Orchard multi-robot multi-target distribution method based on adaptive task segmentation
By employing a multi-robot task allocation method that combines hybrid encoding and adaptive search mode switching, the challenge of multi-robot collaborative scheduling in task-segmentable scenarios was solved, achieving better multi-robot task allocation results in orchards and improving system collaboration efficiency and solution quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHENGZHOU UNIV
- Filing Date
- 2026-01-16
- Publication Date
- 2026-04-28
AI Technical Summary
Existing technologies fail to effectively handle the divisibility of tasks in multi-robot task allocation, resulting in an excessively large decision space and difficulty in algorithm convergence. It is difficult to achieve a balance between coarse-grained path structure exploration and fine-grained task segmentation optimization, which affects the system's collaborative efficiency and solution quality.
Employing a hybrid coding, adaptive search mode switching, and collaborative task segmentation mechanism, the system dynamically adjusts its search strategy through path structure optimization and multi-objective task segmentation optimization modes. By combining path information coding and task segmentation information coding, it identifies bottleneck robots and optimizes task segmentation.
It achieves better multi-robot collaborative scheduling, improves the solution quality and convergence speed of the algorithm, effectively handles the conflict between energy consumption and time objectives, and significantly improves the efficiency of multi-robot task allocation in orchards.
Smart Images
Figure CN121936840A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent agriculture and robot scheduling technology, specifically involving a multi-robot multi-objective allocation method for orchards based on adaptive task segmentation. Background Technology
[0002] In smart agriculture, multi-robot systems play a crucial role in improving operational efficiency and alleviating labor shortages. However, in practical applications such as orchard harvesting, task allocation faces unique complexities. First, task objectives are often multiple and conflicting, such as pursuing the shortest overall completion time while controlling the lowest total system energy consumption. Second, agricultural tasks (such as harvesting a fruit tree) are physically divisible, allowing multiple robots to collaborate on the same task. This provides greater flexibility for optimization but also significantly expands the dimensions of the decision space.
[0003] Existing technologies have significant shortcomings in addressing such problems. Most multi-robot task allocation methods assume that tasks are indivisible, meaning that a task must be completed by a single robot at a time. This does not align with real-world batch processing scenarios and limits the system's collaborative potential. A few methods that do consider task partitioning often suffer from poor convergence and low-quality solutions due to excessively large search spaces or a lack of efficient collaborative partitioning and allocation mechanisms. Furthermore, how to dynamically adjust the search strategy based on the algorithm's progress at different optimization stages, achieving a balance between coarse-grained path structure exploration and fine-grained task partitioning optimization, remains a challenge that current technologies have not effectively addressed.
[0004] Therefore, there is an urgent need for a multi-robot multi-objective optimization algorithm that can efficiently handle the divisible nature of tasks and adaptively adjust the search focus, in order to solve the bottlenecks of existing technologies in terms of task allocation efficiency and solution quality in complex agricultural scenarios. Summary of the Invention
[0005] To address the aforementioned problems, this invention proposes a multi-robot multi-target allocation method for orchards based on adaptive task segmentation. This method employs hybrid coding, adaptive search mode switching, and a collaborative task segmentation mechanism, effectively solving the challenge of collaborative scheduling of multiple robots in scenarios where tasks can be segmented.
[0006] The technical solution adopted in this invention is: a multi-robot multi-target allocation method for orchards based on adaptive task segmentation, comprising the following steps:
[0007] S1, Initialization Phase: Establish a mathematical model with the goal of minimizing the maximum completion time and total energy consumption, and allow the task to be divided among multiple robots; generate the initial population using a hybrid encoding method;
[0008] S2, Adaptive Iterative Optimization Phase: Before the preset termination condition is met, perform the following operations in a loop: Adaptively select the execution path structure optimization mode or the multi-objective task partitioning optimization mode based on the convergence state of the current population.
[0009] The path structure optimization mode involves: conducting coarse-grained exploration of the current solution, adjusting the overall path topology by exchanging and transferring tasks within or between robots, and simultaneously optimizing the internal task sequence of the affected path.
