Multi-robot path planning method based on multi-heterogeneous population coevolution and parallel search

Abstract

The invention discloses a multi-robot path planning method based on multi-heterogeneous population coevolution and parallel search. The method comprises the following steps: firstly, modeling an environment by adopting a two-dimensional grid method, and generating a multi-robot initial path of a total population by utilizing a risk perception A * algorithm; then, dividing the total population into three sub-populations and corresponding external cooperative populations through a hierarchical roulette selection strategy; the three types of heterogeneous algorithms are executed in parallel, namely the collaborative genetic algorithm, the collaborative particle swarm algorithm and the collaborative ant colony algorithm are executed in parallel and synchronously, and a global solution set is generated in combination with a double-solution pool fusion mechanism. And then, a priority-based conflict resolution strategy is adopted, and time-space conflicts are solved by adopting time migration, path detour and local path planning strategies. And finally, path smoothing is realized through a cubic B-spline curve based on path reconstruction. According to the method, heterogeneous population coevolution is combined with parallel calculation, so that the efficiency, safety and cooperation capability of multi-robot path planning are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to a multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search. Background Art

[0002] Multi-robot path planning, as a core technology in multi-robot systems, has important applications in warehousing and logistics, intelligent inspection, disaster relief, and other fields. With the increasing complexity of tasks, task scale, and environmental complexity, planning a set of mutually non-conflicting, optimal, and feasible paths for multiple robots in environments with dense obstacles, while also ensuring both path quality and safety, has become a critical issue that needs to be addressed. Existing multi-robot path planning methods suffer from the following shortcomings: First, they are computationally expensive and difficult to adapt to dynamic environments and large-scale scenarios; second, most methods rely on a single optimization algorithm with limited search capabilities, prone to falling into local optima, and struggling to balance global and local search efficiency; third, they are prone to spatiotemporal conflicts and poor path smoothness, making it difficult to meet the requirements for path feasibility and safety in real-world scenarios. Heterogeneous population coevolution and parallel computing technologies can address these issues. By leveraging a population grouping strategy and introducing a hierarchical roulette wheel selection method, three algorithm subpopulations and an external collaborative population are constructed. The three heterogeneous algorithms are executed in parallel, and the collaborative interaction mechanism between the subpopulations and the external collaborative population effectively improves path search efficiency and diversity. Furthermore, the risk-aware A* algorithm helps improve the initial quality of the population and reduce potential risks along the path. The dual-solution pools, combined with the global solution set to construct a dual elite pool, preserve the evolutionary advantages of elite individuals while maintaining population diversity, while also providing an efficient individual allocation basis for heterogeneous group collaboration. Priority-based conflict resolution, through the combined use of multiple resolution strategies, enables flexible response to different conflict types. Finally, the cubic B-spline curve based path reconstruction ensures path smoothness and security. Summary of the Invention

[0003] The purpose of the present invention is to provide a multi-robot path planning method based on the co-evolution and parallel search of multiple heterogeneous populations, so as to solve the problems of frequent path conflicts, insufficient optimization capability of a single algorithm and poor path executability existing in the multi-robot path planning method in complex environments, so as to improve the path planning efficiency, path quality and system collaboration capability.

[0004] To achieve the above object, the specific solutions of the present invention are as follows:

[0005] Step S101: Use a two-dimensional grid method to establish a multi-robot working environment model:

[0006] The robot working environment is discretized into rectangular grid units to construct a Cartesian coordinate system in which the lower left corner of the grid is set as the coordinate origin (0,0), the horizontal right direction is the positive direction of the x-axis, and the vertical upward direction is the positive direction of the y-axis;

[0007] Assign a value to each grid, mark the grid occupied by the obstacle outline as impassable, recorded as -1, and mark the remaining grids as free passage areas. Calculate the Manhattan distance d from each free passage grid to the nearest obstacle, recorded as d;

[0008] The rasterized map is converted into a matrix for storage, and the robot's starting point and target point are converted into corresponding grids through coordinate mapping, satisfying the following constraints: the eight neighborhoods of the grid where the starting point or target point is located must not all be obstacle grids.

[0009] Step S102: Initialize the paths of the multi-robot system's population using the risk-aware A* algorithm to improve the feasibility and path safety of the initial solution. This initialization process uses the multi-robot joint path as the individual encoding unit, ensuring that each individual represents a complete multi-robot path combination, thus providing a good search starting point for subsequent evolutionary optimization.

[0010] Specifically, suppose the multi-robot system consists of m robots. Each individual in the population is encoded as a joint structure of m independent paths, each of which consists of a series of spatiotemporal path points. The path of each robot i is represented as:

[0011] P={P1,P2,…,P i ,…,P m} (17)

[0012] in, represents the path sequence of the i-th robot, are grid coordinates, is the time when the robot reaches the grid;

[0013] In the path generation process, the risk-aware A* algorithm is used for heuristic path search. The key steps are as follows:

[0014] Step (3.1): For each robot, set its starting point as the initial parent node and its target point as the final parent node, and construct the initial state of the path;

[0015] Step (3.2): Perform feasibility check on the adjacent grid nodes of the current node, eliminate obstacles and illegal nodes, calculate the cost value f(n) for each feasible neighboring node n, and select the domain grid node with the smallest cost value as the next parent node. Use the following heuristic function including risk factors:

[0016] f(n)=g(n)+h(n)+R(n) (18)

[0017] R(n)=exp(-λ·d(n)) (19)

[0018] Where g(n) is the actual distance traveled from the starting point to node n, h(n) is the Manhattan distance from node n to the target point, R(n) represents the risk value of node n, d(n) represents the Euclidean distance from node n to the nearest obstacle, and λ is the attenuation coefficient, which controls the rate at which risk decays with distance.

