A heterogeneous unmanned platform cooperative task planning method

Abstract

The present application relates to a kind of heterogeneous unmanned platform cooperative task planning method, belong to unmanned system task planning technical field.A kind of heterogeneous unmanned platform cooperative task planning method, comprising: based on the task point set and warehouse set of job environment map acquisition;According to unmanned operation body configuration information, obtain unmanned operation body type set and the corresponding task demand information of each type;Task demand information and warehouse set are mixed and arranged coding to generate candidate solution;Candidate solution is decoded to obtain the execution path of multiple unmanned operation bodies;The maximum completion time of candidate solution is calculated;With the double target of minimizing total execution path and maximum completion time, obtain optimal solution set by multi-objective evolutionary search;From optimal solution set, select target solution as cooperative task planning result.Heterogeneous unmanned platform is realized under the double target of minimizing total execution path and maximum completion time Efficient cooperative task planning.

Description

Technical Field

[0001] This invention relates to the field of unmanned system mission planning technology, and in particular to a collaborative mission planning method for heterogeneous unmanned platforms. Background Technology

[0002] With the rapid development of unmanned operation technology, unmanned platforms have been widely used in logistics distribution, environmental monitoring, agricultural plant protection, and emergency rescue. In actual operation scenarios, a single type of unmanned operation unit often cannot meet the needs of complex tasks. Therefore, collaborative operation of heterogeneous unmanned platforms has become an important development direction. Heterogeneous unmanned platforms are usually composed of multiple types of unmanned operation units. Different types of operation units have different functional characteristics and capability constraints, and need to cooperate with each other to complete multiple types of operation services at the same task point. Task planning, as the core link of collaborative operation of unmanned platforms, aims to allocate tasks and plan execution paths for each unmanned operation unit under various constraints to achieve optimal overall operation efficiency.

[0003] Existing technologies are mostly designed for single-type unmanned work entities in terms of coding methods, which makes it difficult to effectively express the service requirements and sequential constraints of multiple types of work entities in heterogeneous platforms for the same task point. In terms of optimization objectives, they usually only consider a single objective and lack a comprehensive trade-off between path cost and time efficiency. In terms of evolutionary search strategies, the crossover and mutation operations of traditional genetic algorithms are prone to destroying the legitimacy of the solution and it is difficult to take into account both global exploration and local optimization, resulting in limited solution efficiency and solution quality, making it difficult to efficiently solve large-scale heterogeneous collaborative task planning problems. Summary of the Invention

[0004] Based on the above analysis, the embodiments of the present invention aim to provide a collaborative task planning method for heterogeneous unmanned platforms, in order to solve the problems in existing collaborative task planning for heterogeneous unmanned platforms, such as the difficulty in effectively expressing multi-type task requirements and cross-type sequential constraints, the inability to comprehensively optimize path cost and time efficiency, and the insufficient search efficiency and solution quality of traditional evolutionary algorithms.

[0005] On one hand, embodiments of the present invention provide a collaborative task planning method for heterogeneous unmanned platforms, including: Obtain the task point set and warehouse set based on the work environment map; Based on the preset unmanned operation configuration information, obtain the set of unmanned operation types and the corresponding task requirements information for each type of unmanned operation. The task requirement information and the warehouse set are mixed and encoded to generate candidate solutions for collaborative task planning of multiple unmanned operations. The candidate solutions are decoded to obtain multiple execution paths for the unmanned operation. The total execution path and maximum completion time of the candidate solutions are calculated based on the execution paths of multiple unmanned operating entities. Using minimizing the total execution path and minimizing the maximum completion time as dual objective functions, a multi-objective evolutionary search is performed on the candidate solutions to obtain the optimal solution set; The target solution is selected from the set of optimal solutions as the result of the collaborative task planning.

[0006] Furthermore, the step of performing a mixed permutation and encoding of the task requirement information and the warehouse set to generate candidate solutions for collaborative task planning of multiple unmanned operations includes: A task requirement matrix is ​​constructed based on the task requirement information, wherein the task requirement matrix is ​​arranged with task points as rows and unmanned operation body types as columns, and the matrix elements are used to characterize whether the corresponding type of unmanned operation body needs to perform operations at the corresponding task points. A corresponding access token is generated for each non-zero element in the task requirement matrix, wherein the access token uniquely corresponds to a service requirement of a task point and an unmanned operation body type, and the total number of access tokens is equal to the sum of all non-zero elements in the task requirement matrix. Several warehouse partition tokens are generated based on the number of warehouses in the warehouse set, wherein the warehouse partition tokens are used to divide the candidate solution into multiple warehouse access segments; Based on the preset execution order of unmanned operations, all access tokens and warehouse separation tokens are arranged to obtain the candidate solutions.

[0007] Further, the decoding of the candidate solutions yields multiple execution paths for the unmanned operation, including: Acquire the capacity of various types of unmanned operating systems; The access sequence is divided into multiple warehouse segments based on the position of the warehouse separator token in the candidate solution; Extract task sequences according to the type of unmanned operation within each warehouse section; The task sequence is segmented based on the capacity of each type of unmanned operation body to obtain multiple unmanned operation body execution paths that meet the capacity constraints.