[0010] The multi-objective task segmentation optimization mode: performs fine-grained deep utilization of non-dominated solutions in the current population, identifies the robot that causes the system bottleneck, and applies at least two task segmentation mechanisms in a coordinated manner to optimize the deviation task points on the path of the bottleneck robot.
[0011] S3, Environment Selection and Output Stage: The new solution generated in S2 is merged with the original population, and the next generation population is selected by non-dominated sorting; when the preset termination condition is reached, the final non-dominated solution set is output.
[0012] Furthermore, in S1, the hybrid encoding includes path information encoding and task segmentation information encoding; the path information encoding is represented by an integer sequence, and two warehouse identifiers with different numbers are used as separators for tasks between robots and for different paths of the same robot, respectively; the task segmentation information encoding uses a list or table to record the segmented task information, and each record includes the task number, the robot number to which it is assigned, the path number, and the proportion of tasks completed on that path.
[0013] Furthermore, in S2, the specific strategy for adaptively selecting the path structure optimization mode or the multi-objective task segmentation optimization mode based on the convergence state of the current population is as follows: Define a probability parameter P, which is dynamically adjusted according to the ratio of the computational resources used by the algorithm to the total computational resources; at the beginning of each iteration, generate a random number. If the random number is less than P, then execute the path structure optimization mode; otherwise, execute the multi-objective task segmentation optimization mode.
[0014] Furthermore, the specific steps of the path structure optimization mode are as follows:
[0015] S211, Generation of Composite Operator Sequence: Based on the complexity of the current solution and the convergence state of the population, using the formula:
[0016] Sequence length L = Average number of paths × (1 - Proportion of non-dominated solutions)
[0017] Calculate the length L of the operator sequence in the composite movement strategy, where the average number of paths is the average number of paths executed by each robot in the current solution; the proportion of non-dominated solutions is the proportion of non-dominated solutions in the current population; randomly select from four basic operators—intra-robot task exchange, intra-robot task transfer, inter-robot task exchange, and inter-robot task transfer—to form an operator sequence of length L.
[0018] S212, Composite operator execution: Each operator in the operator sequence is applied to the current solution in sequence, modifying the total task sequence under one or two robots, without re-dividing the specific physical path for the time being;
[0019] S213, Path Reconstruction and Update: For each robot with a modified task sequence, a path reconstruction process is executed to ensure load feasibility. The path reconstruction process is as follows: Initialize an empty path starting from the warehouse with a current load of zero; sequentially traverse the robot's new task sequence; determine whether adding a task to the current path exceeds the robot's maximum load capacity Q; if not overloaded, add the task to the current path and accumulate the load; if overloaded, close the current path and store it in the robot's new path set, and simultaneously open a new path starting from the overloaded task; repeat the above traversal and judgment operations until all tasks under the robot are assigned to feasible paths; after path reconstruction is completed, if multiple segments of the same original task are assigned to the same new path, merge the segments and update the total completion ratio of the task in the task segmentation information encoding.
[0020] S214, Path Internal Sequence Optimization: For the reconstructed or modified path in S213, the 2-opt optimization method is used to reorder the task points within the path to shorten the robot's walking distance, thereby optimizing energy consumption and completion time.
[0021] Furthermore, the specific steps of the multi-objective task segmentation optimization mode are as follows:
[0022] S221, Bottleneck robot identification: In the current solution, the robot with the longest completion time is identified as the bottleneck robot R_max;
[0023] S222, Deviation Task Point Identification: For each path of the bottleneck robot R_max, the Deviation Task Point Identification (DTPIM) mechanism is used to select the task point with the highest geometric deviation as a candidate task to be segmented.
[0024] S223, Cooperative task partitioning: For each candidate biased task point, execute the proximity-based task partitioning mechanism (PTSM) and the balance-based task partitioning mechanism (BTSM) in sequence to generate a new solution; if the new solution constitutes a Pareto improvement over the original solution, then accept the new solution and iterate the above partitioning operation until a Pareto improvement can no longer be achieved.
