} represents the neighboring neurons of i. In the study, the connection weight between i and j is w ij , where |P i -P j | is the distance between units i and j. represents the external factor of the dynamic grid, which is defined by the state of cell i, and n is the number of grid changes. E is an infinite positive value. The conversion function g(x) ensures that x j (t)∈[0,1].

[0061] Step 3: Introduce a dynamic grid mechanism into the neuromechanical network;

[0062] In the neurodynamic network, each neuron is mapped to a segmented scene unit. The structure of the neurodynamic network is determined by the scene division form, which directly affects the computational workload. In the traditional division form, neurodynamics often maps each neuron to a segmented scene unit and forms a grid structure. Too dense a scene unit grid will greatly increase the number of iterations required for neuron convergence, thereby reducing the efficiency of path output, while too sparse a scene unit grid division will cause the output path to be long, increase the energy waste of the actual driving of the robot, and affect the quality of the output path. This study proposes a dynamic grid scaling method based on an event-triggered mechanism, which dynamically changes the size of the scene and unit grid according to the different driving states of the robot. Where x i (t+1) represents the neural activity value of the current neuron i at iteration t+1. Correspondingly, x j (t) represents the neural activity value of neuron j at the previous iteration t. j∈J={1,2,…,j} represents the neighboring neurons of i. In the study, the connection weight between i and j is w ij , where |P i -P j | is the distance between units i and j. Represents the external factor of the dynamic grid, which is defined by the state of unit i, and n is the number of grid changes. E is an infinite positive value. The proposed lightweight neural dynamic network structure improves the computational efficiency of the proposed algorithm by 25% within ten update iterations in a three-dimensional scene. Compared with the traditional neural dynamic network structure, the global path output efficiency is greatly improved. The following is the specific implementation method of the dynamic grid mechanism:

[0063]

[0064] Step 4: Introduce a global-local hierarchical algorithm into the path planning method for path planning;

[0065] In traditional path planning algorithms, the planning process often focuses on the global environment for unified calculation. In the face of complex and changing environments, such as the highly dynamic and complex obstacle environment in the ocean, a single path planning algorithm often cannot meet the requirements of real-time, accuracy and efficiency. Therefore, a global-local hierarchical algorithm path planning method is proposed, which aims to combine the macro-guidance of global planning and the real-time adjustment capability of local planning to achieve efficient path planning for AUV in complex environments. The path generation method of the proposed global-local hierarchical algorithm is as follows:

[0066] d(p n ,p n+1 )=||p n -p n+1 ||<d th

[0067] x=[X,Y,Z,θ yaw ,θ pitch ,v,ω 1 ,ω 2 ] T

[0068] v∈[max(0,va max ·Δt),min(v max ,v+a max ·Δt)]

[0069] ω 1 ∈[max(-ω max ,ω 1 -α 1,max ·Δt),min(ω max ,ω 1 +α 1,max ·Δt)]

[0070] ω 2 ∈[max(-ω max ,ω 2 -α 2,max ·Δt),min(ω max ,ω 2 +α 2,max ·Δt)]

[0071] where p n ∈P={p 1 ,p 2 ,...,p n} represents the current path coordinate point set. th is the distance threshold for starting the local dynamic window algorithm. x is the current motion state matrix of the AUV, which consists of the position coordinates (X, Y, Z), the yaw angle θ yaw , pitch angle θ pitch , linear velocity v, yaw angular velocity ω 1 , pitch angular velocity ω 2 Among them, v, ω 1 ,ω 2 In the local dynamic window algorithm, the unit time Δt and the maximum linear acceleration a are satisfied. max , maximum angular acceleration α ,max kinematic constraints.

[0072] Step 5: Construct a fusion local path evaluation mechanism based on the global-local hierarchical algorithm.

[0073] On the basis of the global-local hierarchical algorithm path planning method, in order to further optimize the path planning effect, ensure that the path planning results meet the robot's kinematic constraints, and improve the operating efficiency and safety of autonomous underwater robots in complex environments, it is crucial to build a fusion local path evaluation mechanism. This mechanism can conduct a comprehensive and integrated evaluation of the local path planning results, providing a more scientific basis for path selection and adjustment. The specific evaluation methods are divided into target evaluation strategy and safety evaluation strategy. As a key link for underwater robots to autonomously achieve efficient task execution, the target evaluation strategy simultaneously conducts multi-faceted evaluations of the target and global path points to ensure that the robot does not deviate from the global path points during local obstacle avoidance and thus fall into the local optimal solution. This strategy can ensure accurate and efficient arrival at the mission target point. The safety evaluation strategy uses a local algorithm to determine the cost and collision risk of bypassing obstacles when the presence of sudden obstacles may change the accessibility of the target. The specific evaluation methods are as follows:

[0074] G(v,ω 1 ,ω 2 )=a·G heading +b·G dist +c·G velocity +d·G θ +e·G route_heading

[0075] Where G(v,ω 1 ,ω 2 ) is the evaluation function of local path selection. a, b, c, d, e are weight parameters. In order to evaluate the degree of fit between the direction guided by the path and the expected heading, the heading evaluation function G is set heading :

[0076] G heading =CalcHeadingEval(xt,goal)

[0077] In order to ensure navigation safety and reduce the risk of collision during driving, the obstacle distance evaluation function G is set dist Evaluate the safe distance between the current navigation path and the identified obstacles:

[0078]

[0079] In order to ensure that the local path does not deviate from the global path, the global path trajectory evaluation function G is set route_heading :

[0080]

[0081] In order to measure the angle between the current heading and the obstacle, a safety angle evaluation function G is set θ :

[0082]

[0083] In order to ensure that the robot has a higher working efficiency in actual work, the robot needs to have a higher driving speed as much as possible under the premise of ensuring safety. Therefore, the speed evaluation function G is added velocity :

[0084]

[0085] Thanks to the proposed path evaluation selection strategy, it is possible to effectively avoid falling into the local optimal solution due to greedily pursuing a slightly faster navigation speed or a conservative safety distance.

[0086] Step 6: Evaluate the performance of the path planning method.

[0087] To evaluate the performance of the proposed path planning algorithm for autonomous underwater robots based on neurodynamics theory, we conducted 50 simulation experiments with random start and target positions, and compared the simulation results with those of the improved A* algorithm combined with dynamic window algorithm (A*+DWA), the improved biologically inspired neural network algorithm (IBINN), and the improved GBNN algorithm (IGBNN). Figure 2 It is a comparison chart of the navigation time achieved by the method of the present invention and the A*+DWA, IBINN and IGBNN algorithms. Figure 3 It is a comparison chart of the navigation distances achieved by the method of the present invention and the A*+DWA, IBINN and IGBNN algorithms. Figure 4 It is a comparison chart of the safety factors achieved by the method of the present invention and the A*+DWA, IBINN and IGBNN algorithms. Figure 5 It is a comparison chart of algorithm execution time of the method of the present invention and A*+DWA, IBINN and IGBNN algorithms.

[0088] It can be seen from the simulation results that the method proposed in the present invention can achieve a good navigation time T total , total sailing distance NL, safety factor D min and algorithm execution time T CTThe performance is better than A*+DWA, IBINN and IGBNN. Specifically, compared with the directly combined improved A*+DWA, the method proposed in the present invention is more lightweight in algorithm, so the output path is more efficient. For IGBNN and IBINN, although the single global neural dynamics algorithm can efficiently output the path, it cannot respond to unknown sudden obstacles in the path in time, which poses a great safety hazard. In addition, IBINN and IGBNN cannot consider the kinematic constraints of the robot in actual work. In comparison, the algorithm proposed in the present invention can output the path efficiently, smoothly and safely. Therefore, in a highly dynamic and complex underwater environment, the method proposed in the present invention is safer than IBINN and IGBNN, and is better than A*+DWA, IBINN and IGBNN in algorithm execution time. In actual driving, it is more in line with the robot kinematic requirements than A*+DWA, IBINN and IGBNN.

[0089] The embodiments described above are merely descriptions of preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A hierarchical online path planning method for underwater robots based on neurodynamics, characterized by: The specific steps are as follows: Step 1: Construct a traditional underwater robot path planning model for a strong dynamic marine environment; Step 2: Use a neural dynamics-based approach to perform global path planning; Step 3: Introduce a dynamic grid mechanism into the neuromechanical network; Step 4: Introduce a global-local hierarchical algorithm into the path planning method for path planning; Step 5: Construct a fusion local path evaluation mechanism based on the global-local hierarchical algorithm.