[0008] Furthermore, the step of segmenting the task sequence based on the capacity to obtain multiple unmanned operation body execution paths that satisfy the capacity constraints includes: The demand for each task point is accumulated sequentially along the task sequence; When adding the next task point would cause the cumulative demand to exceed the capacity of the corresponding type of unmanned operation body, the currently accumulated task point is taken as an execution path, and the next task point is taken as the starting point of the new execution path, and the demand is accumulated again. Repeat the above process until all task points in the task sequence have been assigned, thereby forming multiple unmanned operation body execution paths that meet capacity constraints.

[0009] Furthermore, the method also includes calculating the total execution path, including: Obtain the coordinates of each task point in the task point set and the coordinates of each warehouse in the warehouse set, and generate a distance matrix based on the length of the passable path between each task point and each warehouse. For each unmanned operation body execution path, the travel distance from the warehouse to the warehouse is calculated based on the distance matrix, sequentially visiting each task point on the path and returning to the warehouse. The total execution path is obtained by summing the travel distances of all unmanned operation paths.

[0010] Furthermore, the step of using minimizing the total execution path and minimizing the maximum completion time as dual objective functions to perform a multi-objective evolutionary search on the candidate solutions to obtain the optimal solution set includes: Step a: Construct an initial population based on the candidate solutions, and use the initial population as the current population; Step b: Calculate the non-dominance level and crowding distance for each individual in the current population, and select elite individuals from the current population based on the non-dominance level and crowding distance to retain for the next generation; Step c: Select parent individuals from the current population using a tournament selection method to obtain the parent population; Step d: Perform mutation operation on the parent population to obtain the offspring population, merge the parent population and the offspring population to obtain the merged population, and select the next generation population based on the merged population. Update the next generation population to the current population, repeat step bd until the preset termination condition is met, and output all non-dominated solutions in the current population as the optimal solution set.

[0011] Furthermore, based on the execution order of the unmanned operation, all access tokens and warehouse separation tokens are arranged to obtain the candidate solutions, including: For each task point, based on the sequential relationship between different types of unmanned operating entities within that task point, the corresponding access tokens are randomly sorted to satisfy that sequential relationship, thus obtaining the access token sequence within that task point. The access token sequences within all task points are randomly and alternately extracted to form a global access sequence; Insert a warehouse separator token into the global access sequence to obtain candidate solutions.

[0012] Furthermore, the offspring population is obtained by performing mutation operations on the parent population, including: Perform an adaptive mixing and crossover operation on the parent individuals in the parent population to generate offspring individuals; Perform a lightweight mutation operation on the generated offspring individuals to obtain mutated offspring individuals; Perform local order fine-tuning on some of the mutated offspring individuals to obtain fine-tuned offspring individuals; The offspring population is obtained based on the finely adjusted offspring individuals.

[0013] Furthermore, the adaptive mixing and crossover operation performed on the parent individuals in the parent population includes: The selection probability of the crossover method is determined based on the evolution progress, where the evolution progress is determined by the ratio of the number of function evaluations completed to the preset maximum number of function evaluations, or by the ratio of the current iteration number to the preset maximum iteration number. The crossover method is determined based on the selection probability; Based on the determined crossover method, the middle segment is selected as the crossover region in the arrangement of the parent individuals, and the elements in the crossover region are recombined according to the element recombination rules corresponding to the crossover method to generate the offspring individuals.

[0014] On the other hand, embodiments of the present invention provide a heterogeneous unmanned platform collaborative task planning system, the system comprising: The acquisition module is used to acquire the task point set and warehouse set based on the operation environment map; and to acquire the unmanned operation body type set and the task requirement information corresponding to each type of unmanned operation body according to the preset unmanned operation body configuration information. The encoding module is used to perform mixed-arrangement encoding on the task requirement information and the warehouse set to generate candidate solutions for collaborative task planning of multiple unmanned operations. The decoding module is used to decode the candidate solutions to obtain multiple execution paths for the unmanned operation body; The calculation module is used to calculate the total execution path and maximum completion time of the candidate solution based on the execution paths of multiple unmanned operation bodies. The search module is used to perform a multi-objective evolutionary search on the candidate solutions, taking minimizing the total execution path and minimizing the maximum completion time as dual objective functions, to obtain the optimal solution set. The output module is used to select the target solution from the set of optimal solutions as the result of the collaborative task planning.

[0015] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. To address the problem of different service requirements for the same task point among multiple types of unmanned platforms in heterogeneous systems, and the need to satisfy cross-type priority constraints, this invention constructs a task requirement matrix and generates access tokens based on non-zero elements. At the same time, it introduces warehouse separator tokens for mixed arrangement encoding, so that candidate solutions can intuitively express the task allocation and warehouse association relationship of each type of unmanned operation. Combined with the priority constraint initialization strategy, the legality of the encoding is effectively guaranteed, avoiding the shortcomings of traditional encoding methods in expressing heterogeneous requirements. This significantly improves the modeling accuracy and solution feasibility of the heterogeneous collaborative task planning problem.