[0025] Furthermore, the execution steps of the DTPIM are as follows:
[0026] S2211, Calculate the centroid of the path: Calculate the average value of the coordinates of all task points on the path to obtain the geometric centroid of the path;
[0027] S2212, Calculate the deviation: For each task point on the path, calculate the comprehensive deviation based on its own position, the centroid position of the path, and the warehouse position. The comprehensive deviation is the product of "the sine of the angle between the vector from the task point to the centroid and the vector to the warehouse" and "the ratio of the distance from the task point to the centroid to the distance to the warehouse".
[0028] S2213, Identify Deviation Points: Determine the task point with the largest overall deviation as the deviation task point of the path.
[0029] Furthermore, the specific steps of the PTSM are as follows:
[0030] S2231, Determine the proximity threshold: Calculate the minimum distance between the task point to be segmented and its neighboring task points in its original path, and use this minimum distance as the proximity distance reference threshold ref_dist;
[0031] S2232, Filter candidate receiving paths: Traverse all paths of all robots except the current robot, filter out paths where "the distance from the task point to the task point to be segmented is less than ref_dist", and use them as candidate receiving paths;
[0032] S2233, Calculate the segmentation ratio and insertion: For each candidate receiving path, calculate the optimal task receiving ratio α based on its available capacity and time margin; find the best insertion position on the candidate receiving path that minimizes the robot's additional travel increment, insert the task with the corresponding ratio α into the position, and simultaneously update the path information encoding and task segmentation information encoding.
[0033] Furthermore, the specific steps of the BTSM are as follows:
[0034] S2234, Identify the idlest robot: Among all robots, the robot with the shortest completion time is identified as the idlest robot R_min;
[0035] S2235, Execute the splitting strategy: Choose the optimal strategy from the following two options:
[0036] Strategy 1, Independent Path Construction: Construct a completely new path for the idlest robot R_min, which only includes a portion of the tasks transferred from the bottleneck robot R_max; the proportion of transferred tasks α is determined according to the principle of "minimizing the difference in completion time between R_max and R_min after the segmentation";
[0037] Strategy 2, existing path integration: Among all existing paths of the idlest robot R_min, select the path with the smallest geometric deviation from the task to be segmented as the receiving path; use the same logic as PTSM to calculate the optimal task reception ratio α and the optimal insertion position of this receiving path;
[0038] S2236, Preserve the optimal solution: Accept the new solution only if the new solution generated by the partitioning operation can Pareto dominate the original solution.
[0039] Furthermore, in S211, the specific execution steps of the four basic operators are as follows:
[0040] The steps involved in the intra-robot task exchange are as follows:
[0041] (1) Randomly select a robot;
[0042] (2) Randomly select two different paths from all the paths that the robot is responsible for;
[0043] (3) Randomly select one task point from each of the two paths mentioned above, and swap the paths to which these two task points belong;
[0044] The steps involved in the intra-robot task transfer are as follows:
[0045] (1) Randomly select a robot;
[0046] (2) From all the paths that the robot is responsible for, randomly select two different paths as the source path and the target path, respectively;
[0047] (3) Randomly select a task point from the source path and move it to a random location in the target path;
[0048] The steps involved in the inter-robot task exchange are as follows:
[0049] (1) Randomly select two different robots;
[0050] (2) Randomly select one path from all the paths each robot is responsible for;
[0051] (3) Randomly select one task point from each of the two selected paths and exchange these two task points between the corresponding paths of the two robots.
[0052] The steps involved in the inter-robot task transfer are as follows:
[0053] (1) Randomly select two different robots as the source robot and the target robot, respectively;
[0054] (2) Randomly select a path from all the paths managed by the source robot and randomly select a task point from that path; at the same time, randomly select a path from all the paths managed by the target robot and determine a random insertion position in that path;
[0055] (3) Remove the task point selected from the source robot from the corresponding path of the source robot and transfer it to the selected path and specified insertion position of the target robot.
[0056] The beneficial effects of this invention are:
[0057] 1. For the first time, a multi-robot multi-objective optimization model that allows for task divisibility is proposed for agricultural scenarios, which is closer to real-world needs and can obtain better solutions that cannot be obtained by traditional methods through task division and collaboration;
[0058] 2. An adaptive search framework is adopted, which can dynamically adjust the search focus according to the optimization process. It conducts extensive structural exploration in the early stage of the algorithm and in-depth convergence mining in the later stage, effectively balancing the algorithm's global exploration and local utilization capabilities.