[0019] Step (3.3): Repeat step (3.2) and search until the target point falls into the neighborhood grid of the target point, indicating that the search is completed, and then generate a complete path by backtracking the parent node.

[0020] Step S103 , dividing the total path population of the multi-robot system into three sub-populations and their corresponding external collaboration populations by evaluating the fitness of path length, smoothness, and path safety and using a hierarchical roulette wheel selection mechanism to divide the population;

[0021] Step (4.1): Arrange the robot individuals in descending order according to their comprehensive fitness F and divide them equally into three levels. The fitness F calculation formula is:

[0022] F=β1·f l +β2·f s +β3·f a (20)

[0023] Among them, f l is the total path length, f s is the path smoothness, f a is the path safety, β1, β2, β3 are weight coefficients, and β1+β2+β3=1;

[0024] Step (4.2): First, the total population is divided into three subpopulations based on the comprehensive fitness F. Then, three corresponding external collaborative populations are constructed using path length, smoothness, and safety as evaluation indicators. All populations are constructed using the stratified roulette wheel selection method. The subpopulations select individuals from the total population, and the external collaborative populations select individuals from the two subpopulations other than their corresponding subpopulations. The stratified roulette wheel selection method combines the hierarchical weight coefficient and the size of the evaluation indicator. The selection probability calculation formula is as follows:

[0025]

[0026] Among them, P i is the probability of individual i being selected, g i is the evaluation index, α i is the hierarchical weight coefficient of individual i.

[0027] Step S104, heterogeneous multi-population collaborative parallel optimization, the collaborative genetic algorithm in parallel branch 1, specifically: first, a roulette wheel selection method is used to combine subpopulation 1 with external collaborative population 1 to construct a parent population; then, individuals are randomly selected from the parent population to perform collaborative crossover and mutation operations; for homologous individuals from the same population, a single-point crossover strategy is used to complete crossover by setting breakpoints in the path and exchanging the path segments following the breakpoint; for heterologous individuals from different populations, a two-point elitist crossover strategy is used to exchange the path segments between the two endpoints. After crossover, the mutation phase begins: homologous individuals use global random mutation to randomly select a feasible grid from the map to replace the current grid; heterologous individuals use local random mutation to randomly select a feasible grid within a radius R centered on the current grid to replace the current grid; finally, the generated path is repaired by obstacle avoidance, backtracking to correct conflicting nodes, and inserting detour nodes for adjustment.

[0028] Furthermore, the collaborative particle swarm algorithm in the parallel branch 2 is specifically as follows:

[0029] Step (6.1): Execute the speed update operation. The collaborative particle swarm algorithm adopts a hybrid speed update operator to integrate the best individual from the individual history, the sub-population, and the external collaborative population to improve the search capability of the particle swarm algorithm. The speed update formula is:

[0030]

[0031] ω=ω0·exp(-η·t / T max ) (twenty three)

[0032] Among them, v i represents the current speed of individual i, x i , pbest i , gbest p , gbest e are the current position of individual i, the historical optimal position, the global optimal position of subpopulation 1, and the group optimal position of the corresponding external collaborative population; represents the direction difference operator, which calculates the difference between the moving direction vector of the optimal path and the current path at the same time step, and rounds the result; c1, c2 and c3 are learning factors, r1, r2 and r3 are independent random numbers, ω is the inertia weight, and the attenuation strategy is adopted, where ω0 is the initial inertia weight, η is the attenuation rate factor, t is the current iteration number, and T max is the total number of iterations;

[0033] Step (6.2): ​​Perform the position update operation. The particle moves according to the updated speed. The position update formula is as follows;

[0034]

[0035] Among them, x i and v i are the current position and velocity of individual i, respectively, and t is the current iteration number;

[0036] Step (6.3): Execute the path obstacle avoidance repair operation. The updated particles must meet the obstacle avoidance requirements. If the path node falls into the obstacle grid, move the infeasible node to bypass the obstacle.

[0037] Furthermore, the collaborative ant colony algorithm in the parallel branch three is executed, specifically:

[0038] Step (7.1): Collaborative pheromone injection operation, extracting path nodes in subpopulation 3 and external collaborative population 3, updating pheromone benchmark distribution, and providing a high-quality search starting point for the ant colony algorithm. The calculation formula for the collaborative pheromone injection is as follows:

[0039]

[0040] Among them, τ ij (t) is the pheromone concentration moving from grid i to adjacent grid j at the tth iteration, t = 0 is the initial pheromone concentration, ρ is the volatility coefficient, γ is the cross-population pheromone fusion coefficient, Q is the pheromone intensity coefficient, L k is the elite path length, Indicates whether ant k passes through the grid edge (i, j), K represents all paths in subpopulation 3 and external collaborative population 3;