[0016] 2. To address the problem that existing methods struggle to simultaneously optimize path cost and time efficiency, this invention constructs a multi-objective optimization model by using minimizing the total execution path and minimizing the maximum completion time as dual objective functions. Through non-dominated sorting and congestion distance mechanisms for environment selection, it can obtain a uniformly distributed optimal solution set in a single run. This provides decision-makers with diverse solutions that balance different degrees of path cost and time efficiency, solving the problem that single-objective optimization cannot meet the multi-dimensional decision-making needs in actual operations.

[0017] 3. To address the problems of low search efficiency, easy violation of solution validity, and insufficient local optimization ability of traditional evolutionary algorithms in heterogeneous task planning problems, this invention designs an adaptive hybrid crossover operation, which dynamically adjusts the selection probability of the crossover method according to the evolutionary progress, taking into account both global exploration and local development. At the same time, it adopts lightweight mutation operation and local order fine-tuning operation, including inversion mutation, insertion mutation, 2-opt operation and local order correction, to eliminate order redundancy without destroying the uniqueness of tokens, effectively improving the convergence speed and solution quality of the algorithm, and realizing efficient solution of large-scale heterogeneous collaborative task planning problems.

[0018] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Figure 1 is a schematic flowchart of a collaborative task planning method for heterogeneous unmanned platforms provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a hybrid permutation encoding provided in an embodiment of the present invention; Figure 3This is a schematic diagram of cross-type sequential constraints provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a multi-objective evolutionary search provided in an embodiment of the present invention. Detailed Implementation

[0020] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0021] The technical solution of the present invention will be described in detail below with reference to specific application scenarios.

[0022] Example 1 This embodiment uses agricultural robot collaborative operation as an example, but the present invention is also applicable to multi-task joint service scenarios such as park logistics distribution, facility inspection, and drone delivery. Figure 1 As shown, a collaborative task planning method for heterogeneous unmanned platforms includes: S1: Obtain the set of task points and the set of warehouses based on the work environment map.

[0023] The task point set refers to the collection of all locations where unmanned vehicles need to perform operations, such as the coordinates of points in a field where fertilization is needed. The warehouse set refers to the collection of all base locations where unmanned vehicles can dock, charge, or load supplies, such as farm machinery warehouses. First, import a raster map or topology map of the target area, such as a 2D occupied raster map generated using open-source map building tools like ROS's gmapping. Extract passable areas from this map and select task points according to preset distribution rules, such as evenly selecting coordinates of points in a field where fertilization or spraying is needed, forming the task point set. The warehouse set is determined based on the actual deployment location; for example, the location of a farm machinery warehouse or charging station may be used as the warehouse coordinates.

[0024] S2: Obtain the set of unmanned operation body types and the task requirement information corresponding to each type of unmanned operation body based on the preset unmanned operation body configuration information.

[0025] Unmanned operation entities include wheeled unmanned vehicles, drones, and robotic arm platforms. Task requirement information refers to the specific task points where each type of unmanned operation entity needs to perform operations. For example, unmanned vehicles need to perform transportation operations at task points 1, 3, and 5, while drones need to perform spraying operations at task points 2 and 4. Task requirement information is uploaded based on actual operational needs and retrieved through task configuration files or a database.

[0026] S3: The task requirement information and the warehouse set are mixed and encoded to generate candidate solutions for collaborative task planning of multiple unmanned operations.

[0027] The task requirements and warehouse information are encoded into sequences and used as the objects of multi-objective search. Candidate solutions refer to possible task allocation and execution order schemes. Furthermore, candidate solutions for multi-unmanned collaborative task planning include: S31: Construct a task requirement matrix based on the task requirement information; The task requirement matrix is ​​a two-dimensional table. Rows correspond to individual task points in the task point set, and columns correspond to individual unmanned operation body types in the unmanned operation body type set. Each element of the matrix takes a value of 0 or 1, where a value of 1 indicates that the task point in that row requires the unmanned operation body type in that column to perform the operation, and a value of 0 indicates that it does not. For example, assuming there are 5 task points and 3 unmanned operation body types, the matrix would be 5 rows and 3 columns. If task point 1 requires both type 1 and type 2, then the elements in the first row and first column, and the elements in the first row and second column, would be 1, and the elements in the first row and third column would be 0. A zero matrix with dimensions equal to the total number of task points multiplied by the total number of types can be created using Python's NumPy library. Then, all task points and types can be iterated through, and the elements at the corresponding positions can be set to 1 according to the actual requirements.

