Improved adaptive ant colony path planning method and system

Abstract

The invention provides an improved adaptive ant colony path planning method and system, and the method comprises the steps: S1, scanning a global map environment through a laser radar, carrying out the modeling, building a three-dimensional point cloud image, and carrying out the obstacle expansion and map rasterization; s2, initializing ant colony parameters, and performing grid search; s3, performing optimization processing on the state transition probability by utilizing the trained neural network model during searching; s4, performing path search planning according to the optimized state transition probability until the ant reaches a final target point; and S5, repeatedly executing the step S4 until the maximum number of iterations is reached, and completing all path planning. According to the method, the trained neural network model is utilized to optimize the state transition probability, the optimized algorithm can quickly find the optimal path in the iteration process, and the method has higher timeliness and better path planning capability.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to an improved adaptive ant colony path planning method and system. Background Art

[0002] With the development of driverless technology, how to achieve efficient, safe and reliable path planning has become an important issue. Traditional path planning methods often rely on static maps and fixed rules, which are difficult to cope with complex and changeable real-world traffic environments. As an effective heuristic search method, the ant colony optimization algorithm has attracted widespread attention because it can simulate the collective intelligent behavior of nature. By simulating the process of ants looking for food, the algorithm can find an approximate optimal solution when solving path planning problems, and has good global search capabilities and robustness; however, the standard ant colony algorithm is prone to fall into local optimality, and the iteration time is longer when optimizing the path.

[0003] Therefore, based on the above technical problems, it is necessary to design an improved adaptive ant colony path planning method. Summary of the invention

[0004] In view of the defects in the prior art, an object of the present invention is to provide an improved adaptive ant colony path planning method and system.

[0005] The improved adaptive ant colony path planning method provided by the present invention includes:

[0006] Step S1: Use laser radar to scan the global map environment and perform modeling, establish a three-dimensional point cloud map, and perform obstacle expansion and map rasterization;

[0007] Step S2: Initialize the ant colony parameters and perform grid search;

[0008] Step S3: Optimizing the state transition probability using the trained neural network model during the search;

[0009] Step S4: Perform path search planning according to the optimized state transition probability until the ant reaches the final target point;

[0010] Step S5: Repeat step S4 until the maximum number of iterations is reached and all path planning is completed.

[0011] Preferably, step S2 comprises:

[0012] Set the number of ants a and the maximum number of iterations N max , place a ants at the starting point s and start searching;

[0013] The state transition probability formula of the ant colony is:

[0014]

[0015] Among them, τ ij (t) is the pheromone concentration from node i to node j at time t, τ ij (0) = τ0; η ij (t) is the heuristic function from node i to node j at time t; α is the pheromone importance factor, β is the heuristic information importance factor;

[0016] The pheromone update formula is:

[0017] τ ij (t+1)=(1-ρ)τ ij (t)+Δτ ij a

[0018] Among them, ρ is the pheromone volatility factor, Δτ ij a is the pheromone increment left by the ath ant from node i to node j; the heuristic function formula is:

[0019]

[0020] Among them, d ij is the distance between node i and node j.

[0021] Preferably, the step S3 comprises:

[0022] Establish a neural network model with n input layer nodes, m hidden layer nodes, and k output layer nodes, where the position of the current node, the distance between the current node and the next feasible node, and the importance of the current node are used as input layer node factors, and the predicted probability of the current node is used as the output layer node factor;

[0023] For the hidden layer, the forward propagation formula is:

[0024]

[0025] a j =σ(z j )

[0026] Among them, x i is the state feature of the input node, ω ij is the weight coefficient from the input layer to the hidden layer, b j is the bias value of the hidden layer; σ is the activation function, and the Sigmoid function is taken as the activation function, that is:

[0027]

[0028] For the output layer, the propagation formula is:

[0029]

[0030] P ij (t) = σ(z k )

[0031] Among them, ω kj is the weight from the hidden layer to the output layer, α j is the state feature of the hidden node, b k is the bias value of the output layer, P ij is the probability of the actual optimal path.

