A robot path planning method and system based on improved gray wolf optimization algorithm
Through the improved gray wolf optimization algorithm, combined with dynamic weighting and reinforcement learning, the problems of low search efficiency and local optimality in robot path planning are solved, and more efficient and accurate path planning is achieved.
Patent Information
- Application Number
- CN202411247795.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-09-06
AI Technical Summary
The existing robot path planning algorithms are inefficient in searching in complex environments, easily fall into local optimal solutions, and have high calculation consumption, especially based on graph search and random sampling algorithms, there are high randomness, non-smooth edges and corners, poor group diversity, and The problem of slow convergence speed.
Using the improved gray wolf optimization algorithm, by introducing adaptive position update of dynamic weights, artificial rabbit optimization algorithm and random center of gravity dynamic reverse learning, a nonlinear adjustment parameter optimization process based on reinforcement learning is constructed, combining dynamic weights and random elements, Enhance the robustness and flexibility of the algorithm.
The search efficiency and accuracy of the algorithm are improved, local optimal traps are avoided, the smoothness and stability of path planning are enhanced, the computational complexity is reduced, and more efficient path planning is achieved.
Smart Images

Figure CN118758320B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of robot path planning, and in particular to a robot path planning method and system based on an improved grey wolf optimization algorithm. Background Art
[0002] Path planning algorithms are categorized into global and local path planning. Global path planning involves planning the optimal trajectory from the robot's current position to its target location, given a static environment where all environmental information is known. This is an offline path planning approach. Local path planning involves the mobile robot, which lacks a complete grasp of the global map, using its onboard sensors to collect environmental information and plan local paths in real time to avoid unknown obstacles. This is a dynamic path planning approach. Numerous global path planning algorithms exist in the related art. These algorithms, based on their planning characteristics, can be broadly categorized as graph search-based path planning algorithms, random sampling-based path planning algorithms, and intelligent bionics-based path planning algorithms. However, graph search-based path planning algorithms are not suitable for planning in complex environments, while random sampling-based path planning algorithms generate highly random paths, are prone to dead zones, and often contain sharp corners and are not smooth. Furthermore, intelligent bionics-based path planning algorithms suffer from poor group diversity, slow convergence, and a tendency to fall into local optima. Furthermore, these algorithms typically require gradient calculations to obtain gradient information or perform complex mathematical transformations to find the optimal solution, which consumes considerable time. Summary of the Invention
[0003] In order to solve the above technical problems, the purpose of the present invention is to provide a robot path planning method and system based on an improved gray wolf optimization algorithm, which can improve the search efficiency of the algorithm and avoid the algorithm falling into local optimality, thereby realizing the planning of the optimal path.
[0004] The first technical solution adopted by the present invention is: a robot path planning method based on an improved gray wolf optimization algorithm, comprising the following steps:
[0005] Get map path feature information;
[0006] Obtain the optimal adaptation parameters of the gray wolf optimization algorithm, introduce the adaptive position update algorithm based on dynamic weights, the artificial rabbit optimization algorithm, and the random center of gravity dynamic reverse learning algorithm to construct an improved gray wolf optimization algorithm;
[0007] Based on the improved grey wolf optimization algorithm, the map path feature information is processed for optimal path planning to obtain the optimal planned path.
[0008] Furthermore, the step of obtaining map path feature information specifically includes:
[0009] The robot's lidar sensor acquires the environmental information of the target area and performs discretization processing to obtain a grid map of the target area;
[0010] Obtain irregular obstacles in the grid map of the target area and perform expansion processing to obtain a grid map with rectangular regular obstacles;
[0011] The grid map with rectangular regular obstacles is divided into grid matrix units and coded and marked to obtain map path feature information.
[0012] Furthermore, the step of obtaining the optimal adaptation parameters of the gray wolf optimization algorithm, introducing the adaptive position update algorithm based on dynamic weights, the artificial rabbit optimization algorithm, and the random center of gravity dynamic reverse learning algorithm to construct an improved gray wolf optimization algorithm specifically includes:
[0013] The nonlinear adjustment parameters in the gray wolf optimization algorithm are selected and processed by a reinforcement learning method to obtain the optimal adaptation parameters of the gray wolf optimization algorithm, wherein the nonlinear adjustment parameters include a nonlinear convergence factor, an energy factor, a first coefficient vector and a second coefficient vector;
[0014] According to the fitness value of the individual position of the gray wolf in the gray wolf optimization algorithm, the individual position of the gray wolf is adaptively updated, and an adaptive position update algorithm based on dynamic weight is constructed;
[0015] Based on the artificial rabbit optimization algorithm, the Lévy flight and Brownian motion strategies are introduced to construct an improved random foraging strategy for the artificial rabbit optimization algorithm;
[0016] Through random center of gravity dynamic reverse learning, the fitness value of the individual position of the gray wolf in the gray wolf optimization algorithm is screened and positioned, and an algorithm based on random center of gravity dynamic reverse learning is constructed;
[0017] According to the optimal adaptation parameters of the gray wolf optimization algorithm, an improved gray wolf optimization algorithm is constructed by combining the adaptive position update algorithm based on dynamic weights, the improved random foraging strategy of the artificial rabbit optimization algorithm, and the dynamic reverse learning algorithm based on random center of gravity.
[0018] Furthermore, the step of selecting and processing the nonlinear adjustment parameters in the gray wolf optimization algorithm by the reinforcement learning method to obtain the optimal adaptation parameters of the gray wolf optimization algorithm specifically includes:
[0019] Determining a search space dimension, a maximum number of iterations, a first nonlinear adjustment parameter, a second nonlinear adjustment parameter, and a benchmark test function;
[0020] Determining a parameter vector and an optimization space for the parameter vector according to the first nonlinear adjustment parameter and the second nonlinear adjustment parameter;
[0021] Randomly assign values to the parameter vector within the optimization space and search space dimensions of the parameter vector to obtain the initial action reward;
[0022] According to the initial action reward, the action corresponding to the first nonlinear adjustment parameter and the second nonlinear adjustment parameter is selected and substituted into the calculation formula of the nonlinear convergence factor and the benchmark test function is optimized to obtain the reward value of the benchmark test function;
[0023] Based on the reward value of the benchmark function, it is iteratively updated through the Q-learning update formula until the maximum number of iterations is met, and the optimal adaptation parameters of the gray wolf optimization algorithm are output.