[0028] S32: Generate a corresponding access token for each non-zero element in the task requirement matrix; An access token is an identifier used to represent a specific job requirement. Each non-zero element in the task requirement matrix, i.e., each position with a value of 1, represents a task point that requires a certain type of unmanned work unit service. A unique access token is generated for each such non-zero element. Access tokens can be encoded using integers, starting from 0 and incrementing. Let the total number of non-zero elements in the task requirement matrix be N, then the set of access tokens is 0, 1, 2, ..., N-1. Each access token is bound to a triple: task point number, unmanned work unit type, and service requirement identifier. The service requirement identifier indicates the specific job task that the task point requires to be performed by a certain type of unmanned work unit.

[0029] S33: Generate several warehouse partition tokens based on the number of warehouses in the warehouse set; like Figure 2As shown, a warehouse separator token is an identifier used to divide a task sequence into different warehouses in a permutation. If the number of warehouses in the warehouse set is m, then m-1 warehouse separator tokens are generated. Warehouse separator tokens and access tokens do not overlap numerically. For example, if the access tokens are numbered from 0 to N-1, then the warehouse separator tokens can take values ​​of N, N+1, ..., N+m-2. Each warehouse separator token is not bound to any task point or type; its function is to act as a separator in the permutation, dividing the permutation into multiple consecutive subsequences, each subsequence corresponding to the task access order of a warehouse. For example, if there are 3 warehouses, then 2 warehouse separator tokens are generated. The portion before the first separator token corresponds to the first warehouse, the portion between the two separator tokens corresponds to the second warehouse, and the portion after the last separator token corresponds to the third warehouse.

[0030] S34: Based on the execution order of the unmanned operation, arrange all access tokens and warehouse separation tokens to obtain the candidate solutions.

[0031] Candidate solutions express the order in which tasks are accessed, the type of unmanned work unit serving each task, and the task division between different warehouses. For example, assuming access tokens are 0, 1, 2, 3, and warehouse separator tokens are A, B, then the arrangement [2, A, 0, 3, B, 1] indicates that the task corresponding to token 2 is processed first, separator token A indicates switching to the second warehouse, then tokens 0 and 3 are processed, separator token B indicates switching to the third warehouse, and finally token 1 is processed. Further, S34 includes: S341: For each task point, according to the sequential relationship between the various types of unmanned operating entities within the task point, the corresponding access tokens are randomly sorted to satisfy the sequential relationship, thus obtaining the access token sequence within the task point. For each task point, a directed acyclic graph (DAG) is constructed based on the sequential relationship between different types of jobs within that task point. Then, a random topological sort is performed to obtain the access token sequence within that task point. A DAG is a graph representing dependencies, where nodes represent different access tokens at that task point, and directed edges represent the sequential constraints between types. For example, if task point i requires unmanned job types 1, 2, and 3, with the constraint that type 1 must precede type 2, and type 2 must precede type 3, then the DAG contains two directed edges: one from type 1 to type 2, and the other from type 2 to type 3. If there is no dependency between two types of unmanned jobs, they can appear in any order.

[0032] S342: Randomly and alternately extract access token sequences within all task points to form a global access sequence; After obtaining the internal sequence of all task points, a multi-way merge random sampling method is adopted: each time, a task point is randomly selected from the remaining tokens of each task point, and its next token is taken out, forming a global access sequence in sequence, thereby increasing the diversity of the population.

[0033] S343: Insert a warehouse separator token into the global access sequence to obtain candidate solutions.

[0034] Let the number of warehouses be m, then there are m-1 warehouse partition tokens. Warehouse partition tokens are repeatedly and randomly inserted into the global access sequence. The resulting permutation is a candidate solution. Each insertion involves randomly selecting m-1 positions to insert a warehouse partition token. After insertion, the global access sequence is divided into m consecutive subsequences, each corresponding to the task access order of a warehouse.

[0035] S4: Decode the candidate solutions to obtain multiple unmanned operation body execution paths.

[0036] The decoding process converts the permutation into a practically executable unmanned work unit path. An unmanned work unit's execution path refers to the route a work unit takes from the warehouse, sequentially visits several task points, and returns to the warehouse. S4 includes: S41: Obtain the capacity of various types of unmanned operating bodies; The capacity of an unmanned operating unit is the maximum load or operating capacity that each unmanned operating unit can carry in a single operation. For example, Type 1 is an unmanned vehicle and Type 2 is a drone. Type 1 can carry a maximum of 100 kg of fertilizer in a single operation, while Type 2 can spray a maximum of 20 mu (approximately 3.3 hectares) in a single operation. The capacity parameters are read through the configuration file.

[0037] S42: Divide the access sequence into multiple warehouse segments according to the position of the warehouse separator token in the candidate solution; Traverse the candidate solutions, recording the index position of the warehouse separator token. Let the permutation length be equal to the total number of tokens. The separator tokens divide the permutation into several consecutive subsequences of warehouses. The portion between any two adjacent warehouse separator tokens is a warehouse segment. The portion before the first warehouse separator token is the first warehouse segment, and the portion after the last warehouse separator token is the last warehouse segment. Each subsequence corresponds to the task access order of a warehouse, and a warehouse segment refers to a sequence of tokens belonging to the same warehouse.