[0041] Step (7.2): Solution path construction operation, the ant selects the path node according to the transition probability and gradually generates the solution path; if the node falls into the obstacle grid, it backtracks to the nearest free grid and increases the invalid path penalty weight during replanning to guide obstacle avoidance; the transition probability The formula is the probability that an ant moves from the current grid i to the adjacent grid j, and its calculation formula is as follows:

[0042]

[0043] Among them, η ij is the heuristic factor, d ij is the Manhattan distance from grid i to the target point, ξ is the anti-zero constant, τ ij is the pheromone concentration moving from the current grid i to the adjacent grid j, α is the pheromone weight factor; β is the heuristic information weight factor, s represents a feasible neighborhood of the current position, and A represents the feasible neighborhood set of ant k at the current node;

[0044] Step (7.3): Pheromone update operation. After the solution path is constructed, the elite retention and global collaboration hybrid model is used to update the pheromone, strengthen the high-quality path, and improve the global search ability and convergence speed. The calculation formula is:

[0045]

[0046] in, It represents the local pheromone increment after the ants release it. Represents the pheromone increment of the global elite path, N a is the number of ants in the current ant colony, λ g is the global update coefficient, M represents the elite path with high fitness in the current ant colony, I is the set of invalid paths, σ is the penalty weight of invalid paths, τ ij is the pheromone concentration moving from the current grid i to the adjacent grid j, t is the current iteration number, k represents the ant individual, Indicates whether ant k passes through the grid edge (i, j), ρ is the volatility coefficient, Q is the pheromone intensity coefficient, L k is the elite path length.

[0047] The dual solution pool fusion mechanism in step S105 generates a global solution set, integrating the elite solution pool and the diversity solution pool generated by the three algorithm branches. The mechanism includes the following steps:

[0048] Step (8.1): Sort the population individuals output from the three parallel branches in descending order according to the comprehensive fitness function F. The individuals in each subpopulation are divided into an elite group and a diversity group based on the sorting results;

[0049] Step (8.2): Initialize two empty sets as the fusion target solution sets: the elite pool is used to store high-fitness individuals from the elite groups of each branch; the diversity pool is used to store representative individuals from the diversity group;

[0050] Step (8.3): Elite pool construction: From the elite group of each branch, use the tournament selection method with a selection pressure of k to randomly select k individuals from the current elite group in each round; compare their fitness and select the best ones to enter the elite pool; repeat the operation until the elite pool capacity is reached;

[0051] Step (8.4): Diversity pool construction: select individuals from the diversity group of each branch by uniform random sampling to fill the diversity pool until the diversity pool capacity is reached;

[0052] Step (8.5): Generate a global solution set. Merge the elite pool and the diversity pool constructed above to generate a new global solution set. This solution set integrates the excellence and diversity of the solution and has high-quality path solutions.

[0053] In step S106, a priority-based conflict resolution strategy is used to resolve multi-robot conflicts. This strategy performs conflict detection, priority division, and conflict path adjustment on the global path solution set to gradually eliminate path conflicts.

[0054] Step (9.1): Perform spatiotemporal conflict detection on all path individuals in the global solution set and identify the following three types of conflicts: (1) occupancy conflict, i.e., two robots occupy the same grid at the same time step; (2) crossing conflict, i.e., two robots cross each other along adjacent paths at the same time step; (3) exchange conflict, i.e., two robots exchange positions in adjacent time steps;

[0055] Step (9.2): Assign priorities according to the urgency of the robot's tasks. The robot only needs to consider conflicts with higher-priority tasks, and resolve conflicts in descending order of priority. Execute the following three resolution strategies in sequence. If the conflict is resolved, a conflict-free path is output: (1) Time offset strategy: When a single-point conflict is detected, insert a waiting node for the low-priority robot to delay its path time step; (2) Path detour strategy: Insert detour path points before and after the conflict area for the low-priority robot; (3) Local replanning strategy: Fix the path of the high-priority robot and replan the path of the low-priority robot in the conflict area;

[0056] After the above strategy is executed, the conflict status is rechecked. If a conflict still exists, the low-priority robot path is regenerated until there is no conflict with other robot paths.

[0057] In step S107, the optimized robot path is smoothed based on the cubic B-spline curve of path reconstruction. This process specifically includes the following two sub-steps:

[0058] Step (10.1): First, directly connect the path points of the optimal individual path to the subsequent non-adjacent path points in sequence, skipping the intermediate points if feasible, and then randomly move each path point towards the midpoint between the two adjacent path points, replacing the current path point if feasible, until all path nodes are processed;

[0059] Step (10.2): Use a cubic B-spline curve to fit the path points to generate a continuous smooth path. The calculation expression of the cubic B-spline method is as follows:

[0060]

[0061] Among them, P is the coordinate value of a point on the cubic B-spline path curve, n is the maximum subscript of the path point, Q i =(x i ,y i ) represents the coordinates of the i-th path point, B i,3(u) is the cubic B-spline basis function, which is represented by a piecewise polynomial. u is the normalized curve parameter, and the path point set is generated by fitting the original path node coordinates using the least squares method.

[0062] Compared with existing robot path planning methods, the present invention has the following advantages:

[0063] 1. The present invention achieves collaborative evolution through subpopulations of collaborative genetic algorithm, collaborative particle swarm algorithm and collaborative ant colony algorithm and external collaborative population, breaking through the limitations of a single algorithm and taking into account both global exploration and local development capabilities.