[0032] Preferably, the probability of the predicted optimal path is Using mean square error as the formula for loss calculation, we have:

[0033]

[0034] Preferably, back-propagation is used to calculate gradients and update weights and biases;

[0035] Calculate the gradient of the output layer:

[0036]

[0037] Calculate the gradient of the hidden layer:

[0038]

[0039] Update weights and biases:

[0040] ω ij '=ω ij -ζδ j x

[0041] b j '=b j -ζδ j

[0042] Where ζ is the learning rate;

[0043] Path search planning is performed based on the optimized state transition probability, and the specific formula is:

[0044] P ij (t) = λa ij (t)+(1-λ)P ij (t)

[0045] Among them, λ is the fusion weight coefficient.

[0046] The improved adaptive ant colony path planning system provided by the present invention comprises:

[0047] Module M1: Use LiDAR to scan the global map environment and build a model, create a 3D point cloud map, expand obstacles, and rasterize the map;

[0048] Module M2: Initialize the ant colony parameters and perform grid search;

[0049] Module M3: Optimize the state transition probability using the trained neural network model during the search;

[0050] Module M4: Perform path search planning based on the optimized state transition probability until the ant reaches the final target point;

[0051] Module M5: Repeatedly trigger module M4 until the maximum number of iterations is reached and all path planning is completed.

[0052] Preferably, the module M2 comprises:

[0053] Set the number of ants a and the maximum number of iterations N max , place a ants at the starting point s and start searching;

[0054] The state transition probability formula of the ant colony is:

[0055]

[0056] Among them, τ ij (t) is the pheromone concentration from node i to node j at time t, τ ij (0) = τ0; η ij (t) is the heuristic function from node i to node j at time t; α is the pheromone importance factor, β is the heuristic information importance factor;

[0057] The pheromone update formula is:

[0058] τ ij (t+1)=(1-ρ)τ ij (t)+Δτ ij a

[0059] Among them, ρ is the pheromone volatility factor, Δτ ij a is the pheromone increment left by the ath ant from node i to node j; the heuristic function formula is:

[0060]

[0061] Among them, d ij is the distance between node i and node j.

[0062] Preferably, the module M3 comprises:

[0063] Establish a neural network model with n input layer nodes, m hidden layer nodes, and k output layer nodes, where the position of the current node, the distance between the current node and the next feasible node, and the importance of the current node are used as input layer node factors, and the predicted probability of the current node is used as the output layer node factor;

[0064] For the hidden layer, the forward propagation formula is:

[0065]

[0066] a j =σ(z j )

[0067] Among them, x i is the state feature of the input node, ω ij is the weight coefficient from the input layer to the hidden layer, b j is the bias value of the hidden layer; σ is the activation function, and the Sigmoid function is taken as the activation function, that is:

[0068]

[0069] For the output layer, the propagation formula is:

[0070]

[0071] P ij (t) = σ(z k )

[0072] Among them, ω kj is the weight from the hidden layer to the output layer, α j is the state feature of the hidden node, b k is the bias value of the output layer, P ij is the probability of the actual optimal path.

[0073] Preferably, the probability of the predicted optimal path is Using mean square error as the formula for loss calculation, we have:

[0074]

[0075] Preferably, back-propagation is used to calculate gradients and update weights and biases;

[0076] Calculate the gradient of the output layer:

[0077]

[0078] Calculate the gradient of the hidden layer:

[0079]

[0080] Update weights and biases:

[0081] ω ij '=ω ij -ζδ j x

[0082] b j '=b j -ζδ j

[0083] Where ζ is the learning rate;

[0084] Path search planning is performed based on the optimized state transition probability, and the specific formula is:

[0085] P ij (t) = λa ij (t)+(1-λ)P ij (t)

[0086] Among them, λ is the fusion weight coefficient.