[0024] Furthermore, the calculation formula of the nonlinear convergence factor is specifically expressed as follows:
[0025] ;
[0026] In the above formula, represents the convergence factor, represents the initial value of the convergence factor, represents the termination value of the convergence factor, represents the first nonlinear adjustment parameter, represents the second nonlinear adjustment parameter, Indicates the current iteration number, Indicates the maximum number of iterations.
[0027] Furthermore, the expression of the adaptive position update algorithm based on dynamic weight is specifically as follows:
[0028] ;
[0029] In the above formula, represents the updated position of the gray wolf individual, represents the position update weight of the first gray wolf individual, represents the current position of the first gray wolf individual, represents the position update weight of the second gray wolf individual, represents the current position of the second gray wolf individual, represents the position update weight of the third gray wolf individual, Indicates the current location of the third gray wolf individual, 、 represents a random number, Indicates the weight of the current position at different iteration times, Indicates the current location of the gray wolf individual, Represents a random individual in the population with a better fitness value than the current position, Represents a constant that controls the amount of randomness.
[0030] Furthermore, the expression of the random foraging strategy of the improved artificial rabbit optimization algorithm is specifically as follows:
[0031] ;
[0032] In the above formula, Indicates the The position of the candidate solution at the iteration, Indicates the The current position at the iteration, 、 represents a random number between [0,1], represents the constant that controls the magnitude of the Brownian motion effect, 、 represents random numbers that follow a standard normal distribution, represents the energy factor, represents the Lévy flight distribution function, represents the dimension of the problem, Represents a random individual in the population with a better fitness value than the current position, represents the running operator, Indicates rounding to the nearest integer. represents the random walk coefficient of Brownian motion.
[0033] Furthermore, the step of performing optimal path planning processing on the map path feature information based on the improved Grey Wolf Optimization Algorithm to obtain the optimal planned path specifically includes:
[0034] Initialize the gray wolf population, nonlinear convergence factor, energy factor, first coefficient vector and second coefficient vector, where the gray wolf individuals in the gray wolf population represent discrete path nodes;
[0035] Calculate the fitness values of the gray wolf individuals in the gray wolf population and sort them, and select and store the corresponding gray wolf individuals with the top three fitness values;
[0036] The first candidate solution of the gray wolf individual is obtained through an adaptive position update algorithm based on dynamic weights, and the nonlinear convergence factor is judged;
[0037] If the nonlinear convergence factor is less than 1, the second candidate solution of the gray wolf individual is obtained by improving the random foraging strategy of the artificial rabbit optimization algorithm, and the fitness value of the second candidate solution of the gray wolf individual is obtained;
[0038] Comparing the fitness value of the candidate solution of the gray wolf individual with the fitness value of the gray wolf individual, if the fitness value of the candidate solution is better, then updating the gray wolf individual with the candidate solution with the best fitness value, and screening it through the random center of gravity dynamic reverse learning algorithm to obtain the optimized gray wolf individual position, the candidate solutions including the first candidate solution and the second candidate solution;
[0039] The nonlinear convergence factor and energy factor are updated, and the positions of gray wolf individuals are iteratively optimized based on the updated nonlinear convergence factor and the updated energy factor until the preset number of iterations is met. The gray wolf population composed of the optimal gray wolf individual positions is output to obtain the optimal planning path.
[0040] The second technical solution adopted by the present invention is: a robot path planning system based on an improved gray wolf optimization algorithm, comprising:
[0041] Acquisition module, used to obtain map path feature information;
[0042] A module is constructed to obtain the optimal adaptation parameters of the gray wolf optimization algorithm. The adaptive position update algorithm based on dynamic weights, the artificial rabbit optimization algorithm, and the random center of gravity dynamic reverse learning algorithm are introduced to construct an improved gray wolf optimization algorithm.
[0043] The planning module is used to perform optimal path planning processing on the map path feature information based on the improved gray wolf optimization algorithm to obtain the optimal planned path.
[0044] The beneficial effects of the method and system of the present invention are as follows: the present invention obtains map path feature information to further obtain the optimal adaptation parameters of the gray wolf optimization algorithm, introduces an adaptive position update algorithm based on dynamic weights, an artificial rabbit optimization algorithm, and a random center of gravity dynamic reverse learning algorithm to construct an improved gray wolf optimization algorithm, and uses reinforcement learning theory to learn and optimize the parameter configuration process in the gray wolf optimization algorithm, and intelligently selects nonlinear adjustment parameters. The adaptive position update based on dynamic weights considers randomly selecting another individual with better fitness value than the current one in the population and the information of the individual with the best fitness value to guide the search of candidate individuals. The introduction of the artificial rabbit optimization algorithm is conducive to a more refined and continuous search of the solution space when approaching the optimal solution in the later stage of iteration, helping the algorithm to find a better solution in the local area to improve the accuracy of the algorithm. The random center of gravity dynamic reverse learning algorithm can enhance the robustness and flexibility of the algorithm by introducing random elements and dynamics, thereby improving the search efficiency of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 This is a flowchart of the steps of a robot path planning method based on an improved gray wolf optimization algorithm of the present invention;
[0046] Figure 2 This is a structural block diagram of a robot path planning system based on an improved gray wolf optimization algorithm of the present invention;
[0047] Figure 3 This is a schematic diagram of optimal path planning based on the improved grey wolf optimization algorithm provided by a specific embodiment of the present invention;
[0048] Figure 4 1 is a schematic diagram showing the convergence comparison of different algorithms provided by a specific embodiment of the present invention on the F1 benchmark test function;
[0049] Figure 5 1 is a schematic diagram showing the convergence comparison of different algorithms provided by a specific embodiment of the present invention on the F2 benchmark test function;
[0050] Figure 6 1 is a schematic diagram showing the convergence comparison of different algorithms provided by a specific embodiment of the present invention on the F3 benchmark test function;
[0051] Figure 7 1 is a schematic diagram showing the convergence comparison of different algorithms provided by a specific embodiment of the present invention on the F4 benchmark test function;
[0052] Figure 8 1 is a schematic diagram showing the convergence comparison of different algorithms provided by a specific embodiment of the present invention on the F5 benchmark test function;
[0053] Figure 9 1 is a schematic diagram showing the convergence comparison of different algorithms provided by a specific embodiment of the present invention on the F6 benchmark test function;
[0054] Figure 10 This is a schematic diagram comparing eight algorithm routes on a 30×30 grid map provided by a specific embodiment of the present invention;
[0055] Figure 11 This is a schematic diagram comparing the convergence curves of eight algorithms on a 30×30 grid map provided by a specific embodiment of the present invention;
[0056] Figure 12 This is a schematic diagram comparing eight algorithm routes on a 40×40 grid map provided by a specific embodiment of the present invention;
[0057] Figure 13 This is a schematic diagram comparing the convergence curves of eight algorithms on a 40×40 grid map provided by a specific embodiment of the present invention. DETAILED DESCRIPTION
[0058] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are provided for ease of description only and do not limit the order of the steps. The order of execution of the steps in the embodiments can be adaptively adjusted based on the understanding of those skilled in the art.