[0064] 2. The present invention adopts the risk-aware A* algorithm to initialize the multi-robot path and introduces a risk value assessment mechanism to effectively avoid the problem of poor path quality generated by traditional initialization strategies and generate an initial path set that is both safe and reliable.

[0065] 3. This invention adopts a dual solution pool fusion mechanism. The elite pool retains the elite solutions, and the diversity pool retains the diversity solutions. The two pools are fused to form a global solution set, which effectively avoids premature convergence of the algorithm and enhances the algorithm's local exploration and global search capabilities.

[0066] 4. The present invention adopts robot priority allocation, combined with conflict identification and time offset strategy, path detour strategy and local replanning strategy, to achieve efficient spatiotemporal conflict handling among multiple robots.

[0067] 5. The present invention uses a cubic B-spline curve post-processing method based on path reconstruction, which reduces the robot path distance and improves the robot path smoothness. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 Flowchart of a multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search provided by an embodiment of the present invention.

[0069] Figure 2 A partial schematic diagram of population grouping provided by an embodiment of the present invention.

[0070] Figure 3 Schematic diagram of heterogeneous multi-population parallel co-evolution provided by an embodiment of the present invention.

[0071] Figure 4 A schematic diagram of the genetic algorithm branch crossover mutation provided by an embodiment of the present invention.

[0072] Figure 5 A fusion diagram of the global solution set provided by an embodiment of the present invention.

[0073] Figure 6 This is a flow chart of a priority-based conflict resolution strategy provided by an embodiment of the present invention.

[0074] Figure 7 This is a smoothed graph of a path curve provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0075] In order to make the features and advantages of the present invention more obvious and easy to understand, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0076] Figure 1 A flow chart of a multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search provided in an embodiment of the present invention, the method comprising:

[0077] Step S101: Establishing an environment model for multi-robot path planning;

[0078] Step S102: Initialize the multi-robot swarm path using the risk-aware A* algorithm;

[0079] Step S103: using a hierarchical roulette wheel selection method to divide the multi-robot population into three sub-populations and corresponding external collaboration populations;

[0080] Step S104: Heterogeneous multi-population collaborative parallel optimization, executing three algorithm branches in parallel, and achieving collaborative evolution through mutual communication between sub-populations and corresponding external collaborative populations;

[0081] Step S105: using the dual solution pool population update mechanism to perform global solution set fusion;

[0082] Step S106: using a priority-based conflict resolution strategy to resolve multi-robot conflicts;

[0083] Step S107: performing path smoothing on the optimal path of the multi-robots using a cubic B-spline curve based on path reconstruction, and finally outputting the smoothed path;

[0084] Step S101 specifically includes:

[0085] The robot working environment is discretized into rectangular grid units to construct a Cartesian coordinate system in which the lower left corner of the grid is set as the coordinate origin (0,0), the horizontal right direction is the positive direction of the x-axis, and the vertical upward direction is the positive direction of the y-axis;

[0086] Assign a value to each grid, mark the grid occupied by the obstacle outline as impassable, recorded as -1, and mark the remaining grids as free passage areas. Calculate the Manhattan distance d from each free passage grid to the nearest obstacle, recorded as d;

[0087] The rasterized map is converted into a matrix for storage, and the robot's starting point and target point are converted into corresponding grids through coordinate mapping, satisfying the following constraints: the eight neighborhoods of the grid where the starting point or target point is located must not all be obstacle grids.

[0088] Step S102: Initialize the paths of the multi-robot system's population using the risk-aware A* algorithm to improve the feasibility and path safety of the initial solution. This initialization process uses the multi-robot joint path as the individual encoding unit, ensuring that each individual represents a complete multi-robot path combination, thus providing a good search starting point for subsequent evolutionary optimization.

[0089] Specifically, suppose the multi-robot system consists of m robots. Each individual in the population is encoded as a joint structure of m independent paths, each of which consists of a series of spatiotemporal path points. The path of each robot i is represented as:

[0090] P={P1,P2,…,P i ,…,P m} (33)

[0091] in, represents the path sequence of the i-th robot, are grid coordinates, is the time when the robot reaches the grid;

[0092] In the path generation process, the risk-aware A* algorithm is used for heuristic path search. The key steps are as follows:

[0093] Step (3.1): For each robot, set its starting point as the initial parent node and its target point as the final parent node, and construct the initial state of the path;

[0094] Step (3.2): Perform feasibility check on the adjacent grid nodes of the current node, eliminate obstacles and illegal nodes, calculate the cost value f(n) for each feasible neighboring node n, and select the domain grid node with the smallest cost value as the next parent node. Use the following heuristic function including risk factors:

[0095] f(n)=g(n)+h(n)+R(n) (34)

[0096] R(n)=exp(-λ·d(n)) (35)

[0097] Where g(n) is the actual distance traveled from the starting point to node n, h(n) is the Manhattan distance from node n to the target point, R(n) represents the risk value of node n, d(n) represents the Euclidean distance from node n to the nearest obstacle, and λ is the attenuation coefficient, which controls the rate at which risk decays with distance.

[0098] Step (3.3): Repeat step (3.2) and search until the target point falls into the neighborhood grid of the target point, indicating that the search is completed, and then generate a complete path by backtracking the parent node.