[0038] S43: Extract task sequences according to the type of unmanned operation within each warehouse section; A task sequence refers to the sequence of access tokens that need to be served sequentially by unmanned work entities of the same type under the same warehouse. For each warehouse segment, the access tokens are classified according to the type of unmanned work entity they are bound to, and the token subsequences of each type are extracted, maintaining their relative order in the original warehouse segment. For example, in the token sequence [5, 2, 7] within the warehouse segment, token 5 is bound to type 1, token 2 is bound to type 2, and token 7 is bound to type 1. Then the sequence for type 1 is [5, 7], and the sequence for type 2 is [2].

[0039] S44: The task sequence is segmented based on the capacity of each type of unmanned work body to obtain multiple unmanned work body execution paths that meet the capacity constraints.

[0040] For each type of unmanned work unit, its corresponding task sequence is sequentially divided into several sub-paths, with each sub-path corresponding to the execution route of one unmanned work unit. The division rule is as follows: S441: Accumulate the demand for each task point sequentially along the task sequence; The requirement for a task point refers to the amount of work that a task point needs to consume for that type of task, such as a task point needing to spray 10 liters of medicine. For each type, starting from the first token in the sequence, the total requirement for all tasks is summed to obtain the total requirement for the task sequence.

[0041] S442: When adding the next task point will cause the cumulative demand to exceed the capacity of the corresponding type of unmanned operation body, the currently accumulated task point is taken as an execution path, and the next task point is taken as the starting point of the new execution path. The demand is accumulated again until all task points in the task sequence are allocated, resulting in multiple unmanned operation body execution paths that meet the capacity constraints.

[0042] Let the capacity be For the current subpath, the current cumulative sum is S, and the demand for the next task point is... .like If the current subpath ends, the accumulated task points are recorded as an execution path. This execution path includes semantic information about sequentially visiting these task points from the respective repository and returning to the repository. Then, the next task point is used as the first point of the new subpath, and the process is reset. .

[0043] After performing the above segmentation on all tokens of the current type, several sub-paths are obtained. Each sub-path corresponds to the actual operation route of an unmanned operation unit. All sub-paths of all types in all warehouses together constitute the execution path set of this candidate solution.

[0044] S5: Calculate the total execution path and maximum completion time of the candidate solution based on the execution paths of multiple unmanned operation bodies; Based on cross-type sequential constraints, a waiting scheduling mechanism is used to calculate the maximum completion time of all paths. Cross-type sequential constraints refer to the sequential dependencies between different unmanned workpiece types at the same task point; for example, fertilization must follow land preparation. The waiting scheduling mechanism means that if a preceding task is not completed when a subsequent task arrives, the subsequent task will actively wait. S5 includes: S51: For each execution path, calculate the time when the unmanned workpiece arrives at each task point; Each unmanned work unit departs from the warehouse at time zero, moves along the path, and the time it arrives at the task point is the cumulative travel time, which is obtained by dividing the distance by the speed.

[0045] During implementation, the coordinates of each task point in the task point set and the coordinates of each warehouse in the warehouse set are obtained, and a distance matrix is ​​generated based on the traversable path length between each task point and each warehouse. The coordinates of each task point and each warehouse are obtained from the map, using Dijkstra's algorithm or A*. The algorithm calculates the shortest possible path length between any two points. For example, for a raster map, an eight-neighbor A... The algorithm uses Euclidean distance as the heuristic function to obtain a distance matrix D, where the element D(i,j) represents the shortest path length from node i to node j, and the nodes include all task points and the repository.

[0046] S52: As Figure 3 As shown, according to the preset cross-type sequential constraint relationship, when the unmanned work body of the subsequent type arrives at the same task point, if the unmanned work body of the preceding type at the task point has not yet completed the work, the unmanned work body of the subsequent type is made to wait until the unmanned work body of the preceding type completes the work. Cross-type precedence constraints are defined by triples (task point i, preceding type). Post-order type The form is given, indicating that at task point i, the type is... The assignment must be of type The task must be completed before the job begins. During scheduling, for each task point i and each type t, its completion time is recorded. The initial completion time is set to the arrival time plus the task duration. The duration of the task is obtained by dividing the demand at the task point by the operating rate of the unmanned work unit. When the subsequent type... When the unmanned workpiece arrives at task point i, if its preceding type completion time If the arrival time is greater than the current arrival time 'a', then the subsequent unmanned work unit must wait, and its actual start time is 'a'. Therefore, its completion time is: .

[0047] The maximum value among the final completion times of all unmanned operation paths is taken as the maximum completion time of the candidate solution.

[0048] The maximum completion time f2(x) is defined as the maximum value among the final completion times of all execution paths, as shown in the following formula: ; in, For candidate solutions Maximum completion time; A set of execution paths; This is one of the execution paths; For execution path The completion time of the last task point.