[0059] Reference Figure 1 The present invention provides a robot path planning method based on an improved gray wolf optimization algorithm, the method comprising the following steps:
[0060] S100, obtaining map path feature information;
[0061] Specifically, the environmental information of the target area is acquired through the robot lidar sensor and discretized to obtain a grid map of the target area; the irregular obstacles in the grid map of the target area are acquired and expanded to obtain a grid map with regular rectangular obstacles; the grid map with regular rectangular obstacles is divided into grid matrix units and coded and marked to obtain map path feature information.
[0062] In an embodiment of the present invention, in an unknown environment, the mobile robot first scans and detects through the onboard sensors such as lidar to obtain environmental information of the logistics warehouse, and then discretizes the environmental model into a grid map. The irregular obstacles in the environment are expanded and set as rectangular regular obstacles. The actual working environment is equivalent to a grid matrix composed of grids of the same size, and the bounded obstacle grid cells and non-obstacle grid cells are marked with 1 and 0 codes respectively.
[0063] S200, obtaining the optimal adaptation parameters of the gray wolf optimization algorithm, introducing an adaptive position update algorithm based on dynamic weights, an artificial rabbit optimization algorithm, and a random center of gravity dynamic reverse learning algorithm to construct an improved gray wolf optimization algorithm;
[0064] Specifically, traditional algorithms generally require gradient operations to obtain gradient information for solution or perform complex mathematical transformation operations when solving the optimal solution, which consumes a lot of time. In contrast, the bio-inspired algorithm only needs to operate the solution vector of the equation according to the heuristic rules within the maximum number of iterations to finally obtain the optimal solution and its corresponding solution vector. When the computing task does not have too high requirements for the solution accuracy, the number of iterations can be appropriately reduced to reduce the solution time and finally obtain an approximate solution close to the global optimum. In recent years, the grey wolf optimizer (GWO) has been developed as a biomimetic intelligent algorithm with the advantages of simple structure, few parameters, and strong global search capabilities. In summary, the embodiment of the present invention adopts the bio-inspired algorithm grey wolf optimization algorithm to optimize and solve the mobile robot in logistics warehousing tasks.
[0065] S210, performing matching processing on nonlinear adjustment parameters in the gray wolf optimization algorithm by a reinforcement learning method to obtain optimal adaptation parameters of the gray wolf optimization algorithm, wherein the nonlinear adjustment parameters include a nonlinear convergence factor, an energy factor, a first coefficient vector, and a second coefficient vector;
[0066] Specifically, swarm intelligence optimization algorithms generally face the problem of balancing global and local search capabilities, and the Gray Wolf Optimization Algorithm is no exception. When |A| ≥ 1, the wolf pack moves away from the prey and conducts a global search across the entire search space. Stronger global search performance effectively helps the algorithm avoid falling into local optimal solutions. When |A| < 1, the wolf pack uses the collected information to conduct precise searches in local areas and gradually approach the prey. Strong local development performance can improve the algorithm's solution accuracy and accelerate its convergence. The magnitude of |A| is determined by the convergence factor a. The convergence factor a of the traditional Gray Wolf Optimization Algorithm decreases linearly from 2 to 0 with iterations. However, this linear change does not reflect the actual search process and is therefore insufficient for complex optimization problems. Therefore, an embodiment of the present invention proposes a new method for determining nonlinear adjustment parameter pairs in a nonlinear function. By utilizing reinforcement learning theory to learn and optimize the parameter configuration process in the Gray Wolf Optimization Algorithm, nonlinear adjustment parameters can be intelligently selected without relying on historical experience.
[0067] In the embodiment of the present invention, the algorithm parameters, the search space dimension D, the maximum number of iterations T are set, and finally the nonlinear adjustment parameters in the new nonlinear convergence factor are learned by reinforcement learning. 、 For intelligent matching, the calculation formula of the nonlinear convergence factor of this solution is as follows:
[0068] ;
[0069] In the above formula: and Convergence factors Initial and final values, is the current iteration number, is the maximum number of iterations, and is the nonlinear adjustment coefficient.
[0070] The steps for reinforcement learning parameter optimization are as follows:
[0071] 1) The present invention selects the unimodal function Step and the multimodal function Penalized1, which are greatly affected by the parameter change algorithm solution performance, as benchmark test functions.
[0072] 2) Reinforcement learning falls under the umbrella of machine learning, learning from experience by mimicking the human skill acquisition process. During the learning iteration process, the agent estimates the accumulated reward and current state, selects the next appropriate action based on the highest Q-value, and stores this in the Q-table, along with the formula for calculating Q-value updates. As the agent interacts with the environment, the environment provides corresponding rewards based on the agent's actions. The knowledge the agent learns is stored in the Q-table, which is continuously updated, as shown in Table 1.
[0073] Table 1 Framework of Q table
[0074]
[0075] By repeatedly selecting actions, executing actions, obtaining rewards, and updating the Q table until the stopping condition is met, the optimal strategy is finally found. The calculation formula for the Q value update is as follows:
[0076] ;
[0077] In the above formula, the variable 、 Represent the current state and the next state respectively. For the current action, is the discount factor, is the learning rate, is Execute actions in status The rewards you get when In state Next action The estimated Q value at . During the algorithm execution, the specific parameter selection and optimization process are described as follows:
[0078] Reinforcement learning is performed on the unimodal function Step and the multimodal function Penalized1 for 1000 training iterations, as shown below:
[0079] Constructed by nonlinear adjustment parameters 、 The parameter vector W (W=[ , ]), set the parameter optimization space to [1,10], the parameter vector W in the process of reinforcement learning 、 The change step size is 1, and finally the initial parameter vector Randomly assign values within the search range and initialize the Q table so that the Q of each state-action pair is initialized to 0.