2. The method for layered online path planning of an underwater robot based on neurodynamics according to claim 1, characterized in that: Step 2: The specific steps are as follows: By simulating the dynamic behavior of the nervous system, the neuron activity value gradient is established according to the path point value and the propagation mode of the neuron activity value, and the neuron with the largest gradient descent is selected as the final generated global optimal path point set. The specific implementation of neurodynamics is as follows: where x i (t+1) represents the neural activity value of the current neuron i at iteration t+1. Correspondingly, x j (t) represents the neural activity value of neuron j at the previous iteration t, j∈J={1,2,...,j} represents the neighboring neuron of i, and the connection weight between i and j is w ij , where |P i -P j | is the distance between units i and j, represents the external factor of the dynamic grid, which is defined by the state of unit i, n is the number of grid changes, E is an infinite positive value, and the conversion function g(x) ensures that x j (t)∈[0,1].

3. The method for layered online path planning of an underwater robot based on neurodynamics according to claim 1, characterized in that: The dynamic grid mechanism in step 3 is specifically a dynamic grid scaling method based on an event trigger mechanism, which dynamically changes the scene and the size of the unit grid according to different driving states of the underwater robot. The specific implementation method is: in Represents the external factor of the dynamic grid, which is defined by the state of unit i, n is the number of grid changes, and E is an infinite positive value.

4. The method for layered online path planning of an underwater robot based on neurodynamics according to claim 1, characterized in that: The global-local hierarchical algorithm in step 4 aims to combine the macro-guidance of global planning and the real-time adjustment capability of local planning to achieve efficient path planning of AUV in complex environments. The path generation method of the global-local hierarchical algorithm is as follows: d(p n ,p n+1 )=‖p n -p n+1 ||<d th x=[X,Y,Z,θ yaw ,i pitch ,v,ω1,ω2] T v∈[max(0,v-a max ·Δt),min(v max ,v+a max ·Δt)] ω1∈[max(-ω max ,ω1-a 1,max ·Δt),min(ω max ,ω1+a 1,max ·Δt)] ω2∈[max(-ω max ,ω2-a 2,max ·Δt),min(ω max ,ω2+a 2,max ·Δt)] where p n ∈P={p1,p2,...,p n } represents the current path coordinate point set, d th is the distance threshold for starting the local dynamic window algorithm, x is the current motion state matrix of the AUV, which consists of the position coordinates (X, Y, Z), the yaw angle θ yaw , pitch angle θ pitch , linear velocity v, yaw angular velocity ω1, pitch angular velocity ω2, where v, ω1, ω2 satisfy the unit time Δt and the maximum linear acceleration a in the local dynamic window algorithm. max , maximum angular acceleration α ,max kinematic constraints.

5. The method for layered online path planning of underwater robots based on neurodynamics according to claim 1, characterized in that: In step 5, the fused local path evaluation mechanism is divided into a target evaluation strategy and a safety evaluation strategy.

6. The method for layered online path planning of an underwater robot based on neurodynamics according to claim 5, characterized in that: The specific evaluation method of the fusion local path evaluation mechanism is as follows: G(v,ω1,ω2)?·G heading +b·G dist +c·G velocity +d·G θ +e·G route_heading Where G(v,ω1,ω2) is the evaluation function of local path selection, a, b, c, d, e are weight parameters, and the heading evaluation function G is set to evaluate the degree of fit between the direction guided by the path and the expected heading. heading : G heading =CalcHeadingEval(xt,goal) In order to ensure navigation safety and reduce the risk of collision during driving, the obstacle distance evaluation function G is set dist Evaluate the safe distance between the current navigation path and the identified obstacles: In order to ensure that the local path does not deviate from the global path, the global path trajectory evaluation function G is set route_heading : In order to measure the angle between the current heading and the obstacle, a safety angle evaluation function G is set θ : In order to ensure that the robot has a higher working efficiency in actual work, the robot needs to have a higher driving speed as much as possible under the premise of ensuring safety. Therefore, the speed evaluation function G is added velocity : By integrating the local path evaluation mechanism, the path planning result is ensured to meet the robot's kinematic constraints, effectively avoiding falling into the local optimal solution due to the greedy pursuit of a slightly faster navigation speed or a conservative safety distance.

7. The method for layered online path planning of an underwater robot based on neurodynamics according to claim 1, characterized in that: It also includes step 6: performance evaluation of the path planning method, specifically: conducting several groups of simulation experiments with random start and target positions, and comparing the simulation results with the improved A* algorithm combined with the dynamic window algorithm (A*+DWA), the improved biologically inspired neural network algorithm (IBINN), and the improved GBNN algorithm (IGBNN).

Images