[0087] Compared with the prior art, the present invention has the following beneficial effects:

[0088] The present invention utilizes a trained neural network model to optimize the state transition probability. The optimized algorithm can find the optimal path more quickly during the iteration process, and has stronger timeliness and better path planning capabilities. The optimized ant colony algorithm is very useful in two-dimensional plane path planning and can meet the actual production needs of factories and enterprises. By utilizing a trained neural network to optimize the state transition probability of the ant colony algorithm, the problem of being easily trapped in a local optimal solution and the problem of too long searching time in a complex environment can be effectively solved, and the search efficiency and quality can be significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0090] Figure 1 is a flow chart of the adaptive ant colony algorithm path planning method based on neural network optimization of the present invention;

[0091] Figure 2 It is a basic structural diagram of the neural network of the present invention;

[0092] Figure 3 It is a schematic diagram of the two-dimensional grid selection of the simulated ant colony of the present invention;

[0093] Figure 4It is a schematic diagram of a two-dimensional map environment model constructed using the grid method.

[0094] Figure 5 It is the trajectory planning path of the traditional ant colony algorithm;

[0095] Figure 6 is the trajectory planning path of the optimized ant colony algorithm of the present invention;

[0096] Figure 7 is the convergence curve of the traditional ant colony algorithm;

[0097] Figure 8 It is the convergence curve of the optimized ant colony algorithm of the present invention. DETAILED DESCRIPTION

[0098] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0099] Example 1

[0100] The present invention provides an improved adaptive ant colony path planning method, such as Figure 1 ,include:

[0101] Step S1: The laser radar scans and models the global map environment and initializes the algorithm parameters;

[0102] Step S2: performing grid search;

[0103] Step S3: While searching, the state transition probability is optimized using the trained neural network model;

[0104] Step S4: Determine whether the target node has been reached;

[0105] Step S5: After reaching the target node, update the optimal path information;

[0106] Step S6: updating the global information of the optimal path;

[0107] Step S7: After reaching the maximum number of iterations, all path planning is completed.

[0108] Furthermore, the unmanned vehicle laser radar is used to scan the environment and create a three-dimensional point cloud map;

[0109] Step 1: Convert the 3D cloud image into a 2D map, expand the obstacles, and rasterize the map. Figure 3 , get the final environment map;

[0110] Step 2: Initialize the ant colony parameters and perform grid search; set the number of ants a and the maximum number of iterations N max . Place a ants at the starting point s and start searching.

[0111] The state transition probability formula of the traditional ant colony is:

[0112]

[0113] Among them, τ ij (t) is the pheromone concentration from node i to node j at time t, τ ij (0) = τ0; η ij (t) is the heuristic function from node i to node j at time t, α is the pheromone importance factor, and β is the heuristic information importance factor.

[0114] Generally, the pheromone update formula is:

[0115] τ ij (t+1)=(1-ρ)τ ij (t)+Δτ ij a

[0116] Among them, ρ is the pheromone volatility factor, Δτ ij a is the increment of pheromone left by the ath ant from node i to node j.

[0117] Generally speaking, the heuristic function formula is:

[0118]

[0119] Among them, d ij is the distance between node i and node j.

[0120] Step 3: Establish a neural network model with n input layer nodes, m hidden layer nodes, and k output layer nodes, such as Figure 2 , where the position of the current node, the distance between the node and the next feasible node, and the importance of this node are used as the node factors of the input layer, and the predicted probability of the current node is used as the node factor of the output layer.

[0121] For the hidden layer, the forward propagation formula is:

[0122]

[0123] a j =σ(z j )

[0124] Among them, xi is the state feature of the input node, ω ij is the weight coefficient from the input layer to the hidden layer, b j is the bias value of the hidden layer, σ is the activation function, and the Sigmoid function is taken as the activation function, that is:

[0125]

[0126] For the output layer,

[0127]

[0128] P ij (t) = σ(z k )

[0129] Among them, ω kj is the weight from the hidden layer to the output layer, b k is the bias value of the output layer.