[0080] In each state, select according to the Q value table 、 Action , and substitute the latest parameter vector W into the nonlinear factor and optimize the Step (Penalized1) benchmark function to obtain the corresponding reward value according to the objective function value.
[0081] Observe the new state and reward and update the Q value using the Q value update calculation formula.
[0082] Repeat the above process until the predetermined termination condition is reached, and finally the optimal parameter configuration is obtained.
[0083] S220, performing adaptive position updating processing on the individual gray wolf positions according to the fitness values of the individual gray wolf positions in the gray wolf optimization algorithm, and constructing an adaptive position updating algorithm based on dynamic weights;
[0084] Specifically, in the basic GWO, the top fitness rankings are recorded during initialization. 、 and Wolf. During the iteration process, if individuals with higher fitness appear, they will replace the existing 、 and Wolf. The wolf does not represent the location of the prey, its location is actually the optimal solution or local optimal solution in the current search space. 、 and The wolves closest to the prey are used to estimate the prey's approximate location. However, if all wolves are trapped in a local optimum, and the ω wolves in the population still tend to the location of these wolves, this means that the pack's leadership has misjudged the prey's location, and the population cannot fully explore the search space and converges prematurely, making it difficult to find a better solution. To enhance the algorithm's ability to escape local optima and maintain population diversity, this paper proposes an adaptive position update formula, as shown below:
[0085] ;
[0086] The information of randomly selecting another individual with better fitness value than the current one in the population and the individual with the best fitness value is considered to guide the search of candidate individuals.
[0087] It is a reflection and The weight of influence at different iteration moments, The calculation formula of weight is as follows:
[0088] ;
[0089] The weight indicates that in the initial stage of the search, the position update equation is more influenced by the information of random excellent individuals, making it random enough to use more useful information for random exploration. In the later stage of the search, more emphasis is placed on the role of the optimal individual, and a more refined local search is performed to generate more promising candidate individuals. It is a constant that controls the magnitude of randomness, and is set to 0.3 in the embodiment of the present invention. and is a random number between (0, 1), is the maximum number of iterations, is the current iteration number, 、 、 express 、 、 The wolf's fitness value. In the traditional gray wolf optimization algorithm, the position iteration update formula 、 、 The average value of the wolf's position is used to update the position of the entire population. The three leadership individuals have the same guidance strength for the group, but the gray wolf optimization algorithm itself is an algorithm based on social hierarchy, so the same guidance makes the algorithm converge slowly. In this regard, the present invention proposes a dynamic proportional weight strategy based on fitness value, which highlights the 、 、 The importance of the three wolves is calculated based on the proportional weight of their fitness values, as shown below:
[0090] ;
[0091] ;
[0092] ;
[0093] Improving the position update formula can make the algorithm converge to the optimal solution faster, improve the efficiency and performance of the algorithm, and the adaptive position update formula based on dynamic weights is as follows:
[0094] ;
[0095] In the above formula, represents the updated position of the gray wolf individual, represents the position update weight of the first gray wolf individual, represents the current position of the first gray wolf individual, represents the position update weight of the second gray wolf individual, represents the current position of the second gray wolf individual, represents the position update weight of the third gray wolf individual, Indicates the current location of the third gray wolf individual, 、 represents a random number, Indicates the weight of the current position at different iteration times, Indicates the current location of the gray wolf individual, Represents a random individual in the population with a better fitness value than the current position, Represents a constant that controls the amount of randomness.
[0096] S230, based on the artificial rabbit optimization algorithm, introduce the Lévy flight and Brownian motion strategies to construct an improved artificial rabbit optimization algorithm random foraging strategy;
[0097] Specifically, the Gray Wolf Optimization (GWO) algorithm is prone to failing to effectively explore the search space when dealing with complex problems such as multi-peak optimization, leading to premature convergence due to being stuck in a local optimum. The Artificial Rabbit Optimization (ARO) algorithm is a swarm intelligence optimization algorithm that simulates rabbits by controlling the transition between foraging and random hiding strategies through energy contraction. Different individuals are updated only along a subset of dimensions in each iteration, reducing the tendency for individuals to converge to the same local region and enabling more efficient exploration of the search space. In the basic GWO algorithm, wolf pack individuals, acting as followers, are guided only by the information of the three leading wolves, resulting in a lack of information sharing among individuals. In contrast, in the ARO algorithm's foraging behavior, each individual searches foraging, ignoring nearby food, and instead updates its position using other randomly selected individuals from the population. Individuals other than the leader have the opportunity to guide the pack, enhancing the pack's exploration capabilities. To address this issue, the present invention proposes a novel hybrid algorithm based on the Gray Wolf Optimization (GWO) algorithm and the Artificial Rabbit Optimization (ARO) algorithm. When a < 1, the Gray Wolf Optimization algorithm avoids stagnation in the exploration phase by incorporating the detour foraging behavior of the ARO. In the MSIAR-GWO algorithm, the gray wolves learn from the detour foraging behavior of the rabbits, enabling the pack to share information among individuals and improving their exploration of candidate solutions. Meanwhile, the wolf pack maintains its original hunting strategy to preserve its exploitation capabilities. To balance exploration and exploitation performance and improve optimization accuracy, the present invention incorporates Lévy flight and Brownian motion strategies into the detour foraging process of the ARO. Inspired by the energy contraction process of the ARO, the present invention simulates the transition from exploration to exploitation using an energy factor. The energy factor E = 1 is used as the cutoff point. When E > 1, the Lévy flight strategy is used for global search. Lévy flight is a random walk strategy with jumping characteristics. Its step length follows a heavy-tailed distribution and is unevenly distributed. Most steps are short, but large jumps occur occasionally. This facilitates leaping to other locations with a higher probability in the early stages of the iteration, allowing the gray wolf individuals to be widely distributed in the search space, thereby improving the global optimization capability. When E≤1, a Brownian motion strategy is used for local detection. Brownian motion is a method that describes the irregular random motion of particles in a fluid over a short distance. The direction and distance of each step are evenly distributed, without obvious long-distance jumps. This facilitates a more detailed and continuous search of the solution space when approaching the optimal solution in the later stages of the iteration, helping the algorithm find a better solution in the local area, thereby improving the accuracy of the algorithm. The improved formula for the random foraging strategy of the artificial rabbit optimization algorithm is as follows:
[0098] ;
[0099] In the above formula, Indicates the The position of the candidate solution at the iteration, Indicates the The current position at the iteration, 、 represents a random number between [0,1], represents the constant that controls the magnitude of the Brownian motion effect, 、 represents random numbers that follow a standard normal distribution, represents the energy factor, represents the Lévy flight distribution function, represents the dimension of the problem, Represents a random individual in the population with a better fitness value than the current position, represents the running operator, Indicates rounding to the nearest integer. represents the random walk coefficient of Brownian motion.
