Optimal path planning method of pulse coupling neural network based on double constraints

Abstract

The invention discloses an optimal path planning method of a pulse coupling neural network based on double constraints. The method comprises the following steps: mapping a path planning environment to a DC-PCNN network; a DC-PCNN neural network model is constructed, and all neurons are initialized; activating a target neuron, and recording the current neuron as a father node; calculating an exponential decay function of the torque deviation, and multiplying the calculated value as a penalty factor by an update item of the internal activity item; calculating a gravitational function value of the flow field constraint, and using the calculated value to update a current neuron dynamic threshold value; comparing the internal activity item with a dynamic threshold; when gt; if yes, activating the neuron, and recording the current neuron as a father node; repeating the steps S4-S6 until the initial neuron is activated; and backtracking all activated nodes, and planning an optimal path. According to the method, the search efficiency is remarkably improved while the path optimality is ensured.

Description

An optimal path planning method based on a dual-constraint pulse-coupled neural network Technical Field

[0001] This invention relates to robot optimal path planning technology, specifically to an optimal path planning method based on a dual-constraint pulse-coupled neural network. Background Technology

[0002] Currently, traditional path planning algorithms, due to their limitations and one-sidedness, struggle to simultaneously satisfy efficiency, completeness, and optimality. Artificial neural network algorithms can handle different scenarios, but training models takes a significant amount of time and also suffers from slow convergence, numerous iterations, and a tendency to get trapped in local optima.

[0003] Current path planning algorithms can be divided into global path planning algorithms and local path planning algorithms. Commonly used global path planning algorithms include the A* algorithm and the Random Tree Search (RRT) method. The A* algorithm is a commonly used pathfinding and graph traversal algorithm, but its real-time performance is poor; the planning time increases exponentially with the size or dimensionality of the map. The core idea of ​​the RRT algorithm is to construct a "tree" that explores outwards from the starting point by randomly sampling in the configuration space until a path connecting the starting point is found. Commonly used local path planning algorithms include the artificial potential field method and the D* algorithm. The artificial potential field method is based on the concept of a virtual potential field, simulating the attractive and repulsive forces experienced by the robot in the environment, enabling it to reach the target point while avoiding obstacles. The D* algorithm mainly searches backwards from the target point to the starting point and can efficiently and incrementally correct the original path when the environment (such as obstacles) changes, without completely recalculating.

[0004] In addition, there is another type of path planning algorithm that simulates natural biological evolution or group social behavior through random search. Because it does not rely on gradient information when solving, it is widely used in practical problems such as path planning. Summary of the Invention

[0005] The main objective of this invention is to provide an optimal path planning method based on a pulse-coupled neural network with dual constraints. By constructing a pulse-coupled neural network with dual constraints of torque and flow field function, and utilizing torque deviation combined with the automatic wave characteristics of the pulse-coupled neural network, the search process is guided along the ideal direction. Combined with the flow field function constraint optimization dynamic threshold adjustment strategy, the search efficiency is significantly improved while ensuring path optimality.

[0006] The technical solution adopted in this invention is: an optimal path planning method based on a dual-constraint pulse-coupled neural network, comprising: S1, mapping the path planning environment onto a DC-PCNN network, wherein the number of neurons in the network is consistent with the number of motion nodes, each neuron corresponds one-to-one with a motion node, and the coupling connection between neurons constitutes the robot's movable path; S2, constructing a DC-PCNN neural network model and initializing all neurons; S3, activating the target neuron and recording the current neuron as the parent node; the DC-PCNN neural network begins its iterative process; S4, calculating the exponential decay function of the torque deviation, using this calculated value as a penalty factor, and multiplying it by the update term of the internal activity term. S5. Calculate the gravitational function value of the flow field constraints, and use this calculated value to update the current neuron's dynamic threshold. S6. Compare internal activity items With dynamic threshold ;when > When the starting neuron is activated, the current neuron is recorded as the parent node; S7, repeat steps S4-S6 until the starting neuron is activated; S8, backtrack through all activated nodes and plan the optimal path.

[0007] Furthermore, the method for calculating the torque deviation in step S4 is as follows: S41 Based on the torque function: S42 defines the torque constraint as: The torque penalty factor for S43 is: The penalty factor S44 is used to adjust the updates of internal activity items. .

[0008] Furthermore, the calculation method for the flow field constraint in step S5 is as follows: S51 Based on the gravitational field in the virtual force field space of the flow field: S52 defines the gravitational function as: The S53 flow field penalty factor is:

[0009] The penalty factor S54 is used to adjust the updates of internal activity items: .