[0059] 3. The hybrid coding method of the design clearly decouples path allocation and task segmentation information, which facilitates efficient local operations and information updates;
[0060] 4. The innovative multi-objective task partitioning and optimization module, through PTSM and BTSM, can accurately optimize bottleneck tasks, effectively handle the conflict between energy consumption and time objectives, and significantly improve solution quality and convergence speed. Attached Figure Description
[0061] Figure 1 This is a flowchart illustrating the overall framework of the present invention;
[0062] Figure 2 This is a schematic diagram of the hybrid encoding structure of the solution in this invention; Detailed Implementation
[0063] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort. To facilitate understanding of the present invention, the present invention will be described in more detail below with reference to the accompanying drawings and specific embodiments.
[0064] like Figure 1 As shown, this invention is a multi-robot multi-objective allocation method for orchards based on adaptive task segmentation, comprising the following steps:
[0065] S1, Initialization Phase: Establish a mathematical model with the objective of minimizing the maximum completion time and total energy consumption, while allowing the task to be divided among multiple robots; generate an initial population using a hybrid encoding method, which includes path information encoding and task division information encoding. For example... Figure 2 As shown, the path information encoding uses an integer sequence, where two different numbers representing the warehouse (such as "0" and "1") are used as separators for tasks between robots and for different paths of the same robot, respectively; the task segmentation information encoding uses a list or table to record the segmented task information, and each record includes the task number, the robot number to which it is assigned, the path number, and the proportion of tasks completed on that path.
[0066] S2, Adaptive Iterative Optimization Phase: Before the termination condition is met, the following operations are performed iteratively: Based on the convergence state of the current population, adaptively select either the path structure optimization mode or the multi-objective task partitioning optimization mode. The path structure optimization mode performs coarse-grained exploration of the current solution, adjusting the overall path topology by exchanging and transferring tasks within or between robots, and simultaneously optimizing the internal task sequences of affected paths. The multi-objective task partitioning optimization mode performs fine-grained deep utilization of non-dominated solutions in the current population, identifying robots causing system bottlenecks, and collaboratively applying both task partitioning mechanisms to optimize the deviation task points on the bottleneck robot's path.
[0067] The adaptive selection strategy described above is as follows: Define a probability parameter P, which is dynamically adjusted based on the ratio of the algorithm's used computational resources to the total computational resources. At the beginning of each iteration, a random number is generated. If the random number is less than P, the path structure optimization mode is executed; otherwise, the multi-objective task partitioning optimization mode is executed. The design of the probability parameter P makes the algorithm more inclined to execute the path structure optimization mode for extensive exploration in the early stages, and more inclined to execute the multi-objective task partitioning optimization mode to accelerate convergence in the later stages.
[0068] The specific steps of the above path structure optimization mode are as follows:
[0069] S211, Generation of Composite Operator Sequence: Based on the complexity of the current solution and the convergence state of the population, dynamically calculate the length of the operator sequence in a composite move strategy, denoted as the sequence length L. This length is calculated using the following formula:
[0070] Sequence length L = Average number of paths × (1 - Proportion of non-dominated solutions)
[0071] The average number of paths is the average number of paths executed by each robot in the current solution, which reflects the structural complexity of the solution; the proportion of non-dominated solutions is the proportion of non-dominated solutions in the current population, which reflects the overall convergence quality of the population.
[0072] This formula generates a longer sequence of operators for a wider neighborhood search when the solution structure is complex (high average path value) or the population quality is poor (low proportion of non-dominated solutions); conversely, it generates a shorter sequence for a more refined local search.
[0073] Subsequently, a sequence of operators of length L is formed by randomly selecting from the following four basic operators:
[0074] (1) Intra-robot task exchange: This operator aims to optimize the task allocation within a single robot; the specific operation is as follows: First, randomly select a robot; then, randomly select two different paths from all the paths that the robot is responsible for; finally, randomly select a task point from each of these two paths for exchange.