[0049] S6: Using minimizing the total execution path and minimizing the maximum completion time as dual objective functions, perform multi-objective evolutionary search on the candidate solutions to obtain the optimal solution set.

[0050] The calculation of the total execution path includes: For each unmanned operation path, the travel distance from the warehouse to the warehouse is calculated based on the distance matrix, sequentially visiting each task point on the path and returning to the warehouse.

[0051] Let the repository of a certain execution path r be... The sequence of task points visited is The travel distance is calculated as follows: ; in, This indicates the distance traveled along the execution path; This represents the shortest path length between two points in the distance matrix; This indicates the repository to which the path belongs; This indicates the first task point visited by this path; This indicates the path accessing the first One task point; This indicates the (k+1)th task point visited by the path; n represents the total number of task points on the path; k is the summation index variable, ranging from 1 to n-1. This represents the summation of k from 1 to n-1; This indicates the distance traveled from the last mission point back to the warehouse.

[0052] The total execution path is obtained by summing the travel distances of all unmanned operation paths.

[0053] The total execution path is the sum of the travel distances of all execution paths, as shown in the following formula: ; in, Let R be the total execution path of candidate solution x; R(x) be the set of all execution paths obtained by decoding candidate solution x; r is a set One of the execution paths in; The distance traveled along path r; Indicates to All paths in Sum.

[0054] like Figure 4 As shown, the improved NSGA-II algorithm can be used to perform multi-objective evolutionary search on the candidate solutions to obtain the optimal solution set.

[0055] Perform a multi-objective evolutionary search on the candidate solutions to obtain the optimal solution set, including: S61: Calculate the non-dominance level and crowding distance for each individual in the current population; The non-dominated ranking is a measure of Pareto optimality. For each individual in the population, the non-dominated ranking is calculated based on two objective values: the total execution path f1 and the maximum completion time f2. First, identify all individuals not dominated by any other individual. Domination means that individual A's two objective values ​​are both no worse than individual B's, and at least one is better than B's; these individuals are ranked 1. After removing these individuals, find a new batch of non-dominated individuals, ranked 2, and so on. Crowding distance is used to maintain population diversity, calculating the sum of distances between each individual and its neighbors in the objective space. The non-dominated ranking and crowding distance functions in Python's DEAP library can be used. DEAP is an open-source framework for fast prototyping and testing evolutionary algorithms.

[0056] S62: Select elite individuals from the current population based on non-dominance level and crowding distance to retain them for the next generation; An elite retention strategy is adopted, in which the 100 individuals with the lowest rank (i.e., the best) and the highest crowding in the current population are directly copied to the next generation, with a retention ratio of up to 10%.

[0057] S63: Select parent individuals from the current population using a tournament selection method; A binary tournament selection method is used, where two individuals are randomly selected from the population, and their non-dominance levels are compared. The individual with the lower dominance level wins; if the levels are the same, the individual with the higher crowding level wins. This process is repeated 100 or 200 times to select the parent individual.

[0058] S64: performing adaptive hybrid crossover operation on the selected parent individuals to generate child individuals; The crossover operation includes: S641: determining the selection probability of the crossover mode according to the evolution progress; Evolution progress is the current number of completed function evaluations divided by the preset maximum number of function evaluations, or the current number of iteration generations divided by the preset maximum number of iteration generations. A typical value of the preset maximum number of function evaluations is 10000, and a typical value of the maximum number of iteration generations is 200.

[0059] The probability of selecting ordered crossover (OX) is and the probability of selecting partially mapped crossover (PMX) is . Ordered crossover (OX) is a crossover operator that retains the order information of parents, and partially mapped crossover (PMX) is a crossover operator that retains the absolute positions of parents. For example, in the early stage of evolution = 0.2, then OX is used with a probability of 80% and PMX is used with a probability of 20%.

[0060] S642: determining the crossover mode based on the selection probability; Generate a random number r between 0 and 1, if r < 1- , ordered crossover (OX) is selected; otherwise, partially mapped crossover (PMX) is selected.

[0061] S643: selecting an intermediate fragment from the permutation of parent individuals as the crossover region based on the determined crossover mode, and recombining the elements in the crossover region according to the element recombination rule corresponding to the crossover mode to generate a child permutation; For example, the length of each of the two parent permutations is the total number of tokens L. Randomly select two crossover points c1 and c2 that satisfy 0 < c1 < c2 < L, and extract the intermediate fragment. For ordered crossover (OX), the child inherits the fragment of one parent, and then fills the missing tokens in order from the other parent. For partially mapped crossover (PMX), a mapping relationship of elements in the fragment is established, and the corresponding positions are exchanged. For specific implementation, reference can be made to a classic genetic algorithm library, and two child individuals are generated.

[0062] S65: performing a light-weight mutation operation on the generated child individuals to obtain mutated child individuals; Performing the light-weight mutation operation includes: performing a local order adjustment operation on the permutation composed of access tokens and warehouse separation tokens based on a mutation probability that gradually decreases with the evolution progress; Mutation probability is initially set to 0.2, and decreases linearly to 0.01 along with the evolution progress , that is: .