[0010] Advantages of this invention: The method of this invention constructs a DC-PCNN model, introducing torque function constraints and flow field function constraints to work collaboratively in the path planning process, each playing a dominant role at different stages. In the early stage of DC-PCNN propagation, torque function constraints play a dominant role, ensuring that neuron activation proceeds in the direction of the starting point and quickly establishing a roughly correct path framework. In the middle stage of propagation, torque function constraints and flow field function constraints work in balance, optimizing path details while maintaining the correct direction. In the later stage of propagation, flow field function constraints become more important, finely adjusting the path to obtain a shorter length and minimize path cost. Results show that, compared with traditional PCNN path planning algorithms, torque function constraints can guide the direction of neuron activation, reduce unnecessary neuron activation, and shorten computation time; by using flow field function constraints to optimize path selection, search efficiency is significantly improved while ensuring path optimality, thus enabling this method to quickly plan high-quality paths in complex obstacle environments.

[0011] This path planning algorithm ensures the completeness and optimality of path planning, and can quickly plan a smooth and optimal feasible path in complex obstacle environments. Compared with traditional path planning methods, it has significantly improved in terms of path quality and search efficiency, and can effectively improve the speed of path planning.

[0012] From the perspective of path practicality, the generated path has good executability, a moderate number of turns, and a smooth change in path curvature. These characteristics are of great significance for practical robot navigation applications.

[0013] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0015] Figure 1 is a flowchart of the optimal path planning method based on DC-PCNN of the present invention; Figure 2 is a topology diagram of the path planning grid map and DC-PCNN neuron mapping (map size is 7×7; actual usage size is 100×100, black represents obstacle areas); Figure 3 is a schematic diagram of the DC-PCNN model; Figure 4 is a schematic diagram of the optimal path planning result of DC-PCNN. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] In traditional PCNN path planning, after the current neuron is activated, neighboring neurons with similar properties are subsequently activated, generating automatic waves that propagate between neurons. These automatic waves search all possible paths in parallel. However, since all neurons have the same threshold, neighboring neurons are activated sequentially, and the computational load increases exponentially with the size of the map, reducing path planning speed. To address this problem, this invention introduces a torque function and a flow field function to build a Dual Constraint Pulse Coupled Neural Network (DC-PCNN) algorithm model. This model controls the propagation speed and direction of the waves, ensuring that neurons within the path planning region from the starting point to the target point are more easily activated, while neurons outside the region are less easily activated or not activated at all. This achieves the goal of reducing computational load, improving path planning speed, and ensuring the completeness of the algorithm. A schematic diagram of the specific process is shown in Figure 1, which will be explained in detail below.

[0018] Specifically, the steps include: S1. Mapping the path planning environment (grid map) onto a DC-PCNN network.

[0019] The number of neurons in the network is consistent with the number of motion nodes, and each neuron corresponds one-to-one with a motion node. The coupling connections between neurons constitute the robot's movable path; S2. Construct the DC-PCNN neural network model and initialize all neurons; S3. Activate the target neuron and record the current neuron as the parent node; the DC-PCNN neural network begins its iterative process; S4. Calculate the exponential decay function of the torque deviation and use this calculated value as a penalty factor, multiplying it by the update term of the internal activity term. S5. Calculate the gravitational function value of the flow field constraints and use this calculated value to update the current neuron's dynamic threshold. S6. Compare internal activity items With dynamic threshold ,when > When the starting neuron is activated, the current neuron is recorded as the parent node; S7. Repeat steps S4-S6 until the starting neuron is activated; S8. Backtrack all activated nodes and plan the optimal path; In step S1, simulate the actual map environment and build a 100*100 grid map. In this map, each node corresponds to a neuron in the DC-PCNN network. The coupling connections between neurons are the robot's movable path. Normally, each neuron is connected to eight surrounding neurons, but in obstacle areas (black nodes), the corresponding neuron is not coupled to surrounding neurons, as shown in Figure 2.

[0020] In step S2, the DC-PCNN neural network model is built (as shown in Figure 3), and each neuron is initialized; , , In step S3, the target neuron is activated, n Record the current neuron as the parent node; , Activated_Neuron[n]=true; the pulse-coupled neural network begins its iterative process; in step S4, a torque function constraint is introduced to guide the neuron activation direction. The torque constraint calculates the deviation between the direction from the current node to the target point and the ideal direction from the starting point to the target point, and penalizes this deviation. When the current direction matches the ideal direction, the internal activity term increases faster, making the neuron easier to activate. When the direction deviation is large, the internal activity term increases slowly, thereby reducing the priority of the neuron's activation and guiding the path search along the ideal direction.

