A global path planning method based on a random sampling genetic algorithm

Abstract

A global path planning method based on a random sampling genetic algorithm is proposed. The method initializes the population by alternating random sampling and Gaussian sampling to increase the diversity of the initial population and the connectivity of the path segments. An expansion layer penalty factor is added to the fitness function to improve the security of the path. The population is divided into high-fitness and low-fitness populations according to the fitness value. Path crossing operations are performed by determining the encoding repetition rate and integrating a simulated annealing algorithm to avoid falling into local optimal solutions. Deduplication operations are performed to improve population quality by removing redundant parent nodes at a fixed frequency in the early and late stages of iteration.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned vehicle path planning, and more specifically to a global path planning method based on a random sampling genetic algorithm. Background Art

[0002] With the rapid development of global technology, autonomous mobile robot technology has expanded beyond traditional industrial production and has gained widespread application across a wide range of industries, including military, agriculture, healthcare, and logistics and warehousing. Unmanned vehicles, as specialized wheeled mobile robots, consist of environmental perception, decision-making, and control systems, enabling them to autonomously complete tasks, reducing labor costs and the risk of human error.

[0003] Path planning technology is a key indicator of the degree of automation and intelligence in autonomous vehicles, and it also serves as the foundation for completing other tasks. The quality of its algorithms directly impacts the safety of autonomous vehicles during operation and the effectiveness of their missions. Currently, many experts, both domestically and internationally, are dedicated to the research of path planning methods. Commonly used algorithms include Dijkstra's algorithm, A* algorithm, genetic algorithms, and artificial potential field methods. Genetic algorithms are computational models of biological evolution that mimic the natural selection and genetic mechanisms of Darwinian evolution. They search for optimal solutions by simulating the natural evolutionary process. The overall search strategy does not rely on gradient or other auxiliary information during the calculation process, requiring only a target function for the search direction and a corresponding evaluation function. Therefore, it has applications in many scientific fields. However, it is prone to premature convergence and local optimality during the search process.

[0004] The existing technologies are as follows:

[0005] Publication number: CN113219986A, application title: "Robot Global Path Planning Method Based on Genetic Algorithm and Cubic Spline Interpolation." This method uses a multifaceted model representation to represent the robot's environment; two-dimensional coordinates to represent the robot's path; random and directed search strategies to generate an initial path set, thereby ensuring the diversity of the initialization population; a suitable collision detection method based on prior knowledge such as environmental map information and obstacle characteristics; a fitness function with penalties; and genetic manipulation of the initialization population using roulette wheel selection, single-point crossover, and random mutation. Simulation experiments demonstrate that this method is both feasible and effective for solving the global path planning problem of robots. This invention provides a new approach to solving the global path planning problem of mobile robots.

[0006] Its approach determines whether the generated path collides with an obstacle, adding a penalty if a collision occurs. The fitness function is based on the path length and the number of collisions between the path and obstacles. The crossover operation involves randomly selecting two individuals from the population. Our approach, on the other hand, discards path nodes located within the obstacle zone before path generation. The fitness function incorporates a path length factor, a path smoothness factor, and a penalty factor for whether or not the path passes through an expansion layer. Based on the fitness value, the population is divided into high-fitness and low-fitness path zones, from which a path is extracted for crossover. Our method generates paths that maintain a safe distance from obstacles, resulting in safer, smoother, and more consistent with robot dynamics.

[0007] Publication number: CN109799820B, application title: "Local Path Planning Method for Unmanned Vessels Based on Comparative Random Landmark Graphs," includes the following steps: S1: acquiring environmental information; S2: acquiring a starting point and a target point, determining the number of landmarks K and the maximum number of iterations G; S3: using a multi-target collision detection method to obtain an initial set of ship navigation path plans pop0; S4: iteratively processing pop0 using an improved genetic algorithm to obtain a set of iterated paths popm; S5: obtaining the fitness value of each path plan in the iterated path set popm; S6: selecting the shortest ship navigation path plan for output based on the fitness value. This method addresses the problems of existing technologies such as difficulty in storing process data, slow ship response speed, and increased energy consumption.