[0099] Step S103 specifically includes:

[0100] Step (4.1): Arrange the robot individuals in descending order according to their comprehensive fitness F and divide them equally into three levels. The fitness F calculation formula is:

[0101] F=β1·f l +β2·f s +β3·f a (36)

[0102] Among them, f l is the total path length, f s is the path smoothness, f a is the path safety, β1, β2, β3 are weight coefficients, and β1+β2+β3=1;

[0103] Step (4.2): First, the total population is divided into three subpopulations based on the comprehensive fitness F. Then, three corresponding external collaborative populations are constructed using path length, smoothness, and safety as evaluation indicators. All populations are constructed using the stratified roulette wheel selection method. The subpopulations select individuals from the total population, and the external collaborative populations select individuals from the two subpopulations other than their corresponding subpopulations. The stratified roulette wheel selection method combines the hierarchical weight coefficient and the size of the evaluation indicator. The selection probability calculation formula is as follows:

[0104]

[0105] Among them, P i is the probability of individual i being selected, g i is the evaluation index, α i is the hierarchical weight coefficient of individual i;

[0106] Figure 2 The steps to complete the construction process of the three subpopulations are described in

[15] .

[0107] Step S104 specifically includes:

[0108] Figure 3 The parallel coevolution process of heterogeneous multi-populations is illustrated in

[15] .

[0109] The collaborative genetic algorithm in parallel branch 1 is as follows: First, a roulette wheel selection method is used to combine subpopulation 1 with external collaborative population 1 to construct a parent population. Subsequently, individuals are randomly selected from the parent population to perform collaborative crossover and mutation operations. For homologous individuals from the same population, a single-point crossover strategy is used to complete crossover by setting breakpoints in the path and exchanging the path segments following the breakpoints. For heterologous individuals from different populations, a two-point elitist crossover strategy is used to exchange the path segments between the two endpoints. After crossover, the mutation phase begins: homologous individuals undergo global random mutation, randomly selecting a feasible grid from the map to replace the current grid. For heterologous individuals, local random mutation is used, randomly selecting a feasible grid within a radius R centered on the current grid to replace the current grid. Finally, the generated path is repaired by avoiding obstacles, backtracking to correct conflicting nodes, and inserting detour nodes for adjustment.

[0110] Figure 4 The cooperative crossover and mutation operations of the parallel branch-cooperative genetic algorithm are described in.

[0111] Furthermore, the collaborative particle swarm algorithm in the parallel branch 2 is specifically as follows:

[0112] Step (6.1): Execute the speed update operation. The collaborative particle swarm algorithm adopts a hybrid speed update operator to integrate the best individual from the individual history, the sub-population, and the external collaborative population to improve the search capability of the particle swarm algorithm. The speed update formula is:

[0113]

[0114] ω=ω0·exp(-η·t / T max ) (39)

[0115] Among them, v i represents the current speed of individual i, x i , pbest i , gbest p , gbest e are the current position of individual i, the historical optimal position, the global optimal position of subpopulation 1, and the group optimal position of the corresponding external collaborative population; represents the direction difference operator, which calculates the difference between the moving direction vector of the optimal path and the current path at the same time step, and rounds the result; c1, c2 and c3 are learning factors, r1, r2 and r3 are independent random numbers, ω is the inertia weight, and the attenuation strategy is adopted, where ω0 is the initial inertia weight, η is the attenuation rate factor, t is the current iteration number, and T max is the total number of iterations;

[0116] Step (6.2): ​​Perform the position update operation. The particle moves according to the updated speed. The position update formula is as follows;

[0117]

[0118] Among them, x i and v i are the current position and velocity of individual i, respectively, and t is the current iteration number;

[0119] Step (6.3): Execute the path obstacle avoidance repair operation. The updated particles must meet the obstacle avoidance requirements. If the path node falls into the obstacle grid, move the infeasible node to bypass the obstacle.

[0120] Furthermore, the collaborative ant colony algorithm in the parallel branch three is executed, specifically:

[0121] Step (7.1): Collaborative pheromone injection operation, extracting path nodes in subpopulation 3 and external collaborative population 3, updating pheromone benchmark distribution, and providing a high-quality search starting point for the ant colony algorithm. The calculation formula for the collaborative pheromone injection is as follows:

[0122]

[0123] Among them, τ ij (t) is the pheromone concentration moving from grid i to adjacent grid j at the tth iteration, t = 0 is the initial pheromone concentration, ρ is the volatility coefficient, γ is the cross-population pheromone fusion coefficient, Q is the pheromone intensity coefficient, L k is the elite path length, Indicates whether ant k passes through the grid edge (i, j), K represents all paths in subpopulation 3 and external collaborative population 3;

[0124] Step (7.2): Solution path construction operation, the ant selects the path node according to the transition probability and gradually generates the solution path; if the node falls into the obstacle grid, it backtracks to the nearest free grid and increases the invalid path penalty weight during replanning to guide obstacle avoidance; the transition probability The formula is the probability that an ant moves from the current grid i to the adjacent grid j, and its calculation formula is as follows:

[0125]

[0126] Among them, η ij is the heuristic factor, d ij is the Manhattan distance from grid i to the target point, ξ is the anti-zero constant, τ ijis the pheromone concentration moving from the current grid i to the adjacent grid j, α is the pheromone weight factor; β is the heuristic information weight factor, s represents a feasible neighborhood of the current position, and A represents the feasible neighborhood set of ant k at the current node;