[0130] Step 4: Let the probability of the actual optimal path be P ij , the predicted probability is The formula for calculating the loss using mean square error is:

[0131]

[0132] Step 5: Use backpropagation to calculate gradients and update weights and biases.

[0133] Calculate the gradient of the output layer:

[0134]

[0135] Calculate the gradient of the hidden layer:

[0136]

[0137] Update weights and biases:

[0138] ω ij =ω ij -ζδ j x

[0139] b j =b j -ζδ j

[0140] Where ζ is the learning rate.

[0141] Step 5: Optimize the state transition probability based on the neural network prediction and the traditional ant colony algorithm. The specific formula is:

[0142] P ij (t) = λaij (t)+(1-λ)P ij (t)

[0143] Among them, λ is the fusion weight coefficient.

[0144] Step 6: The ants perform path search planning based on the optimized state transition probability;

[0145] Step 7: Repeat step 6 until the ant reaches the final target point;

[0146] Step 8: Repeat steps 6 and 7 until the maximum number of iterations is reached.

[0147] Specifically, MATLAB was used to complete the simulation search analysis. Assume that the unmanned vehicle moves in a 30×30m plane with several obstacles of varying widths. From the starting point (1,1) to the terminal (30,30), the grid method is used to complete the environment modeling. The grid block width is 1×1m. For details, see Figure 4 , the specific parameters are shown in Table 1 below:

[0148] Table 1 Simulation parameters

[0149] Parameter name symbol Numeric Number of ants a 200 Maximum number of iterations <![CDATA[N max ]]> 50 Important factors of pheromones α 1.5 Heuristic Information Importance Factor β 6.4 Initial pheromone concentration <![CDATA[τ0]]> 1.5

[0150] In order to verify the superiority of the optimized algorithm, in a given space, by comparing the traditional ant colony algorithm Figure 5 and optimized ant colony algorithm Figure 6 ,like Figure 7 , Figure 8 The convergence curve is obtained, which shows that the optimized algorithm can find the optimal path faster during the iteration process. Therefore, the optimized algorithm has stronger timeliness and better path planning ability.

[0151] Table 2 shows the comparison results obtained from 50 implementations:

[0152] Table 2 Comparison results

[0153] algorithm Traditional Algorithms Optimization Algorithm Average length of optimal path / m 55.69 45.48 Average number of iterations / times 75.41 52.90 Average running time / s 44.27 38.29

[0154] Comparison of the two algorithms shows that, without sacrificing accuracy, the optimized ant colony algorithm is significantly better than the traditional ant colony algorithm in terms of convergence speed, and is very useful in two-dimensional plane path planning, which can meet the actual production needs of factories and enterprises. By using the trained neural network to optimize the state transition probability of the ant colony algorithm, it can effectively solve the problem of being easily trapped in the local optimal solution and the long search time in complex environments, and can significantly improve the search efficiency and quality.

[0155] Example 2

[0156] The present invention also provides an improved adaptive ant colony path planning system, which can be implemented by executing the process steps of the improved adaptive ant colony path planning method, that is, those skilled in the art can understand the improved adaptive ant colony path planning method as a preferred implementation of the improved adaptive ant colony path planning system.

[0157] The improved adaptive ant colony path planning system provided by the present invention includes: module M1: using laser radar to scan the global map environment and perform modeling, establish a three-dimensional point cloud map and perform obstacle expansion and map rasterization; module M2: initializing ant colony parameters and performing grid search; module M3: using a trained neural network model to optimize the state transition probability during the search; module M4: performing path search planning according to the optimized state transition probability until the ant reaches the final target point; module M5: repeatedly triggering module M4 until the maximum number of iterations is reached to complete the entire path planning.