[0075] (2) Intra-robot task transfer: This operator aims to adjust the task execution order and path composition within a single robot; the specific operation is as follows: First, randomly select a robot; then, randomly select two different paths from all the paths that the robot is responsible for, as the source path and the target path respectively; finally, randomly select a task point from the source path and move it to a random position on the target path.
[0076] (3) Inter-robot task exchange: This operator aims to optimize the task load balancing among different robots. The specific operation is as follows: First, randomly select two different robots; then, randomly select one path from the paths each of the two robots is responsible for; finally, randomly select one task point from each of the two selected paths for exchange.
[0077] (4) Inter-robot task transfer: This operator aims to achieve more flexible inter-robot task redistribution; the specific operation is as follows: First, randomly select two different robots as the source robot and the target robot respectively; then, randomly select a path and a task point from the source robot, and randomly select a path and an insertion position from the target robot; finally, move the selected task point from the source robot to the specified path and position of the target robot.
[0078] S212, Composite Operator Execution: Each operator in the operator sequence generated in S211 is applied sequentially to the current solution, modifying the overall task sequence under one or two robots. This process only changes the task order and attribution; it does not re-divide the specific physical paths.
[0079] S213, Path Reconstruction and Update: After the entire operator sequence has been executed, for each robot whose overall task sequence has been modified, perform the following path reconstruction process to ensure load feasibility:
[0080] (1) Initialize an empty path starting from the warehouse, with the current load being zero;
[0081] (2) Iterate through the new total sequence of tasks for the robot in sequence. For each task, determine whether adding it to the current path will cause the total load of the path to exceed the robot's maximum load capacity Q;
[0082] (3) If not overloaded, add the task to the current path and add its load;
[0083] (4) If overloaded, close the current path (i.e. add a node to return to the warehouse), store it in the robot's new path set, and start a new path with the task that caused the overload as the starting point;
[0084] (5) Repeat steps (2) to (4) until all tasks under the robot are assigned to feasible paths;
[0085] (6) After all paths are reconstructed, check each new path. If multiple segments of the same original task are assigned to the same new path, merge these segments and update their total proportion in the task segmentation information encoding.
[0086] S214, Path Internal Sequence Optimization: For all paths that were reconstructed or modified in S213, the 2-opt optimization method is used to adjust the task sequence within the path, that is, to reorder the task points within the path in order to find a shorter travel distance, thereby optimizing its energy consumption and time.
[0087] The specific steps of the multi-objective task segmentation optimization mode are as follows:
[0088] S221, Bottleneck Robot Identification: In the current solution, find the robot with the longest completion time, denoted as the bottleneck robot R_max, and use it as the optimization target.
[0089] S222, Deviation Task Point Identification: For each path of the bottleneck robot R_max, perform DTPIM to find the task point with the highest geometric deviation in the path, which is then used as a candidate task to be segmented; the specific steps of DTPIM are as follows:
[0090] (1) Calculate the centroid of the path: take the average of the coordinates of all task points on a path to obtain the geometric centroid of the path;
[0091] (2) Calculate the deviation: For each task point on the path, calculate a comprehensive deviation Dev based on its own position, the centroid position of the path, and the warehouse position. The deviation is composed of the sine of the angle between the vector from the task point to the centroid and the vector to the warehouse, and the product of the ratio of the distance from the task point to the centroid to the distance to the warehouse.
[0092] (3) Identify deviation points: In a path, the task point with the largest deviation value is identified as the deviation task point of that path.
[0093] S223, Cooperative task splitting: For each candidate biased task point, PTSM and BTSM are executed sequentially to generate a new solution; if the new solution constitutes a Pareto improvement, the new solution is accepted and the process is iterated until no improvement can be made.
[0094] The specific steps of PTSM in S223 are as follows:
[0095] (1) Determine the proximity threshold: Calculate the minimum distance between the task point to be segmented and its neighboring task points in its original path, and use it as the proximity distance reference threshold ref_dist;
[0096] (2) Filter candidate receiving paths: Traverse all paths of all other robots except the current robot, and look for paths where the distance from the task point to the task point to be segmented is less than the reference threshold. These paths are selected as candidate receiving paths.