[0008] The approach used here randomly determines waypoints, applies multi-target collision detection to path segments to generate feasible paths, and then uses traditional crossover and mutation operations to compare randomly generated probabilities with set crossover and mutation thresholds to determine whether to perform crossover and mutation. Our approach, however, combines random sampling with Gaussian sampling to initialize the population. This maintains the randomness of the genetic algorithm while adding directionality. The crossover process considers the code repetition rate to reduce the probability of two overly similar paths crossing over. The mutation operation is replaced by an adaptive deduplication operation, thereby eliminating redundant nodes. Our approach improves the quality of the initial population, reduces the overall number of iterations, and reduces the number of ineffective optimization operations. Summary of the Invention

[0009] To solve the above technical problems, the present invention proposes a global path planning method based on a random sampling genetic algorithm. This method initializes the population by alternating random sampling and Gaussian sampling to increase the diversity of the initial population and the connectivity of the path segments; adds an expansion layer penalty factor to the fitness function to increase the security of the path; divides the population into a high-fitness population and a low-fitness population according to the fitness value, and performs path crossing operations by determining the coding repetition rate and integrating a simulated annealing algorithm to avoid falling into a local optimal solution; and through deduplication operations, redundant parent nodes are deleted at a fixed frequency in the early and late stages of iteration to improve the population quality.

[0010] To achieve the above object, the technical solution adopted by the present invention is:

[0011] The present invention provides a global path planning method based on a random sampling genetic algorithm, the specific steps of which are as follows:

[0012] Step 1: Use LiDAR to map the environment and rasterize the resulting 2D map to expand irregular obstacles.

[0013] Step 2: Set the number of samples, sampling threshold, and node neighborhood step size, perform random sampling and Gaussian sampling according to the number of samples in the grid map, and delete the sampling points that fall in the obstacle area until the number of idle sampling points reaches the sampling threshold;

[0014] Step 3: Connect other adjacent nodes within the neighborhood step of each node and store the formed path into the path set;

[0015] Step 4: Calculate the fitness of each path in the path set, add the two paths with the highest fitness to the temporary path set, and select and eliminate the remaining paths based on their fitness and the set elimination rate. The selected paths are added to the temporary path set.

[0016] Step 5: Divide the path set into high fitness path area and low fitness path area according to the fitness size, and optimize the crossover operation based on the crossover probability, evaluation parameters, encoding repetition rate and simulated annealing idea;

[0017] Step 6: Adaptive deduplication is performed on the path set. The quality of the path is good in the later stage. d The frequency of deduplication is poor in the early stage, and 1.5f d Frequency deduplication;

[0018] Step 7: Repeat the above steps until the individual fitness reaches the set threshold or the number of iterations reaches the upper limit, then the algorithm ends and the optimal global path is found.

[0019] As a further improvement of the present invention, step 2 specifically includes the following process:

[0020] 2.1 Set the number of samples m1, the sampling threshold m2, and the node neighborhood step size l;

[0021] 2.2 Randomly select m1 sampling points in the map. If the sampling point is located in an idle grid, it is a valid sampling point, recorded as q u ;

[0022] 2.3 Use Gaussian sampling and random sampling to cross-sample, and the sampling points q obtained by random sampling u Further Gaussian sampling is performed to q u Define a Gaussian distribution for the center and generate new sampling points q on both sides n1 ,q n2 , if q n1 With q n2 If each of them is located in the idle area and the obstacle area, the new sampling point located in the idle area is recorded as a valid sampling point;

[0023] 2.4 The two samples are cross-sampled until the effective sampling point reaches the sampling threshold m2.

[0024] As a further improvement of the present invention, the step 3 specifically includes the following steps:

[0025] 3.1 Connect to other adjacent nodes within the neighborhood step of each node;

[0026] 3.2 Perform collision detection on each line segment. If it contacts an obstacle, it is considered invalid and removed;

[0027] 3.3 Put all valid paths into the path set.

[0028] As a further improvement of the present invention, step 4 specifically includes the following steps:

[0029] The following steps are involved:

[0030] 4.1 Set the fitness function according to the path length, path smoothness and penalty factor: Fit = ω1*fit1+ω2*fit2-ω3*fit3

[0031] Among them, fit1 is the adaptation factor related to the path length, fit2 is the adaptation factor related to the path smoothness, and fit3 is the penalty factor related to the number of expansion layers occupied by the path;

[0032] 4.2 Sort the paths by fitness, add the two paths with the highest fitness to the temporary path set, and assign the remaining paths a probability of being selected based on their fitness. The higher the fitness, the greater the probability of being selected.