[0158] The module M2 comprises:

[0159] Set the number of ants a and the maximum number of iterations N max , place a ants at the starting point s and start searching;

[0160] The state transition probability formula of the ant colony is:

[0161]

[0162] Among them, τ ij (t) is the pheromone concentration from node i to node j at time t, τ ij (0) = τ0; η ij (t) is the heuristic function from node i to node j at time t; α is the pheromone importance factor, β is the heuristic information importance factor;

[0163] The pheromone update formula is:

[0164] τ ij (t+1)=(1-ρ)τ ij (t)+Δτ ij a

[0165] Among them, ρ is the pheromone volatility factor, Δτ ij a is the pheromone increment left by the ath ant from node i to node j; the heuristic function formula is:

[0166]

[0167] Among them, d ijis the distance between node i and node j.

[0168] The module M3 comprises:

[0169] Establish a neural network model with n input layer nodes, m hidden layer nodes, and k output layer nodes, where the position of the current node, the distance between the current node and the next feasible node, and the importance of the current node are used as input layer node factors, and the predicted probability of the current node is used as the output layer node factor;

[0170] For the hidden layer, the forward propagation formula is:

[0171]

[0172] a j =σ(z j )

[0173] Among them, x i is the state feature of the input node, ω ij is the weight coefficient from the input layer to the hidden layer, b j is the bias value of the hidden layer; σ is the activation function, and the Sigmoid function is taken as the activation function, that is:

[0174]

[0175] For the output layer, the propagation formula is:

[0176]

[0177] P ij (t) = σ(z k )

[0178] Among them, ω kj is the weight from the hidden layer to the output layer, α j is the state feature of the hidden node, b k is the bias value of the output layer, P ij is the probability of the actual optimal path.

[0179] Suppose the probability of the predicted optimal path is Using mean square error as the formula for loss calculation, we have:

[0180]

[0181] Use backpropagation to calculate gradients and update weights and biases;

[0182] Calculate the gradient of the output layer:

[0183]

[0184] Calculate the gradient of the hidden layer:

[0185]

[0186] Update weights and biases:

[0187] ω ij '=ω ij -ζδ j x

[0188] b j '=b j -ζδ j

[0189] Where ζ is the learning rate;

[0190] Path search planning is performed based on the optimized state transition probability, and the specific formula is:

[0191] P ij (t) = λa ij (t)+(1-λ)P ij (t)

[0192] Among them, λ is the fusion weight coefficient.

[0193] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0194] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. An improved adaptive ant colony path planning method, characterized in that: include: Step S1: Use laser radar to scan the global map environment and perform modeling, establish a three-dimensional point cloud map, and perform obstacle expansion and map rasterization; Step S2: Initialize the ant colony parameters and perform grid search; Step S3: Optimizing the state transition probability using the trained neural network model during the search; Step S4: Perform path search planning according to the optimized state transition probability until the ant reaches the final target point; Step S5: Repeat step S4 until the maximum number of iterations is reached and all path planning is completed.

2. The improved adaptive ant colony path planning method according to claim 1, characterized in that: The step S2 comprises: Set the number of ants a and the maximum number of iterations N max , place a ants at the starting point s and start searching; The state transition probability formula of the ant colony is: Among them, τ ij (t) is the pheromone concentration from node i to node j at time t, τ ij (0) = τ0; η ij (t) is the heuristic function from node i to node j at time t; α is the pheromone importance factor, β is the heuristic information importance factor; The pheromone update formula is: t ij (t+1)=(1-ρ)τ ij (t)+Δτ ij a Among them, ρ is the pheromone volatility factor, Δτ ij a is the pheromone increment left by the ath ant from node i to node j; The heuristic function formula is: Among them, d ij is the distance between node i and node j.

3. The improved adaptive ant colony path planning method according to claim 2, characterized in that: The step S3 comprises: Establish a neural network model with n input layer nodes, m hidden layer nodes, and k output layer nodes, where the position of the current node, the distance between the current node and the next feasible node, and the importance of the current node are used as input layer node factors, and the predicted probability of the current node is used as the output layer node factor; For the hidden layer, the forward propagation formula is: a j =σ(z j ) Among them, x i is the state feature of the input node, ω ij is the weight coefficient from the input layer to the hidden layer, b j is the bias value of the hidden layer; σ is the activation function, and the Sigmoid function is taken as the activation function, that is: For the output layer, the propagation formula is: P ij (t)=σ(z k ) Among them, ω kj is the weight from the hidden layer to the output layer, α j is the state feature of the hidden node, b k is the bias value of the output layer, P ij is the probability of the actual optimal path.