[0097] (3) Calculate the segmentation ratio and insertion: For each candidate receiving path, calculate an optimal task receiving ratio α based on its available capacity and time margin; then, find the best insertion position on the path that minimizes the extra travel increment, insert the corresponding ratio of tasks, and update the path information and segmentation information synchronously.
[0098] The specific steps of BTSM in S223 are as follows:
[0099] (1) Identify the idlest robot: Among all robots, find the robot with the shortest completion time and denot it as the idlest robot R_min;
[0100] (2) Execute the segmentation strategy: Choose the optimal strategy from the following two options:
[0101] Strategy 1, Independent Path Construction: Construct a completely new path for the idlest robot R_min, which only includes the part of the task transferred from the bottleneck robot R_max; the proportion of transferred tasks α is determined according to the principle of minimizing the difference in completion time between the two robots after the split.
[0102] Strategy 2, existing path integration: Among all existing paths of the idlest robot R_min, select the path with the smallest geometric deviation from the task to be segmented as the receiving path, and use logic similar to PTSM to calculate the segmentation ratio and the optimal insertion point.
[0103] (3) Retain the optimal solution: Only accept the new solution generated by the partitioning operation if it can Pareto dominate the original solution.
[0104] S3, Environment Selection and Output: Merge the new solution generated in S2 with the original population, and select the next generation population through non-dominated sorting; when the termination condition is met, output the final non-dominated solution set.
[0105] To further illustrate the superiority of this invention in solving multi-objective multi-robot task allocation, Table 1 presents the results obtained by this invention and some excellent algorithms for solving multi-robot task allocation, namely Multi-objective Discrete ArtificialBee Colony (MODABC), Reconstructed MOEA / D (RMOEA / D), and Non-dominated SortingGenetic Algorithm-Ⅱ (NSGA-Ⅱ), on the generated orchard multi-robot task allocation test set.
[0106] Table 1 Comparison of the optimal number of times the dataset experiments were conducted
[0107] Methods\Test Set 50×50 60×60 70×70 MODABC 0 1 0 RMOEA / D 0 0 0 NSGA-II 0 0 0 This invention 10 9 10
[0108] The examples present experimental results on a generated test set of multi-objective, multi-robot task allocation in an orchard. These problems vary in difficulty, with greenhouse sizes ranging from 50×50 to 70×70 square meters, containing 320, 500, and 720 fruit trees (task loads), respectively. The fruit tree maturity rate is set to 0.8 for each scenario, and the number of available robots is uniformly 4. Therefore, each combination of fruit tree number and maturity represents a different test problem to evaluate the algorithm's task allocation performance under different work scenarios and task difficulties. The experiment termination condition is set to a maximum CPU runtime of 0.5 × task load, and a population size of 30. To avoid the influence of randomness on the experimental results, each algorithm is run 10 times for each test problem. Hypervolume is used as the evaluation metric. Table 1 shows the number of problems in which each algorithm achieves optimal results in each test problem scenario. The comparison shows that the proposed method demonstrates superior optimization performance in solving multi-objective, multi-robot task allocation with power constraints. Specifically, regardless of the test problem scenario, the number of times the optimal result is obtained is greater than the number of times other algorithms obtain the optimal result.
[0109] In summary, this invention effectively addresses multi-objective, multi-robot task allocation problems, including those with separable tasks, providing decision-makers with a range of ideal compromise solutions. By adaptively switching between coarse-grained path structure optimization and fine-grained task partitioning optimization, this invention effectively manages the allocation of computational resources while ensuring solution quality, offering valuable insights for solving other complex combinatorial optimization problems.
[0110] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.
Claims
1. A method for multi-robot multi-objective allocation in orchards based on adaptive task segmentation, characterized in that, Includes the following steps: S1, Initialization Phase: Establish a mathematical model with the goal of minimizing the maximum completion time and total energy consumption, and allow the task to be divided among multiple robots; generate the initial population using a hybrid encoding method; S2, Adaptive Iterative Optimization Phase: Before the preset termination condition is met, perform the following operations in a loop: Adaptively select the execution path structure optimization mode or the multi-objective task partitioning optimization mode based on the convergence state of the current population. The path structure optimization mode involves: conducting coarse-grained exploration of the current solution, adjusting the overall path topology by exchanging and transferring tasks within or between robots, and simultaneously optimizing the internal task sequence of the affected path. The multi-objective task segmentation optimization mode: performs fine-grained deep utilization of the non-dominated solutions in the current population, identifies the robots that cause system bottlenecks, and optimizes the deviation task points on the path of the bottleneck robot by coordinating the application of at least two task segmentation mechanisms. S3, Environment Selection and Output Stage: The new solution generated in S2 is merged with the original population, and the next generation population is selected by non-dominated sorting; when the preset termination condition is reached, the final non-dominated solution set is output.