[0100] in:
[0101] ;
[0102] ;
[0103] ;
[0104] ;
[0105] ;
[0106] ;
[0107] ;
[0108] in It is The position of the candidate solution at the iteration; It is The current position at the iteration; is the maximum number of iterations; Indicates the movement speed when performing detour foraging; 、 、 、 and is a random number between [0,1]; Indicates from 1 to A random integer between ; In order to control the constant of the Brownian motion effect, the embodiment of the present invention is set to 10; and is a random number that follows a standard normal distribution; is the upper limit function; is the energy factor; is the Lévy flight distribution function, is the problem dimension, which is calculated as follows:
[0109] ;
[0110] in, is a fixed constant of 0.005; It is a related parameter. In the embodiment of the present invention, its value is set to 1.5; and is a random number in the interval [0,1]; The calculation formula is as follows:
[0111] ;
[0112] Represents the standard Gamma function, similarly The value of is 1.5. is the random walk coefficient of Brownian motion, which is essentially a random value that obeys the standard normal distribution. It can be obtained by The calculation formula is obtained.
[0113] The calculation formula for the random walk coefficient of Brownian motion is as follows:
[0114] ;
[0115] Although the improved detour foraging behavior based on the Lévy flight and Brownian motion strategies can obtain the location information of other gray wolves in the population except the leader wolf and guide them to update their positions, there is no guarantee that the fitness of the new solution obtained by the improved artificial rabbit optimization algorithm random foraging strategy formula is better than the adaptive position update formula based on dynamic weights. Therefore, a greedy mechanism is used to compare the fitness of the two solutions to retain the one with better fitness. The calculation formula is as follows:
[0116] ;
[0117] In the above formula, Represents a comparison function.
[0118] S240, screening and positioning the fitness values of the individual positions of the gray wolves in the gray wolf optimization algorithm through random center of gravity dynamic reverse learning, and constructing a random center of gravity dynamic reverse learning algorithm;
[0119] Specifically, in the traditional gray wolf algorithm, inferior individuals contribute less to the process of updating positions and searching for the best solution, but ineffective individuals still occupy computing resources during the evolution process, reducing the overall search efficiency. Due to the limited exploration ability of inferior individuals, it is impossible to effectively explore the entire search space. Their existence causes the search path to be too concentrated in certain inefficient areas, limiting the algorithm's ability to escape local optimality, making it difficult for the population to maintain a stable and efficient search path, affecting the search for the global optimal solution. Therefore, the present invention uses random center of gravity dynamic reverse learning to eliminate the inferior individuals ranked in the bottom 10% of the fitness value and re-update their positions, which can effectively alleviate these adverse effects, increase the search range of wolf pack individuals, prevent the algorithm from falling into local optimality, enhance population diversity, and accelerate the convergence process. Tizhoosh's research shows that the reverse solution of elite reverse learning has a 50% probability of being closer to the global optimal solution. Random center of gravity dynamic reverse learning takes into account the concepts of center of gravity and antagonistic learning. Considering the importance of elite individual guidance, it randomly selects elite individuals in the population according to the dynamic changes in the current iteration process to calculate a center of gravity point and generates reverse individuals based on this. By introducing random elements and dynamics, the robustness and flexibility of the algorithm are enhanced.
[0120] Generate a random integer , is the population size, and the best-fit wolf in the current pack is selected. Individual 、 , and calculate this The center of gravity of an individual, The formula for calculating the center of gravity of an individual is as follows:
[0121] ;
[0122] Based on the desired center of gravity, the reverse solution is calculated and the calculation formula for random center of gravity dynamic reverse learning is as follows:
[0123] ;
[0124] The greedy strategy is used to select the optimal solution individual from the current solution and its reverse solution as the new generation individual. Its expression is as follows:
[0125] ;
[0126] S250. According to the optimal adaptation parameters of the gray wolf optimization algorithm, combined with the adaptive position update algorithm based on dynamic weights, the improved artificial rabbit optimization algorithm random foraging strategy and the random center of gravity dynamic reverse learning algorithm, an improved gray wolf optimization algorithm is constructed.
[0127] S300: Perform optimal path planning processing on the map path feature information based on the improved grey wolf optimization algorithm to obtain the optimal planned path.
[0128] Specifically, the gray wolf population, the nonlinear convergence factor, the energy factor, the first coefficient vector and the second coefficient vector are initialized, and the gray wolf individuals in the gray wolf population represent discrete path nodes; the fitness values of the gray wolf individuals in the gray wolf population are calculated and sorted, and the corresponding gray wolf individuals with the top three fitness values are selected and stored; the first candidate solution of the gray wolf individual is obtained by an adaptive position update algorithm based on dynamic weights, and the nonlinear convergence factor is judged; if the nonlinear convergence factor is less than 1, the second candidate solution of the gray wolf individual is obtained by the random foraging strategy of the improved artificial rabbit optimization algorithm, and the fitness of the second candidate solution of the gray wolf individual is obtained value; compare the fitness value of the candidate solution of the gray wolf individual with the fitness value of the gray wolf individual. If the fitness value of the candidate solution is better, update the gray wolf individual with the candidate solution with the best fitness value, and screen it through the random center of gravity dynamic reverse learning algorithm to obtain the optimized gray wolf individual position, and the candidate solution includes the first candidate solution and the second candidate solution; update the nonlinear convergence factor and the energy factor, and iteratively optimize the position of the gray wolf individual based on the updated nonlinear convergence factor and the updated energy factor until the preset number of iterations is met, output the gray wolf population composed of the optimal gray wolf individual positions, and obtain the optimal planning path.