[0033] 4.3 Determine the threshold number of individuals in the temporary path set of this round based on the set elimination rate, and add the selected paths to the temporary path set until the threshold is reached.

[0034] As a further improvement of the present invention, the step 5 specifically includes the following steps:

[0035] Step 5 includes the following steps:

[0036] 5.1 Based on the fitness, the path set is divided into a high fitness path area and a low fitness path area. One path is selected from each area and a crossover operation is performed with the set crossover probability and evaluation parameters. If the encoding repetition rate of the two paths is lower than the evaluation parameter, a crossover operation is performed.

[0037] 5.2 Compare the fitness of the crossover parent and offspring. If the offspring’s fitness is greater than that of the parent, the offspring is selected for the next round of iteration. If the offspring’s fitness is less than that of the parent, the offspring is selected for the next round of iteration with probability P. c Decide whether to select a child for the next iteration, where

[0038]

[0039] Among them, Fit1 is the maximum fitness of the parent generation, Fit2 is the maximum fitness of the offspring generation, and T is the gradually decreasing temperature coefficient.

[0040] As a further improvement of the present invention, step 6 specifically includes the following steps:

[0041] 6.1 The iteration is divided into early iteration and late iteration. The quality of the late path is good, while the quality of the early path is poor. Therefore, the frequency of deduplication in the late iteration is set to f. d , the frequency of early iteration deduplication is 1.5f d ;

[0042] 6.2 Deduplication is performed at the specified number of iterations according to the deduplication frequency. The specific operation is as follows: starting from the starting point, directly connect the path nodes after the path one by one until the line connecting to the t+1th node passes through the obstacle. The tth node is then considered to be the middle point. The previous middle point or starting point is connected to the middle point, and the above operation is performed starting from the tth node and the nodes after it until it is connected to the end point.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention provides a global path planning method based on a random sampling genetic algorithm. The method initializes a population by alternating random sampling and Gaussian sampling, thereby increasing the diversity of the initial population and the connectivity of path segments. An expansion layer penalty factor is added to the fitness function to increase the security of the path. The population is divided into a high-fitness population and a low-fitness population according to the fitness value. A path crossover operation is performed by determining the coding repetition rate and integrating a simulated annealing algorithm to avoid falling into a local optimal solution. A deduplication operation is performed to delete redundant parent nodes at a fixed frequency in the early and late stages of iteration, thereby improving the population quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flowchart based on random sampling genetic algorithm;

[0046] Figure 2 Sets the schematic for the raster map. DETAILED DESCRIPTION

[0047] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0048] As a specific embodiment of the present invention, the present invention provides a random sampling genetic algorithm flow chart as follows Figure 1 A global path planning algorithm based on a random sampling genetic algorithm is shown, comprising the following steps:

[0049] (1) Use laser radar to map the environment and rasterize the obtained two-dimensional map. The schematic diagram of the raster map setting is as follows: Figure 2 As shown, the irregular obstacles are expanded.

[0050] (2) Set the number of samples, sampling threshold, and node neighborhood step size, perform random sampling and Gaussian sampling in the grid map according to the number of samples, and delete the sampling points that fall in the obstacle area until the number of idle sampling points reaches the sampling threshold. The specific steps are as follows:

[0051] Step 1: Set the sampling number m1, sampling threshold m2, and node neighborhood step length l.

[0052] Step 2: Randomly select m1 sampling points in the map. If the sampling point is located in an idle grid, it is a valid sampling point, recorded as q u .

[0053] Step 3: Use Gaussian sampling and random sampling to cross-sample, and obtain the sampling points q by random sampling. u Further Gaussian sampling is performed to q u Define a Gaussian distribution for the center and generate new sampling points q on both sides n1 ,q n2 , if q n1With q n2 If they are located in the idle area and the obstacle area respectively, the new sampling point located in the idle area is recorded as a valid sampling point.

[0054] Step 4: Cross sampling is performed until the valid sampling point reaches the sampling threshold m2.