2. The orchard multi-robot multi-objective allocation method based on adaptive task segmentation according to claim 1, characterized in that, In S1, the hybrid encoding includes path information encoding and task segmentation information encoding; the path information encoding is represented by an integer sequence, and two warehouse identifiers with different numbers are used as the separators between tasks between robots and between different paths of the same robot, respectively; the task segmentation information encoding uses a list or table to record the segmented task information, and each record includes the task number, the robot number to which it is assigned, the path number, and the proportion of tasks completed on that path.
3. The orchard multi-robot multi-objective allocation method based on adaptive task segmentation according to claim 1, characterized in that, In S2, the specific strategy for adaptively selecting the execution path structure optimization mode or the multi-objective task segmentation optimization mode based on the convergence state of the current population is as follows: define a probability parameter P, which is dynamically adjusted according to the ratio of the computational resources used by the algorithm to the total computational resources. At the start of each iteration, a random number is generated. If the random number is less than P, the path structure optimization mode is executed; otherwise, the multi-objective task partitioning optimization mode is executed.
4. The orchard multi-robot multi-objective allocation method based on adaptive task segmentation according to claim 1, characterized in that, The specific steps of the path structure optimization mode are as follows: S211, Generation of Composite Operator Sequence: Based on the complexity of the current solution and the convergence state of the population, using the formula: Sequence length L = Average number of paths × (1 - Proportion of non-dominated solutions) Calculate the length L of the operator sequence in the composite movement strategy, where the average number of paths is the average number of paths executed by each robot in the current solution; the proportion of non-dominated solutions is the proportion of non-dominated solutions in the current population; randomly select from four basic operators—intra-robot task exchange, intra-robot task transfer, inter-robot task exchange, and inter-robot task transfer—to form an operator sequence of length L. S212, Composite operator execution: Each operator in the operator sequence is applied to the current solution in sequence, modifying the total task sequence under one or two robots, without re-dividing the specific physical path for the time being; S213, Path Reconstruction and Update: For each robot with a modified task sequence, a path reconstruction process is executed to ensure load feasibility. The path reconstruction process is as follows: Initialize an empty path starting from the warehouse with a current load of zero; sequentially traverse the robot's new task sequence; determine whether adding a task to the current path exceeds the robot's maximum load capacity Q; if not overloaded, add the task to the current path and accumulate the load; if overloaded, close the current path and store it in the robot's new path set, and simultaneously open a new path starting from the overloaded task; repeat the above traversal and judgment operations until all tasks under the robot are assigned to feasible paths; after path reconstruction is completed, if multiple segments of the same original task are assigned to the same new path, merge the segments and update the total completion ratio of the task in the task segmentation information encoding. S214, Path Internal Sequence Optimization: For the reconstructed or modified path in S213, the 2-opt optimization method is used to reorder the task points within the path to shorten the robot's walking distance, thereby optimizing energy consumption and completion time.
5. A method for multi-robot multi-objective allocation in orchards based on adaptive task segmentation according to claim 1, characterized in that, The specific steps of the multi-objective task segmentation optimization mode are as follows: S221, Bottleneck robot identification: In the current solution, the robot with the longest completion time is identified as the bottleneck robot R_max; S222, Deviation Task Point Identification: For each path of the bottleneck robot R_max, the deviation task point identification mechanism is used to select the task point with the highest geometric deviation as the candidate task to be segmented. S223, Cooperative task partitioning: For each candidate biased task point, execute the task partitioning mechanism based on proximity and the task partitioning mechanism based on equilibrium in sequence to generate a new solution; if the new solution constitutes a Pareto improvement over the original solution, then accept the new solution and iterate the above partitioning operation until a Pareto improvement can no longer be achieved.