[0129] In the embodiment of the present invention, Figure 3 As shown:
[0130] 1) Initialize the population, one gray wolf is one path, the gray wolf population Depend on Only dimension The gray wolf individuals are composed of the upper limit of the search space , the lower limit is , obtained by the calculation formula of random position initialization, as shown below:
[0131] ;
[0132] Randomly generate a set of possible solutions. The gray wolf population can be represented by the position matrix of the following formula: , whose expression is:
[0133] ;
[0134] in Indicates the Gray Wolf The location of the solution, , is the population size, , Represents dimension; and Indicates the The lower and upper bounds of the dimensional variables; is a random number between [0,1];
[0135] 2) Discrete path nodes of dimension Connected by interpolation into a collision-free continuous path, and Respectively represent the coordinates of the starting point and the target point of the path P, and finally merge some redundant nodes on the path between the starting point and the target point to form an integrated planning path. The calculation formula of the fitness function for evaluating the path optimization ability of the algorithm is constructed as shown in the following formula. The fitness value of each individual in the group is calculated and ranked, and the fitness value of each individual in the group is used. 、 and They represent the positions of the top three individuals in terms of fitness value, and their expressions are:
[0136] ;
[0137] in, is the number of obstacles that the particle passes through, and are the horizontal and vertical coordinates of the end point, and are the horizontal and vertical coordinates of the continuous feasible path of each individual.
[0138] Determine whether the algorithm meets the termination condition. If so, the algorithm ends and outputs the optimal gray wolf individual. Otherwise, execute step 3).
[0139] 3) Calculate the convergence factor according to the calculation formula of nonlinear convergence factor Then calculate A and C according to the coefficient vector calculation formula of A and C. Finally, calculate the energy factor according to the calculation formula value.
[0140] 4) Update the position of each individual gray wolf in the group according to a new adaptive position update formula based on dynamic weights.
[0141] 5) By judgment Value, if , an improved artificial rabbit optimization algorithm is introduced through an adaptive position update formula based on dynamic weights to increase the randomness of the detour foraging behavior, and finally a greedy strategy is used to retain the solution with better fitness.
[0142] 6) According to The calculation formula of the center of gravity of each individual, the calculation formula of the random center of gravity dynamic reverse learning and the calculation formula of the greedy strategy are used to eliminate and reposition the inferior individuals in the population. , return to step 2).
[0143] In summary, the embodiment of the present invention first uses a grid method to model the environment based on known map information, and converts the actual physical environment in which the mobile robot is working into map feature information that can be processed and stored by the algorithm. Secondly, it establishes a fitness suitable for two-dimensional path planning scenarios. Finally, through the gray wolf optimization algorithm, a random optimization algorithm that imitates group social behavior, it generally follows the shortest path, shortest time, minimum energy consumption, and highest safety indicators to evaluate and select an optimal collision-free path from the starting point to the target point, while efficiently completing the task. Path planning algorithms are widely used in the fields of logistics, warehousing, and transportation. This improves operational efficiency and reduces labor costs.
[0144] In addition, it should be noted that the gray wolf optimization algorithm is a group intelligence optimization algorithm designed to simulate the hierarchy and hunting behavior of gray wolves in nature. When designing GWO, it is necessary to first establish a gray wolf social hierarchy model. After calculating the fitness of each individual in the population, the gray wolves with the top three fitness rankings are marked in order. 、 、 , other gray wolves are marked The optimization process of GWO is mainly composed of the best three solutions in each generation ( 、 、 ) is completed with guidance. Among them, The wolf is the leader of the wolf pack and is primarily responsible for making decisions about hunting, habitat, and rest. assist Make decisions and take over when necessary, second only to ; Submission and , responsible for tasks such as sentry duty and reconnaissance; At the bottom level are the indispensable common wolves, responsible for maintaining the balance within the population. The optimization process divides the predation task into three phases: encirclement, pursuit, and attack, which are completed by gray wolf packs at all levels, thus achieving a globally optimized search process.
[0145] Surrounding the prey:
[0146] When searching for prey, gray wolves gradually approach and surround their prey. This behavior can be described by the following mathematical model:
[0147] The distance formula between the individual and the prey is:
[0148] ;
[0149] Gray wolf position update formula:
[0150] ;
[0151] In the above formula, and is the coefficient vector, which is calculated as follows:
[0152] ;
[0153] ;
[0154] 、 The modulus of gets a random number between [0,1], is the convergence factor, which decreases linearly from 2 to 0 as the number of iterations decreases. The expression of the convergence factor is as follows:
[0155] ;
[0156] hunting:
[0157] Gray wolves cannot determine the exact location of their prey, but they have the ability to identify the location of potential prey. In order to simulate the search behavior of gray wolves, other gray wolves are based on 、 、 During the hunting process, the gray wolf's basic position update method is defined as:
[0158] ;
[0159] in , , Candidate individuals and 、 and The distance between them. , and Respectively represent 、 and Current location, , , is a random vector, This is the current location of the Gray Wolf.
[0160] ;
[0161] ;
[0162] Attacking prey:
[0163] In the calculation formula of the convergence factor, as the iteration When the value of decreases from 0, the corresponding The value is in the interval Change, when If >1, it will cause the gray wolves to stay away from prey, prompting the wolf pack to disperse to various areas and search for prey for a global search. When the gray wolf approaches the prey, it will concentrate on searching for prey in one or certain areas, prompting the wolf pack to conduct a local search.
[0164] Furthermore, to test the optimization performance of the Multi-Strategy Improved Gray Wolf Optimization Algorithm (MSIAR-GWO) algorithm proposed in this embodiment of the present invention, it was compared with other swarm intelligence optimization algorithms, such as the Artificial Rabbit Optimization (ARO), the Dung Beetle Optimization (DBO), and the Whale Optimization (WOA). Simulation experiments were also conducted on six standard test functions, the specific characteristics of which are shown in Table 2. F1-F2 are single-peak benchmark functions, F3-F4 are benchmark functions, and F5-F6 are fixed-dimensional multi-peak benchmark functions. To verify the performance of the proposed IGWO algorithm and the fairness of the simulation experiments, the maximum number of iterations for all algorithms was set to 500, and the number of populations was set to 30. Each algorithm was executed 20 times, and the optimal value, average value, worst value, and standard deviation of each algorithm were calculated. The average value was used to evaluate the optimization accuracy, and the standard deviation was used to evaluate the stability performance.