[0055] (3) Connect other adjacent nodes within the neighborhood step of each node and store the formed paths into the path set. The specific steps are as follows:

[0056] Step 1: Connect other adjacent nodes within the neighborhood step of each node.

[0057] Step 2: Perform collision detection on each line segment. If it contacts an obstacle, it is considered invalid and removed.

[0058] Step 3: Put all valid paths into the path set.

[0059] (4) Calculate the fitness of each path in the original path set, add the two paths with the highest fitness to the temporary path set, and select and eliminate the remaining paths according to their fitness and the set elimination rate. The selected paths are added to the temporary path set. The specific steps are as follows:

[0060] Step 1: Set the fitness function based on path length, path smoothness and penalty factor: Fit = ω1*fit1+ω2*fit2-ω3*fit3

[0061] fit2=arccos((b 2 +c 2 -a 2 ) / 2bc),

[0062] Where d is the total length of the path, a, b, and c are the distances between three adjacent points on the path, n is the number of path grids, Γ = 1 if the grid is an expansion layer grid, and Γ = 0 if the grid is an idle grid, and ω1, ω2, and ω3 are weight coefficients, respectively.

[0063] Step 2: Sort the paths by fitness, add the two paths with the highest fitness to the temporary path set, and assign the probability of selection to the remaining paths based on their fitness. The higher the fitness, the greater the probability of selection.

[0064] Step 3: Determine the threshold number of individuals in the temporary path set of this round according to the set elimination rate, and add the selected paths to the temporary path set until the threshold is reached.

[0065] (5) According to the fitness, the path set is divided into high fitness path area and low fitness path area. The crossover operation is optimized according to the crossover probability, evaluation parameters, coding repetition rate and simulated annealing idea. The specific steps are as follows:

[0066] Step 1: Divide the path set into a high-fitness path area and a low-fitness path area according to the fitness size. Take one path from each area and perform a crossover operation with the set crossover probability and evaluation parameters. If the encoding repetition rate of the two paths is lower than the evaluation parameter, the crossover operation is performed.

[0067] Step 2: Compare the fitness of the crossover parent and offspring. If the offspring’s fitness is greater than that of the parent, the offspring is selected for the next iteration. If the offspring’s fitness is less than that of the parent, the offspring is selected for the next iteration with probability P. c Decide whether to select a child for the next iteration, where

[0068]

[0069] Among them, Fit1 is the maximum fitness of the parent generation, Fit2 is the maximum fitness of the offspring generation, and T is the gradually decreasing temperature coefficient.

[0070] (6) Adaptive deduplication is performed on the path set, and the quality of the later paths is good, with f d The frequency of deduplication is poor in the early stage, and 1.5f d The specific steps are as follows:

[0071] Step 1: The iteration is divided into early iteration and late iteration. The quality of the late path is good, while the quality of the early path is poor. Therefore, the frequency of deduplication in the late iteration is set to f. d , the frequency of early iteration deduplication is 1.5f d ;

[0072] Step 2: Deduplication is performed at the specified number of iterations according to the deduplication frequency: starting from the starting point, directly connect the path nodes after the path one by one until the line connecting to the t+1th node passes through the obstacle. The tth node is then considered to be the middle point, and the previous middle point (starting point) is connected to the middle point. The above operation is repeated from the tth node to the nodes after it until it is connected to the end point.

[0073] (7) Repeat the above steps until the individual fitness reaches the set threshold or the number of iterations reaches the upper limit, then the algorithm ends and the optimal global path is found.