6. A method for multi-robot multi-objective allocation in orchards based on adaptive task segmentation according to claim 5, characterized in that, The execution steps of the deviation task point identification mechanism are as follows: S2211, Calculate the centroid of the path: Calculate the average value of the coordinates of all task points on the path to obtain the geometric centroid of the path; S2212, Calculate the deviation: For each task point on the path, calculate the comprehensive deviation based on its own position, the centroid position of the path, and the warehouse position. The comprehensive deviation is the product of "the sine of the angle between the vector from the task point to the centroid and the vector to the warehouse" and "the ratio of the distance from the task point to the centroid to the distance to the warehouse". S2213, Identify Deviation Points: Determine the task point with the largest overall deviation as the deviation task point of the path.
7. A method for multi-robot multi-objective allocation in orchards based on adaptive task segmentation according to claim 5, characterized in that, The specific steps of the proximity-based task segmentation mechanism are as follows: S2231, Determine the proximity threshold: Calculate the minimum distance between the task point to be segmented and its neighboring task points in its original path, and use this minimum distance as the proximity distance reference threshold ref_dist; S2232, Filter candidate receiving paths: Traverse all paths of all robots except the current robot, filter out paths where "the distance from the task point to the task point to be segmented is less than ref_dist", and use them as candidate receiving paths; S2233, Calculate the segmentation ratio and insertion: For each candidate receiving path, calculate the optimal task receiving ratio α based on its available capacity and time margin; find the best insertion position on the candidate receiving path that minimizes the robot's additional travel increment, insert the task with the corresponding ratio α into the position, and simultaneously update the path information encoding and task segmentation information encoding.
8. A method for multi-robot multi-objective allocation in orchards based on adaptive task segmentation according to claim 5, characterized in that, The specific steps of the task partitioning mechanism based on balance are as follows: S2234, Identify the idlest robot: Among all robots, the robot with the shortest completion time is identified as the idlest robot R_min; S2235, Execute the splitting strategy: Choose the optimal strategy from the following two options: Strategy 1, Independent Path Construction: Construct a completely new path for the idlest robot R_min, which only includes a portion of the tasks transferred from the bottleneck robot R_max; the proportion of transferred tasks α is determined according to the principle of "minimizing the difference in completion time between R_max and R_min after the segmentation"; Strategy 2, existing path integration: Among all existing paths of the idlest robot R_min, select the path with the smallest geometric deviation from the task to be segmented as the receiving path; use the same logic as the proximity-based task segmentation mechanism to calculate the optimal task reception ratio α and the optimal insertion position of this receiving path. S2236, Preserve the optimal solution: Accept the new solution only if the new solution generated by the partitioning operation can Pareto dominate the original solution.
9. A method for multi-robot multi-objective allocation in orchards based on adaptive task segmentation according to claim 4, characterized in that, In S211: The steps involved in the intra-robot task exchange are as follows: (1) Randomly select a robot; (2) Randomly select two different paths from all the paths that the robot is responsible for; (3) Randomly select one task point from each of the two paths mentioned above, and swap the paths to which these two task points belong; The steps involved in the intra-robot task transfer are as follows: (1) Randomly select a robot; (2) From all the paths that the robot is responsible for, randomly select two different paths as the source path and the target path, respectively; (3) Randomly select a task point from the source path and move it to a random location in the target path; The steps involved in the inter-robot task exchange are as follows: (1) Randomly select two different robots; (2) Randomly select one path from all the paths each robot is responsible for; (3) Randomly select one task point from each of the two selected paths and exchange these two task points between the corresponding paths of the two robots. The steps involved in the inter-robot task transfer are as follows: (1) Randomly select two different robots as the source robot and the target robot, respectively; (2) Randomly select a path from all the paths managed by the source robot and randomly select a task point from that path; at the same time, randomly select a path from all the paths managed by the target robot and determine a random insertion position in that path; (3) Remove the task point selected from the source robot from the corresponding path of the source robot and transfer it to the selected path and specified insertion position of the target robot.