[0165] Table 2 Classic standard benchmark functions
[0166]
[0167] In the single-peak test function F1~F2, the MSIAR-GWO algorithm and the RSMGWO algorithm proposed in this invention show similar performance in F1, and both can obtain the theoretical optimal solution. Figure 4 It can be seen that the convergence speed of MSIAR-GWO in the optimization process is faster than that of RSMGWO. Compared with other comparison algorithms, F2 always maintains the first place in optimization performance in terms of optimal value, average value, worst value and standard deviation, showing better ability and excellent stability. Since a unimodal function has only one global optimal solution, it can be used to evaluate the performance of algorithm development. Figure 4 and Figure 5As shown in the figure, the simulation results of F1~F2 prove that the nonlinear convergence factor proposed in the present invention and the intelligent selection of adjustable parameters through reinforcement learning can well measure the exploration and development process of the group, and the adaptive position update formula based on dynamic weights can well highlight the importance of different advantage individuals according to the hierarchical relationship, effectively demonstrating the convergence performance of the improved algorithm and the accuracy of the solution.
[0168] like Figure 6 and Figure 7 As shown in the data, our MSIAR-GWO method demonstrates a clear advantage over other classic optimization algorithms and other improved gray wolf optimization algorithms in all statistical data for multimodal test functions F3 and F4. For multimodal test functions, many local optima exist in the search space, effectively testing an algorithm's ability to solve complex problems. Simulation experiments demonstrate that the MSIAR-GWO algorithm, by improving the artificial rabbit optimization algorithm's foraging strategy and incorporating a random center of gravity dynamic reverse learning elimination and relocation strategy, increases population diversity, effectively overcomes the problem of being trapped in local optima, and improves the algorithm's search capabilities.
[0169] like Figure 8 and Figure 9 As shown in the figure, on the fixed-dimensional multimodal benchmark functions F5 and F6, many algorithms can find the optimal solution, with no significant differences. In F6, both the MSIAR-GWO and IGWO algorithms achieve higher convergence accuracy than the other comparison algorithms. However, a comparison of the standard deviation evaluation metric reveals that the MSIAR-GWO algorithm has an absolute advantage and is more robust than the other algorithms.
[0170] In summary, for most test functions, MSIAR-GWO is significantly superior to other algorithms in terms of search accuracy, convergence speed, and stability. The specific experimental data are shown in Table 3.
[0171] Table 3 Comparative experimental results of MSIAR-GWO algorithm effectiveness verification
[0172]
[0173] Further simulation for mobile robot path planning:
[0174] In order to verify the performance of the MSIAR-GWO algorithm, the improved GWO algorithm was applied to the path planning problem of a mobile robot in environments of different complexity. 30×30 and 40×40 grids were used, and the starting point and end point of the robot path were set at the lower left corner and upper right corner respectively. Eight different algorithms were run independently 20 times in different grid environments. Finally, the effectiveness of the algorithms was evaluated by calculating the optimal value, average value and standard deviation of each algorithm. Figure 10 and Figure 11 As shown. From the path planning results under the 30×30 grid map in Table 4, although the MSIAR-GWO and ARO algorithms converge to the same optimal value, the MSIAR-GWO algorithm is better than the ARO algorithm in terms of both average and standard deviation. The improved algorithm shows strong stability and robustness, which is due to the fact that the nonlinear convergence factor after enhanced parameter optimization can well control the process of global exploration and local development of the wolf pack. In the 30×30 grid path, the values were reduced by 17.77%, 4.27%, 11.99%, 14.52%, 6.09%, 15.02% and 14.59% respectively compared with other algorithms; as shown in Table 5, in the 40×40 high-dimensional complex map, whether it is the optimal value, average value or standard deviation, it has an absolute advantage over other algorithms, reducing them by 22.88%, 4.47%, 7.10%, 8.32%, 19.40%, 22.65% and 21.97% respectively compared with other algorithms. Figure 12 and Figure 13 As the grid map dimension increases and the environment becomes more complex, other algorithms lose their search capabilities in the later stages of iteration, causing them to become trapped in local optima. This invention, by combining and improving the rabbits' detour foraging behavior and eliminating and relocating the wolf pack's inferior individuals, enriches the population diversity, improves the algorithm's global search and ability to escape local optima, and enhances the mobile robot's ability to fully search for paths.
[0175] Table 4 Comparative experimental analysis results of path planning under 30×30 grid map
[0176]
[0177] Table 5 Comparative experimental analysis results of path planning under 40×40 grid map
[0178]
[0179] Reference Figure 2 , a robot path planning system based on an improved gray wolf optimization algorithm, comprising:
[0180] Acquisition module 201, used to obtain map path feature information;
[0181] Construction module 202 is used to obtain the optimal adaptation parameters of the gray wolf optimization algorithm, introduce the adaptive position update algorithm based on dynamic weights, the artificial rabbit optimization algorithm and the random center of gravity dynamic reverse learning algorithm to construct an improved gray wolf optimization algorithm;
[0182] The planning module 203 is used to perform optimal path planning processing on the map path feature information based on the improved gray wolf optimization algorithm to obtain the optimal planned path.
[0183] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0184] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. A robot path planning method based on an improved grey wolf optimization algorithm, characterized in that: The following steps are involved: Get map path feature information; The nonlinear adjustment parameters in the gray wolf optimization algorithm are selected and processed by a reinforcement learning method to obtain the optimal adaptation parameters of the gray wolf optimization algorithm, wherein the nonlinear adjustment parameters include a nonlinear convergence factor, an energy factor, a first coefficient vector and a second coefficient vector; According to the fitness value of the individual position of the gray wolf in the gray wolf optimization algorithm, the individual position of the gray wolf is adaptively updated, and an adaptive position update algorithm based on dynamic weight is constructed; Based on the artificial rabbit optimization algorithm, the Levy flight and Brownian motion strategies are introduced to construct an improved random foraging strategy for the artificial rabbit optimization algorithm; Through random center of gravity dynamic reverse learning, the fitness value of the individual position of the gray wolf in the gray wolf optimization algorithm is screened and positioned, and a random center of gravity dynamic reverse learning algorithm is constructed; According to the optimal adaptation parameters of the gray wolf optimization algorithm, the improved gray wolf optimization algorithm is constructed by combining the adaptive position update algorithm based on dynamic weights, the random foraging strategy of the improved artificial rabbit optimization algorithm, and the dynamic reverse learning algorithm based on random center of gravity. Based on the improved grey wolf optimization algorithm, the map path feature information is processed for optimal path planning to obtain the optimal planned path.