[0021] The mathematical expression for torque is:

[0022] in: : is the torque vector; : is the displacement vector from the axis of rotation to the point of application of the force; : is the force vector; the magnitude of the torque is:

[0023] in: It is the angle between the displacement vector r and the force vector F.

[0024] Using an exponentially decaying function of torque deviation as a penalty factor, multiplied by the update term of the internal activity term, a constraint on the propagation direction of the neural network is achieved. The torque constraint can be defined as:

[0025] in: Neuron The torque; It is the torque constant; It is the ideal direction (the direction from the starting point to the target point); It is the direction from the current neuron i to the target point.

[0026] Convert torque into a penalty factor:

[0027] This penalty factor can be used to adjust the updates of internal activity items:

[0028] Where: C is the amplification factor, and μ is the weighting factor.

[0029] In step S5, by introducing flow field function constraints, the activation threshold of neurons is dynamically adjusted so that neurons closer to the target have lower activation thresholds, thus gaining preferential activation opportunities. The target point acts as an attraction source, generating a gravitational field; obstacles act as a repulsion source, generating a repulsive field; the robot, as a point mass in the field, moves along the direction of "potential energy decrease" under the action of the resultant force; gravitational field:

[0030] in: It is a scale factor; This is the robot's current position. The target location for the robot; This represents the difference between the robot's current position and the target position; the gravitational force is:

[0031] Normalize gravity:

[0032] in: It is the gravitational scale factor; This represents the gravitational pull between the two farthest neurons in the network; and the dynamic threshold in the neural network can be adjusted as:

[0033] Flow field function constraints optimize path length by adjusting dynamic thresholds to make neurons closer to the target more easily activated.

[0034] In step S6, the internal activity items are compared. With dynamic threshold

[0035]

[0036] when When the current node is activated, the neuron is set as the parent node; when... If the starting neuron is not activated, the DC-PCNN network iterates continuously in step S7, repeating steps S4-S6 until the starting neuron is activated, and then terminates the DC-PCNN network iteration. In step S8, based on the activated neurons, the optimal path from the starting point to the target point is backtracked.

[0037] Among them, the DC-PCNN model exhibits unique dynamic characteristics in its neuron activation patterns. The design concept of the dual constraint mechanism is to achieve constraints through the inherent neural network dynamics. The advantage of this intrinsic constraint mechanism is that it avoids the complexity of parameter tuning while maintaining the parallel computing characteristics of PCNN.

[0038] In PCNN, torque function constraints and flow field function constraints are introduced and implemented by adjusting the connection weights and impulse propagation rules between neurons. Torque function constraints are primarily achieved by adjusting the connection strength of neighboring neurons, preferentially activating neurons aligned with the endpoint direction and suppressing neurons with larger directional deviations. Flow field function constraints are achieved by dynamically adjusting the activation threshold of neurons; neurons closer to the target have lower activation thresholds, thus gaining preferential activation opportunities.

[0039] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An optimal path planning method based on a dual-constraint pulse-coupled neural network, characterized in that, include: S1. Map the path planning environment onto a DC-PCNN network. The number of neurons in the network is the same as the number of motion nodes. Each neuron corresponds one-to-one with a motion node, and the coupling connection between neurons is the robot's movable path. S2. Construct the DC-PCNN neural network model and initialize all neurons. S3. Activate the target neuron and record the current neuron as the parent node. The DC-PCNN neural network begins its iterative process. S4. Calculate the exponential decay function of the torque deviation, and use this calculated value as a penalty factor to multiply by the update term of the internal activity term. ； S5. Calculate the gravity function value of the flow field constraints, and use this calculated value to update the current neuron's dynamic threshold. S6. Compare internal activity items With dynamic threshold ;when > When the starting neuron is activated, the current neuron is recorded as the parent node; S7, repeat steps S4-S6 until the starting neuron is activated; S8, backtrack through all activated nodes and plan the optimal path.

2. The optimal path planning method based on a dual-constraint pulse-coupled neural network according to claim 1, characterized in that, The method for calculating the torque deviation in step S4 is as follows: S41 Based on the torque function: S42 defines the torque constraint as: The torque penalty factor for S43 is: The penalty factor S44 is used to adjust the updates of internal activity items. 。 3. The optimal path planning method based on a dual-constraint pulse-coupled neural network according to claim 1, characterized in that, The calculation method for the flow field constraint in step S5 is as follows: S51 Based on the gravitational field in the virtual force field space of the flow field: S52 defines the gravitational function as: The S53 flow field penalty factor is: The penalty factor S54 is used to adjust the updates of internal activity items. 。

Images