4. The improved adaptive ant colony path planning method according to claim 3, characterized in that: Suppose the probability of the predicted optimal path is Using mean square error as the formula for loss calculation, we have:

5. The improved adaptive ant colony path planning method according to claim 4, characterized in that: Use backpropagation to calculate gradients and update weights and biases; Calculate the gradient of the output layer: Calculate the gradient of the hidden layer: Update weights and biases: oh ij '=ω ij -zd j x b j ’=b j -ζδ j Where ζ is the learning rate; Path search planning is performed based on the optimized state transition probability, and the specific formula is: P ij (t)=λa ij (t)+(1-λ)P ij (t) Among them, λ is the fusion weight coefficient.

6. An improved adaptive ant colony path planning system, characterized in that: include: Module M1: Use LiDAR to scan the global map environment and build a model, create a 3D point cloud map, expand obstacles, and rasterize the map; Module M2: Initialize the ant colony parameters and perform grid search; Module M3: Optimize the state transition probability using the trained neural network model during the search; Module M4: Perform path search planning based on the optimized state transition probability until the ant reaches the final target point; Module M5: Repeatedly trigger module M4 until the maximum number of iterations is reached and all path planning is completed.

7. The improved adaptive ant colony path planning system according to claim 6, characterized in that: The module M2 comprises: Set the number of ants a and the maximum number of iterations N max , place a ants at the starting point s and start searching; The state transition probability formula of the ant colony is: Among them, τ ij (t) is the pheromone concentration from node i to node j at time t, τ ij (0) = τ0; η ij (t) is the heuristic function from node i to node j at time t; α is the pheromone importance factor, β is the heuristic information importance factor; The pheromone update formula is: t ij (t+1)=(1-ρ)τ ij (t)+Δτ ij a Among them, ρ is the pheromone volatility factor, Δτ ij a is the increment of pheromone left by the ath ant from node i to node j; The heuristic function formula is: Among them, d ij is the distance between node i and node j.

8. The improved adaptive ant colony path planning system according to claim 7, characterized in that: The module M3 comprises: Establish a neural network model with n input layer nodes, m hidden layer nodes, and k output layer nodes, where the position of the current node, the distance between the current node and the next feasible node, and the importance of the current node are used as input layer node factors, and the predicted probability of the current node is used as the output layer node factor; For the hidden layer, the forward propagation formula is: a j =σ(z j ) Among them, x i is the state feature of the input node, ω ij is the weight coefficient from the input layer to the hidden layer, b j is the bias value of the hidden layer; σ is the activation function, and the Sigmoid function is taken as the activation function, that is: For the output layer, the propagation formula is: P ij (t)=σ(z k ) Among them, ω kj is the weight from the hidden layer to the output layer, α j is the state feature of the hidden node, b k is the bias value of the output layer, P ij is the probability of the actual optimal path.

9. The improved adaptive ant colony path planning system according to claim 8, characterized in that: Suppose the probability of the predicted optimal path is Using mean square error as the formula for loss calculation, we have:

10. The improved adaptive ant colony path planning system according to claim 9, characterized in that: Use backpropagation to calculate gradients and update weights and biases; Calculate the gradient of the output layer: Calculate the gradient of the hidden layer: Update weights and biases: oh ij '=ω ij -zd j x b j ’=b j -ζδ j Where ζ is the learning rate; Path search planning is performed based on the optimized state transition probability, and the specific formula is: P ij (t)=λa ij (t)+(1-λ)P ij (t) Among them, λ is the fusion weight coefficient.

Images