2. According to claim 1, a robot path planning method based on an improved gray wolf optimization algorithm is characterized in that: The step of obtaining map path feature information specifically includes: The robot laser radar sensor obtains the environmental information of the target area and performs discretization processing to obtain a grid map of the target area; Obtain irregular obstacles in the grid map of the target area and perform expansion processing to obtain a grid map with rectangular regular obstacles; The grid map with rectangular regular obstacles is processed by grid matrix unit division and coding marking to obtain map path feature information.
3. According to claim 2, a robot path planning method based on an improved gray wolf optimization algorithm is characterized in that: The step of performing matching processing on the nonlinear adjustment parameters in the gray wolf optimization algorithm by the reinforcement learning method to obtain the optimal adaptation parameters of the gray wolf optimization algorithm specifically includes: Determining the search space dimension, the maximum number of iterations, the first nonlinear adjustment parameter, the second nonlinear adjustment parameter, and the benchmark test function; Determining a parameter vector and an optimization space of the parameter vector according to the first nonlinear adjustment parameter and the second nonlinear adjustment parameter; Randomly assign values to the parameter vector within the optimization space and search space dimension range of the parameter vector to obtain the initial action reward; According to the initial action reward, select the action corresponding to the first nonlinear adjustment parameter and the second nonlinear adjustment parameter, substitute them into the calculation formula of the nonlinear convergence factor, and optimize the benchmark test function to obtain the reward value of the benchmark test function; Based on the reward value of the benchmark function, it is iteratively updated through the Q-learning update formula until the maximum number of iterations is met, and the optimal adaptation parameters of the gray wolf optimization algorithm are output.
4. According to claim 3, a robot path planning method based on an improved gray wolf optimization algorithm is characterized in that: The calculation formula of the nonlinear convergence factor is specifically expressed as follows: ; In the above formula, represents the convergence factor, represents the initial value of the convergence factor, represents the final value of the convergence factor, represents the first nonlinear adjustment parameter, represents the second nonlinear adjustment parameter, Indicates the current iteration number, Indicates the maximum number of iterations.
5. A robot path planning method based on an improved grey wolf optimization algorithm according to claim 4, characterized in that: The expression of the adaptive position update algorithm based on dynamic weight is specifically as follows: ; In the above formula, represents the updated position of the individual gray wolf, represents the position update weight of the first gray wolf individual, represents the current position of the first gray wolf individual, represents the position update weight of the second gray wolf individual, represents the current position of the second gray wolf individual, represents the position update weight of the third gray wolf individual, represents the current position of the third gray wolf individual, Represents a random number, Represents the weight of the current position at different iteration times, Indicates the current location of the gray wolf individual, Represents a random individual in the population with a better fitness value than the current position, represents a constant that controls the size of randomness, Represents the individual with the best fitness value in the population Wolf's location.
6. A robot path planning method based on an improved grey wolf optimization algorithm according to claim 5, characterized in that: The expression of the random foraging strategy of the improved artificial rabbit optimization algorithm is specifically as follows: ; In the above formula, Indicates The position of the candidate solution at the iteration, Indicates The current position at the iteration, represents a random number between [0,1], represents the constant that controls the magnitude of the Brownian motion effect, represents random numbers that follow a standard normal distribution, represents the energy factor, represents the Levy flight distribution function, represents the dimension of the problem, Represents a random individual in the population with a better fitness value than the current position, represents the running operator, Indicates rounding to the nearest integer. represents the random walk coefficient of Brownian motion, Represents the individual with the best fitness value in the population Wolf's location.
7. A robot path planning method based on an improved grey wolf optimization algorithm according to claim 6, characterized in that: The step of performing optimal path planning processing on the map path feature information based on the improved grey wolf optimization algorithm to obtain the optimal planned path specifically includes: Initialize the gray wolf population, nonlinear convergence factor, energy factor, first coefficient vector and second coefficient vector, and the gray wolf individuals in the gray wolf population represent discrete path nodes; Calculate the fitness values of the gray wolf individuals in the gray wolf population and sort them, and select and store the corresponding gray wolf individuals with the top three fitness values; The first candidate solution of the individual gray wolf is obtained through an adaptive position update algorithm based on dynamic weights, and the nonlinear convergence factor is judged; If the nonlinear convergence factor is less than 1, the second candidate solution of the gray wolf individual is obtained by improving the random foraging strategy of the artificial rabbit optimization algorithm, and the fitness value of the second candidate solution of the gray wolf individual is obtained; Compare the fitness value of the candidate solution of the gray wolf individual with the fitness value of the gray wolf individual. If the fitness value of the candidate solution is better, update the gray wolf individual with the candidate solution with the best fitness value, and perform screening processing through the random center of gravity dynamic reverse learning algorithm to obtain the optimized gray wolf individual position, wherein the candidate solution includes a first candidate solution and a second candidate solution; The nonlinear convergence factor and energy factor are updated, and the positions of the gray wolf individuals are iteratively optimized based on the updated nonlinear convergence factor and the updated energy factor until the preset number of iterations is met, and the gray wolf population composed of the optimal gray wolf individual positions is output to obtain the optimal planning path.
8. A robot path planning system based on an improved grey wolf optimization algorithm, characterized in that: Includes the following modules: An acquisition module is used to obtain map path feature information; A construction module is used to select and process the nonlinear adjustment parameters in the gray wolf optimization algorithm through a reinforcement learning method to obtain the optimal adaptation parameters of the gray wolf optimization algorithm, wherein the nonlinear adjustment parameters include a nonlinear convergence factor, an energy factor, a first coefficient vector and a second coefficient vector; According to the fitness value of the individual position of the gray wolf in the gray wolf optimization algorithm, the individual position of the gray wolf is adaptively updated, and an adaptive position update algorithm based on dynamic weight is constructed; Based on the artificial rabbit optimization algorithm, the Levy flight and Brownian motion strategies are introduced to construct an improved random foraging strategy for the artificial rabbit optimization algorithm; Through random center of gravity dynamic reverse learning, the fitness value of the individual position of the gray wolf in the gray wolf optimization algorithm is screened and positioned, and a random center of gravity dynamic reverse learning algorithm is constructed; According to the optimal adaptation parameters of the gray wolf optimization algorithm, the improved gray wolf optimization algorithm is constructed by combining the adaptive position update algorithm based on dynamic weights, the random foraging strategy of the improved artificial rabbit optimization algorithm, and the dynamic reverse learning algorithm based on random center of gravity. The planning module is used to perform optimal path planning processing on the map path feature information based on the improved gray wolf optimization algorithm to obtain the optimal planned path.