[0074] The above description is merely a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent variation based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A global path planning method based on a random sampling genetic algorithm, comprising the following specific steps, characterized in that: Step 1: Use LiDAR to map the environment and rasterize the resulting 2D map to expand irregular obstacles. Step 2: Set the number of samples, sampling threshold, and node neighborhood step size, perform random sampling and Gaussian sampling according to the number of samples in the grid map, and delete the sampling points that fall in the obstacle area until the number of idle sampling points reaches the sampling threshold; Step 3: Connect other adjacent nodes within the neighborhood step of each node and store the formed path into the path set; Step 4: Calculate the fitness of each path in the path set, add the two paths with the highest fitness to the temporary path set, and select and eliminate the remaining paths based on their fitness and the set elimination rate. The selected paths are added to the temporary path set. The step 4 is specifically The following steps are involved: The following steps are involved: 4.1 Set the fitness function according to the path length, path smoothness and penalty factor: Fit = ω1*fit1+ω2*fit2-ω3*fit3; Among them, fit1 is the adaptation factor related to the path length, fit2 is the adaptation factor related to the path smoothness, and fit3 is the penalty factor related to the number of expansion layers occupied by the path; ω1, ω2, and ω3 are weight coefficients respectively. 4.2 Sort the paths by fitness, add the two paths with the highest fitness to the temporary path set, and assign the remaining paths a probability of being selected based on their fitness. The higher the fitness, the greater the probability of being selected. 4.3 Determine the threshold number of individuals in the temporary path set for this round based on the set elimination rate, and add the selected paths to the temporary path set until the threshold is reached; Step 5: Divide the path set into high fitness path area and low fitness path area according to the fitness size, and optimize the crossover operation based on the crossover probability, evaluation parameters, encoding repetition rate and simulated annealing idea; Step 6: Adaptive deduplication is performed on the path set. The quality of the path is good in the later stage. d The frequency of deduplication is poor in the early stage, and 1.5f d Frequency deduplication; Step 7: Repeat the above steps until the individual fitness reaches the set threshold or the number of iterations reaches the upper limit, then the algorithm ends and the optimal global path is found.

2. The global path planning method based on random sampling genetic algorithm according to claim 1, characterized in that: The step 2 specifically includes the following process: 2.1 Set the number of samples m1, the sampling threshold m2, and the node neighborhood step size l; 2.2 Randomly select m1 sampling points in the map. If the sampling point is located in an idle grid, it is a valid sampling point, recorded as q u ; 2.3 Use Gaussian sampling and random sampling to cross-sample, and the sampling points q obtained by random sampling u Further Gaussian sampling is performed to q u Define a Gaussian distribution for the center and generate new sampling points q on both sides n1 ,q n2 , if q n1 With q n2 If each of them is located in the idle area and the obstacle area, the new sampling point located in the idle area is recorded as a valid sampling point; 2.4 The two samples are cross-sampled until the effective sampling point reaches the sampling threshold m2.

3. The global path planning method based on random sampling genetic algorithm according to claim 1, characterized in that: The step 3 specifically includes the following steps: 3.1 Connect to other adjacent nodes within the neighborhood step of each node; 3.2 Perform collision detection on each line segment. If it contacts an obstacle, it is considered invalid and removed; 3.3 Put all valid paths into the path set.

4. The global path planning method based on random sampling genetic algorithm according to claim 1, characterized in that: The step 5 specifically includes the following steps: Step 5 includes the following steps: 5.1 Based on the fitness, the path set is divided into a high fitness path area and a low fitness path area. One path is selected from each area and a crossover operation is performed with the set crossover probability and evaluation parameters. If the encoding repetition rate of the two paths is lower than the evaluation parameter, a crossover operation is performed. 5.2 Compare the fitness of the crossover parent and offspring. If the offspring’s fitness is greater than that of the parent, the offspring is selected for the next round of iteration. If the offspring’s fitness is less than that of the parent, the offspring is selected for the next round of iteration with probability P. c Decide whether to select a child for the next iteration, where in, Fit1 is the maximum fitness of the parent generation, Fit2 is the maximum fitness of the offspring generation, and T is the gradually decreasing temperature coefficient.

5. The global path planning method based on random sampling genetic algorithm according to claim 1, characterized in that: The step 6 specifically includes the following steps: 6.1 The iteration is divided into early iteration and late iteration. The quality of the late path is good, while the quality of the early path is poor. Therefore, the frequency of deduplication in the late iteration is set to f. d , the frequency of early iteration deduplication is 1.5f d ; 6.2 Deduplication is performed at the specified number of iterations according to the deduplication frequency. The specific operation is as follows: starting from the starting point, directly connect the path nodes after the path one by one until the line connecting to the t+1th node passes through the obstacle. The tth node is then considered to be the middle point. The previous middle point or starting point is connected to the middle point, and the above operation is performed starting from the tth node and the nodes after it until it is connected to the end point.

Images