[0063] For each offspring individual, with probability Perform mutation. Mutation methods include inversion mutation and insertion mutation, one of which is randomly selected. Inversion mutation: Randomly select two positions and invert the subsequence between these two positions. Insertion mutation: Randomly select a token, remove it, and then randomly insert it into another position.

[0064] S66: Perform local order fine-tuning on some of the mutated offspring individuals to obtain fine-tuned offspring individuals; Local optimization is performed on offspring individuals with a preset fine-tuning probability of 0.1. Local order fine-tuning includes the 2-opt operation and local order correction. The 2-opt operation is a path optimization local search method that randomly selects two positions and reverses the subsequence between the two points to eliminate path intersections. Local order correction swaps the positions of tokens that violate the sequence constraint within the same task point to make them satisfy the constraint.

[0065] S67: Merge the parent population with the offspring population composed of the finely adjusted offspring individuals, and obtain the next generation population based on the merged population. The parent and offspring populations are merged into a temporary population. Non-dominated sorting and crowding distance calculation are performed on this temporary population, and then the top 100 individuals are selected as the next generation population according to the principle of priority of rank and priority of greater crowding within the same rank.

[0066] The next generation of the population is updated to the current population. S61-S67 are repeated until a preset termination condition is met. All non-dominated solutions in the current population are output as the optimal solution set. The termination condition is typically set to reaching the maximum number of iterations, such as 200 generations, or that the optimal solution has not improved for several consecutive generations. Finally, all individuals with a non-dominated level of 1 in the population constitute a Pareto optimal solution set, where solutions are mutually non-dominated. The initial population size can be 100. 100 solutions are randomly selected from all candidate solutions to construct the initial population, which is the initial current population.

[0067] S7: Select the target solution from the set of optimal solutions as the result of the collaborative task planning.

[0068] Based on actual preferences, such as if the user is more concerned with the shortest total execution path, the solution with the shortest total execution path is selected from the optimal solution set; if the user is more concerned with the maximum completion time, the solution with the shortest completion time is selected; a weighted summation method or user interaction method can also be used to select the solution. After decoding the solution, the specific task sequence and driving route of each unmanned operation body are obtained, and the output is a planning scheme in JSON or XML format.

[0069] Example 2 A collaborative task planning system for heterogeneous unmanned platforms, characterized in that the system comprises: The acquisition module is used to acquire the task point set and warehouse set based on the operation environment map; and to acquire the unmanned operation body type set and the task requirement information corresponding to each type of unmanned operation body according to the preset unmanned operation body configuration information. The encoding module is used to perform mixed-arrangement encoding on the task requirement information and the warehouse set to generate candidate solutions for collaborative task planning of multiple unmanned operations. The decoding module is used to decode the candidate solutions to obtain multiple execution paths for the unmanned operation body; The calculation module is used to calculate the total execution path and maximum completion time of the candidate solution based on the execution paths of multiple unmanned operation bodies. The search module is used to perform a multi-objective evolutionary search on the candidate solutions, taking minimizing the total execution path and minimizing the maximum completion time as dual objective functions, to obtain the optimal solution set. The output module is used to select the target solution from the set of optimal solutions as the result of the collaborative task planning.

[0070] It is understandable that the modules recorded in this heterogeneous unmanned platform collaborative task planning system are similar to those in the reference system. Figure 1 The steps described in the heterogeneous unmanned platform collaborative task planning method correspond to each other. Therefore, the operations, characteristics, and beneficial effects described above for the heterogeneous unmanned platform collaborative task planning method are also applicable to the heterogeneous unmanned platform collaborative task planning system and its included modules, and will not be repeated here.

[0071] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0072] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A collaborative task planning method for heterogeneous unmanned platforms, characterized in that, include: Obtain the task point set and warehouse set based on the work environment map; Based on the preset unmanned operation configuration information, obtain the set of unmanned operation types and the corresponding task requirements information for each type of unmanned operation. The task requirement information and the warehouse set are mixed and encoded to generate candidate solutions for collaborative task planning of multiple unmanned operations. The candidate solutions are decoded to obtain multiple execution paths for the unmanned operation. The total execution path and maximum completion time of the candidate solutions are calculated based on the execution paths of multiple unmanned operating entities. Using minimizing the total execution path and minimizing the maximum completion time as dual objective functions, a multi-objective evolutionary search is performed on the candidate solutions to obtain the optimal solution set; The target solution is selected from the set of optimal solutions as the result of the collaborative task planning.

2. The method according to claim 1, characterized in that, The process of mixing and encoding the task requirement information and the warehouse set to generate candidate solutions for collaborative task planning of multiple unmanned operations includes: A task requirement matrix is ​​constructed based on the task requirement information, wherein the task requirement matrix is ​​arranged with task points as rows and unmanned operation body types as columns, and the matrix elements are used to characterize whether the corresponding type of unmanned operation body needs to perform operations at the corresponding task points. Generate a corresponding access token for each non-zero element in the task requirement matrix; Generate several warehouse partition tokens based on the number of warehouses in the warehouse set; Based on the execution order of the unmanned operation, all access tokens and warehouse separation tokens are arranged to obtain the candidate solutions.