[0127] Step (7.3): Pheromone update operation. After the solution path is constructed, the elite retention and global collaboration hybrid model is used to update the pheromone, strengthen the high-quality path, and improve the global search ability and convergence speed. The calculation formula is:

[0128]

[0129] in, It represents the local pheromone increment after the ants release it. Represents the pheromone increment of the global elite path, N a is the number of ants in the current ant colony, λ g is the global update coefficient, M represents the elite path with high fitness in the current ant colony, I is the set of invalid paths, σ is the penalty weight of invalid paths, τ ij is the pheromone concentration moving from the current grid i to the adjacent grid j, t is the current iteration number, k represents the ant individual, Indicates whether ant k passes through the grid edge (i, j), ρ is the volatility coefficient, Q is the pheromone intensity coefficient, L k is the elite path length.

[0130] Step S105 specifically includes:

[0131] The individuals of each branch subpopulation are divided into elite group and diversity group according to the comprehensive fitness value F;

[0132] Then, two empty solution pools are initialized, namely the elite pool and the diversity pool. The elite pool is populated by selecting individuals from the elite group of each subpopulation through a tournament selection method with a selection pressure of k. The diversity pool is populated by selecting individuals from the diversity group of each subpopulation using random sampling. Finally, the elite pool and the diversity pool are merged to generate a new global solution set.

[0133] Figure 5 The process of forming the global solution set for one of the parallel algorithm branches is shown.

[0134] Step S106 specifically includes:

[0135] Step (9.1): Perform spatiotemporal conflict detection on all path individuals in the global solution set and identify the following three types of conflicts: (1) occupancy conflict, i.e., two robots occupy the same grid at the same time step; (2) crossing conflict, i.e., two robots cross each other along adjacent paths at the same time step; (3) exchange conflict, i.e., two robots exchange positions in adjacent time steps;

[0136] Step (9.2): Assign priorities according to the urgency of the robot's tasks. The robot only needs to consider conflicts with higher-priority tasks, and resolve conflicts in descending order of priority. Execute the following three resolution strategies in sequence. If the conflict is resolved, a conflict-free path is output: (1) Time offset strategy: When a single-point conflict is detected, insert a waiting node for the low-priority robot to delay its path time step; (2) Path detour strategy: Insert detour path points before and after the conflict area for the low-priority robot; (3) Local replanning strategy: Fix the path of the high-priority robot and replan the path of the low-priority robot in the conflict area;

[0137] After the above strategy is executed, the conflict status is rechecked. If a conflict still exists, the low-priority robot path is regenerated until there is no conflict with other robot paths.

[0138] Figure 6 The multi-robot hybrid conflict resolution flow chart is shown.

[0139] Step S107 specifically includes:

[0140] Step (10.1): First, directly connect the path points of the optimal individual path to the subsequent non-adjacent path points in sequence, skipping the intermediate points if feasible, and then randomly move each path point towards the midpoint between the two adjacent path points, replacing the current path point if feasible, until all path nodes are processed;

[0141] Step (10.2): Use a cubic B-spline curve to fit the path points to generate a continuous smooth path. The calculation expression of the cubic B-spline method is as follows:

[0142]

[0143] Among them, P is the coordinate value of a point on the cubic B-spline path curve, n is the maximum subscript of the path point, Q i =(x i ,y i ) represents the coordinates of the i-th path point, B i,3 (u) is the cubic B-spline basis function, which is represented by a piecewise polynomial. u is the normalized curve parameter, and the path point set is generated by fitting the original path node coordinates using the least squares method.

[0144] Figure 7 The path after processing based on the cubic B-spline curve of path reconstruction is shown. It can be seen that the optimized path is smoother.

Claims

1. A multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search, characterized by: The method comprises: Step S101: Use the two-dimensional grid method to establish a multi-robot working environment model and initialize parameter information, including map size, number of robots, size of each robot population, number of path nodes, static obstacle information, maximum number of iterations, each robot's starting point, target point, and parameters related to each algorithm branch; Step S102: Initialize the total population of multiple robots using the risk-aware A* algorithm; Step S103: Using the population grouping strategy, based on the fitness evaluation of path length, smoothness and safety, the hierarchical roulette wheel selection method is used to divide the total population into three heterogeneous sub-populations and the corresponding external collaborative population; Step S104: Heterogeneous multi-population collaborative parallel optimization, executing three algorithm branches in parallel, and realizing collaborative evolution through mutual communication between the sub-population and the corresponding external collaborative population. Parallel branch 1 executes the collaborative genetic algorithm, parallel branch 2 executes the collaborative particle swarm algorithm, and parallel branch 3 executes the collaborative ant colony algorithm; Step S105: Generate a global solution set using a dual solution pool fusion mechanism, retain the elite solution and diversity solution of each branch to form an elite pool and a diversity pool; merge the two solution pools to form a global solution set; Step S106: Use a priority-based conflict resolution strategy to resolve conflicts among multiple robots. First, perform spatiotemporal conflict detection on the paths of multiple robots. Then, assign priorities to the robots based on their task urgency. Finally, perform conflict resolution in descending order of priority to resolve conflicts among multiple robots. Step S107: Determine whether the maximum number of iterations has been reached. If not, add 1 to the number of iterations and return to S103. If the maximum number of iterations has been reached, smooth the optimal individual paths of the multiple robots using a cubic B-spline curve based on path reconstruction, and finally output the smoothed path.

2. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1 is characterized in that: In step S101, a two-dimensional grid method is used to establish a multi-robot working environment model, specifically: First, the working environment is discretized into rectangular grid cells, and a Cartesian coordinate system with the lower left corner as the origin is constructed. Then, a matrix is ​​used to store the map, where the obstacle grid is coded as -1 and the passable area grid is coded as d, where d is the Manhattan distance between the grid and the nearest obstacle.

3. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1 is characterized in that: The risk-aware A* algorithm is used in step S102 to initialize the total population of multiple robots, specifically: The path individuals are initialized by the risk-aware A* algorithm. Each individual is encoded as the joint path of all robots. The encoding structure is: P={P1,P2,…,P i ,…,P m } (1) Where m represents the total number of robots, represents the path sequence of the i-th robot, The grid coordinates of path point j, is the time when the robot reaches the grid; The specific steps of the risk-aware A* algorithm are as follows: Step (3.1): Set the starting point and target point as the initial parent node and the ending parent node respectively; Step (3.2): Check the feasibility of the domain grid of the current node, calculate the cost value of the feasible domain nodes, and select the domain grid node with the smallest cost value as the next parent node. The cost value calculation adopts the heuristic cost function f(n) that considers risk factors. Its calculation formula is as follows: f(n)=g(n)+h(n)+R(n) (2) R(n)=exp(-λ·d(n)) (3) Where g(n) is the actual distance traveled from the starting point to node n, h(n) is the Manhattan distance from node n to the target point, R(n) represents the risk value of node n, d(n) represents the Euclidean distance from node n to the nearest obstacle, and λ is the attenuation coefficient, which controls the rate at which risk decays with distance. Step (3.3): Repeat step (3.2) until the target point falls into the neighborhood grid node of the target point, trace back to the parent node, and connect the lines in sequence to generate the initial path.

4. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1 is characterized in that: The population grouping strategy in step S103 is specifically: Step (4.1): Arrange the robot individuals in descending order according to their comprehensive fitness F and divide them equally into three levels. The fitness F calculation formula is: F=β1·f l +β2·f s +β3·f a (4) Among them, f l is the total path length, f s is the path smoothness, f a is the path safety, β1, β2, β3 are weight coefficients, and β1+β2+β3=1; Step (4.2): First, the total population is divided into three subpopulations based on the comprehensive fitness F. Then, three corresponding external collaborative populations are constructed using path length, smoothness, and safety as evaluation indicators. All populations are constructed using the stratified roulette wheel selection method. The subpopulations select individuals from the total population, and the external collaborative populations select individuals from the two subpopulations other than their corresponding subpopulations. The stratified roulette wheel selection method combines the hierarchical weight coefficient and the size of the evaluation indicator. The selection probability calculation formula is as follows: Among them, P i is the probability of individual i being selected, g i is the evaluation index, α i is the hierarchical weight coefficient of individual i.

5. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1 is characterized in that: The heterogeneous multi-population collaborative parallel optimization in step S104, and the collaborative genetic algorithm in the parallel branch 1, are specifically: First, a roulette wheel selection method is used to combine subpopulation 1 and external collaborative population 1 to construct a parent population. Subsequently, individuals are randomly selected from the parent population to perform collaborative crossover and mutation operations. For homologous individuals from the same population, a single-point crossover strategy is adopted to complete the crossover by setting breakpoints in the path and exchanging the path segments following the breakpoints. For heterologous individuals from different populations, a two-point elite crossover strategy is adopted to exchange the path segments between the two end points. After crossover, the mutation phase begins: homologous individuals undergo global random mutation to randomly select a feasible grid from the map to replace the current grid; heterologous individuals undergo local random mutation to randomly select a feasible grid within a radius R centered on the current grid to replace the current grid. Finally, the generated path is repaired by obstacle avoidance, conflicting nodes are backtracked, and detour nodes are inserted for adjustment.

6. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1 is characterized in that: The heterogeneous multi-population collaborative parallel optimization in step S104 and the collaborative particle swarm algorithm in the parallel branch 2 are specifically as follows: Step (6.1): Execute the speed update operation. The collaborative particle swarm algorithm adopts a hybrid speed update operator to integrate the best individual from the individual history, the sub-population, and the external collaborative population to improve the search capability of the particle swarm algorithm. The speed update formula is: ω=ω0·exp(-η·t / T max ) (7) Among them, v i represents the current speed of individual i, x i , pbest i , gbest p , gbest e are the current position of individual i, the historical optimal position, the global optimal position of subpopulation 1, and the group optimal position of the corresponding external collaborative population; represents the direction difference operator, which calculates the difference between the moving direction vector of the optimal path and the current path at the same time step, and rounds the result; c1, c2 and c3 are learning factors, r1, r2 and r3 are independent random numbers, ω is the inertia weight, and the attenuation strategy is adopted, where ω0 is the initial inertia weight, η is the attenuation rate factor, t is the current number of iterations, and T max is the total number of iterations; Step (6.2): ​​Perform the position update operation. The particle moves according to the updated speed. The position update formula is as follows; Among them, x i and v i are the current position and velocity of individual i, respectively, and t is the current iteration number; Step (6.3): Execute the path obstacle avoidance repair operation. The updated particles must meet the obstacle avoidance requirements. If the path node falls into the obstacle grid, move the infeasible node to bypass the obstacle.

7. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1 is characterized in that: The heterogeneous multi-population collaborative parallel optimization in step S104 and the collaborative ant colony algorithm in the parallel branch three are specifically as follows: Step (7.1): Collaborative pheromone injection operation, extracting path nodes in subpopulation 3 and external collaborative population 3, updating pheromone benchmark distribution, and providing a high-quality search starting point for the ant colony algorithm. The calculation formula for the collaborative pheromone injection is as follows: Among them, τ ij (t) is the pheromone concentration moving from grid i to adjacent grid j at the tth iteration, t = 0 is the initial pheromone concentration, ρ is the volatility coefficient, γ is the cross-population pheromone fusion coefficient, Q is the pheromone intensity coefficient, L k is the elite path length, Indicates whether ant k passes through the grid edge (i, j), K represents all paths in subpopulation 3 and external collaborative population 3; Step (7.2): Solution path construction operation, the ant selects the path node according to the transition probability and gradually generates the solution path; if the node falls into the obstacle grid, it backtracks to the nearest free grid and increases the invalid path penalty weight during replanning to guide obstacle avoidance; the transition probability The formula is the probability that an ant moves from the current grid i to the adjacent grid j, and its calculation formula is as follows: Among them, η ij is the heuristic factor, d ij is the Manhattan distance from grid i to the target point, ξ is the anti-zero constant, τ ij is the pheromone concentration moving from the current grid i to the adjacent grid j, α is the pheromone weight factor; β is the heuristic information weight factor, s represents a feasible neighborhood of the current position, and A represents the feasible neighborhood set of ant k at the current node; Step (7.3): Pheromone update operation. After the solution path is constructed, the elite retention and global collaboration hybrid model is used to update the pheromone, strengthen the high-quality path, and improve the global search ability and convergence speed. The calculation formula is: in, It represents the local pheromone increment after the ants release it. Represents the pheromone increment of the global elite path, N a is the number of ants in the current ant colony, λ g is the global update coefficient, M represents the elite path with high fitness in the current ant colony, I is the set of invalid paths, σ is the penalty weight of invalid paths, τ ij is the pheromone concentration moving from the current grid i to the adjacent grid j, t is the current iteration number, k represents the ant individual, Indicates whether ant k passes through the grid edge (i, j), ρ is the volatility coefficient, Q is the pheromone intensity coefficient, L k is the elite path length.

8. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1 is characterized in that: The dual solution pool fusion mechanism in step S105 generates a global solution set, specifically: The individuals of each branch subpopulation are divided into an elite group and a diversity group according to the comprehensive fitness value F; then two empty solution pools, namely the elite pool and the diversity pool, are initialized; among them, the elite pool is filled by selecting individuals from the elite group of each subpopulation through the tournament selection method with a selection pressure of k; the diversity pool is filled by selecting individuals from the diversity group of each subpopulation by random sampling; finally, the elite pool and the diversity pool are merged to generate a new global solution set.

9. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1 is characterized in that: The priority-based conflict resolution strategy in step S106 is used to resolve multi-robot conflicts, specifically: Step (9.1): Perform spatiotemporal conflict detection on all path individuals in the global solution set and identify the following three types of conflicts: (1) occupancy conflict, i.e., two robots occupy the same grid at the same time step; (2) crossing conflict, i.e., two robots cross each other along adjacent paths at the same time step; (3) exchange conflict, i.e., two robots exchange positions in adjacent time steps; Step (9.2): Assign priorities according to the urgency of the robot's tasks. The robot only needs to consider conflicts with higher-priority tasks, and resolve conflicts in descending order of priority. Execute the following three resolution strategies in sequence. If the conflict is resolved, a conflict-free path is output: (1) Time offset strategy: When a single-point conflict is detected, insert a waiting node for the low-priority robot to delay its path time step; (2) Path detour strategy: Insert detour path points before and after the conflict area for the low-priority robot; (3) Local replanning strategy: Fix the path of the high-priority robot and replan the path of the low-priority robot in the conflict area; After the above strategy is executed, the conflict status is rechecked. If a conflict still exists, the low-priority robot path is regenerated until there is no conflict with other robot paths.

10. The multi-robot path planning method based on multi-heterogeneous population co-evolution and parallel search according to claim 1, characterized in that: The path smoothing based on the cubic B-spline curve of the path reconstruction in step S107 is specifically as follows: Step (10.1): First, directly connect the path points of the optimal individual path to the subsequent non-adjacent path points in sequence, skipping the intermediate points if feasible, and then randomly move each path point towards the midpoint between the two adjacent path points, replacing the current path point if feasible, until all path nodes are processed; Step (10.2): Use a cubic B-spline curve to fit the path points to generate a continuous smooth path. The calculation expression of the cubic B-spline method is as follows: Among them, P is the coordinate value of a point on the cubic B-spline path curve, n is the maximum subscript of the path point, Q i =(x i ,y i ) represents the coordinates of the i-th path point, B i,3 (u) is the cubic B-spline basis function, which is represented by a piecewise polynomial. u is the normalized curve parameter, and the path point set is generated by fitting the original path node coordinates using the least squares method.

Images