3. The method according to claim 1, characterized in that, The process of decoding the candidate solutions yields multiple execution paths for the unmanned operation, including: Acquire the capacity of various types of unmanned operating systems; The candidate solutions are divided into multiple warehouse segments based on the position of the warehouse separator token in the candidate solutions; Extract task sequences according to the type of unmanned operation within each warehouse section; The task sequence is segmented based on the capacity of each type of unmanned operation body to obtain multiple unmanned operation body execution paths that meet the capacity constraints.

4. The method according to claim 3, characterized in that, The task sequence is segmented based on the capacity of each type of unmanned operator to obtain multiple unmanned operator execution paths that satisfy capacity constraints, including: The demand for each task point is accumulated sequentially along the task sequence; When adding the next task point would cause the cumulative demand to exceed the capacity of the corresponding type of unmanned work body, the currently accumulated task point is taken as an execution path, and the next task point is taken as the starting point of the new execution path. The demand is then accumulated again until all task points in the task sequence are allocated, resulting in multiple unmanned work body execution paths.

5. The method according to claim 1, characterized in that, The method further includes calculating the total execution path, including: Obtain the coordinates of each task point in the task point set and the coordinates of each warehouse in the warehouse set, and generate a distance matrix based on the length of the passable path between each task point and each warehouse. For each unmanned operation body execution path, the travel distance from the warehouse to the warehouse is calculated based on the distance matrix, sequentially visiting each task point on the path and returning to the warehouse. The total execution path is obtained by summing the travel distances of all unmanned operation paths.

6. The method according to claim 1, characterized in that, A multi-objective evolutionary search is performed on the candidate solutions to obtain the optimal solution set, including: Step a: Construct an initial population based on the candidate solutions, and use the initial population as the current population; Step b: Calculate the non-dominance level and crowding distance for each individual in the current population, and select elite individuals from the current population based on the non-dominance level and crowding distance to retain for the next generation; Step c: Select parent individuals from the current population using a tournament selection method to obtain the parent population; Step d: Perform mutation operation on the parent population to obtain the offspring population, merge the parent population and the offspring population to obtain the merged population, and select the next generation population based on the merged population. Update the next generation population to the current population, repeat step bd until the preset termination condition is met, and output all non-dominated solutions in the current population as the optimal solution set.

7. The method according to claim 2, characterized in that, Based on the execution order of the unmanned operation, all access tokens and warehouse separation tokens are arranged to obtain the following candidate solutions: For each task point, based on the sequential relationship between different types of unmanned operating entities within that task point, the corresponding access tokens are randomly sorted to satisfy that sequential relationship, thus obtaining the access token sequence within that task point. The access token sequences within all task points are randomly and alternately extracted to form a global access sequence; Insert a warehouse separator token into the global access sequence to obtain candidate solutions.

8. The method according to claim 6, characterized in that, The offspring population obtained by performing mutation operations on the parent population includes: Perform an adaptive crossover operation on the parent individuals in the parent population to generate offspring individuals; Perform a light mutation operation on the generated offspring individuals to obtain mutated offspring individuals; Perform local order fine-tuning on some of the mutated offspring individuals to obtain fine-tuned offspring individuals; The offspring population is obtained based on the finely adjusted offspring individuals.

9. The method according to claim 8, characterized in that, The adaptive mixing and crossover operation performed on parent individuals in the parent population includes: Determine the probability of choosing the crossover method based on the evolutionary progress; The crossover method is determined based on the selection probability; Based on the determined crossover method, the middle segment is selected as the crossover region in the arrangement of the parent individuals, and the elements in the crossover region are recombined according to the element recombination rules corresponding to the crossover method to generate the offspring individuals.

10. A collaborative task planning system for heterogeneous unmanned platforms, characterized in that, The system includes: The acquisition module is used to acquire the task point set and warehouse set based on the operation environment map; and to acquire the unmanned operation body type set and the task requirement information corresponding to each type of unmanned operation body according to the preset unmanned operation body configuration information. The encoding module is used to perform mixed arrangement encoding on the task requirement information and the warehouse set to generate candidate solutions for collaborative task planning of multiple unmanned operations. The decoding module is used to decode the candidate solutions to obtain multiple execution paths for the unmanned operation body; The calculation module is used to calculate the total execution path and maximum completion time of the candidate solution based on the execution paths of multiple unmanned operation bodies. The search module is used to perform a multi-objective evolutionary search on the candidate solutions, taking minimizing the total execution path and minimizing the maximum completion time as dual objective functions, to obtain the optimal solution set. The output module is used to select the target solution from the set of optimal solutions as the result of the collaborative task planning.

Images