Orchard multi-robot task allocation optimization method based on hierarchical path reconstruction

Abstract

The invention discloses an orchard multi-robot task allocation optimization method based on hierarchical path reconstruction, and the method comprises the steps: S1, building a multi-target mathematical model, optimizing the maximum completion time and total energy consumption, including robot load, speed, energy consumption and battery capacity constraints, and employing VLDIM to generate an initial solution population; s2, optimizing the task sequence of each path by using DRRM in the microscopic layer; s3, robot-level DRRM optimization, TRRM task redistribution, CRRM charging path reconstruction and environment selection are executed iteratively; and S4, splitting the longest path by using the SRRM, redistributing the longest path through the MILP1, and outputting a non-dominated solution set. Through hierarchical coding, multi-stage collaborative optimization and a special reconstruction mechanism for core constraints, the task allocation efficiency and scheme quality of a multi-robot system under complex constraints are significantly improved, and the method is suitable for agricultural automation and warehouse logistics scenes.

Description

Technical Field

[0001] This invention belongs to the field of multi-robot collaboration and path planning technology, specifically involving an orchard multi-robot task allocation optimization method based on hierarchical path reconstruction. Background Technology

[0002] In recent years, with rising labor costs, multi-robot systems have been increasingly widely used in agriculture, logistics, and other fields. However, how to efficiently coordinate robot systems to cope with complex real-world constraints remains a key research focus and challenge. In typical scenarios such as orchard harvesting, task allocation not only needs to consider multiple conflicting objectives, such as minimizing maximum completion time and total energy consumption, but also must address a series of practical constraints, such as the dynamic impact of robot load on travel speed and limited battery capacity.

[0003] Existing research on multi-robot task allocation often simplifies or ignores the aforementioned key constraints. For example, some methods fail to fully consider the dynamic coupling relationship between load, speed, energy consumption, and battery management, resulting in solutions with low feasibility in practical applications. Furthermore, traditional research mostly uses flattened linear encoding to represent task sequences, which limits the ability to perform fine-grained optimization on individual robots or paths, making efficient searching in complex combinatorial spaces difficult. Simultaneously, there is a lack of dedicated optimization mechanisms to address path structure disruptions caused by battery charging or swapping.

[0004] Therefore, there is an urgent need for a multi-robot task allocation method that can systematically integrate multiple real-world constraints and adopt more flexible and efficient optimization strategies to address the shortcomings of existing technologies in problem modeling and solution efficiency. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a hierarchical path reconstruction optimization method for multi-objective task allocation. This method significantly improves the task allocation efficiency and solution quality of multi-robot systems under complex constraints through hierarchical coding, multi-stage collaborative optimization, and a dedicated reconstruction mechanism for core constraints.

[0006] The technical solution adopted in this invention is: a multi-robot task allocation optimization method for orchards based on hierarchical path reconstruction, comprising the following steps:

[0007] S1, Problem Modeling and Initialization: Establish a multi-objective mathematical model that simultaneously optimizes the maximum completion time and total energy consumption. The model includes dynamic constraints on robot load, speed, energy consumption, and battery capacity. A two-stage initialization mechanism with variable load constraints (VLDIM) is used to generate an initial solution population, and each solution adopts a hierarchical encoding structure.

[0008] The VLDIM includes the following steps:

[0009] S11, Path construction stage: An independent task path is constructed using a distance-based greedy strategy, while a linearly varying load limit strategy is introduced to make the upper limit of robot load different for different solutions in the population.

[0010] S12, Path-Robot Allocation Phase: If the number of constructed paths is greater than or equal to the number of robots, the first mixed integer linear programming model (MILP1) is used for allocation to minimize the maximum completion time;

[0011] The MILP1 defines a binary decision variable zᵢⱼ; if zᵢⱼ=1, it means that the i-th prefabricated path is assigned to the j-th robot, otherwise it is 0; the objective function is to minimize the maximum completion time C. max The constraints include path allocation constraints, completion time calculation constraints (the robot's completion time is the sum of the execution times of all its allocated paths), and maximum completion time definition constraints.

[0012] The path allocation constraint is that each pre-built path is assigned to only one robot; the completion time calculation constraint is that the robot's completion time is the sum of the execution times of all its assigned paths; the maximum completion time definition constraint is C. max ≥ Completion time of any robot;

[0013] If the number of constructed paths is less than the number of robots, iteratively split the longest path until the number of paths equals the number of robots, and then perform balanced distribution.

[0014] The iterative splitting is performed in a loop: the path with the longest execution time in the current path set is identified and selected, and an iterative greedy strategy is used to split it into two sub-paths with approximately equal execution times. That is, task nodes are transferred one by one from one end of the longest path to the new sub-path, and the execution time difference between the two sub-paths is calculated in real time to find the splitting point that minimizes the time difference.

[0015] The balanced allocation is to distribute all paths after the split to each robot;

[0016] S2, Initial Path Optimization: For each initial solution generated in S1, a distance-based path reordering mechanism (DRRM) is applied at the micro-path layer to optimize the execution order of each independent task path; the specific steps of the DRRM are as follows:

[0017] S21, Distance Reordering: Calculate the distance from each task point within the path to the warehouse, and reorder the task execution order in descending order of distance;

[0018] S22, 2-opt local search: Based on the reordered sequence, explore neighborhood solutions by exchanging two disjoint edges using the 2-opt operator;

[0019] S3, Iterative Optimization: Repeat the following steps before the preset termination condition is met:

[0020] S31, Robot-level Path Sequence Optimization: Apply DRRM at the macro-scheduling layer to optimize the complete task sequence assigned to each robot;

[0021] S32, Inter-robot task redistribution: Applying a task-based path reallocation mechanism (TRRM) to optimize global task load balancing;

[0022] S33, Charge-based path reconstruction: The charge-based path reconstruction mechanism (CRRM) is performed on the non-dominated solutions in the current population. This mechanism extracts the task sequence after the last charging operation of each robot, performs local optimization, and then redistributes the task to reduce the interference of charging behavior on the path structure.

[0023] S34, Environment Selection: The optimized new solution is merged with the original population, and the next generation population is selected by non-dominated sorting and crowding calculation.

[0024] S4, Final Refinement and Output: Execute a split-based path reconstruction mechanism, which includes:

[0025] S41, Identify the longest path: Identify the path with the longest execution time among all independent task paths;

[0026] S42, Path Splitting: The longest path is split into two sub-paths with approximately equal execution times using an iterative greedy strategy;

[0027] S43, Global Reassignment: Merge the split sub-paths with other unchanged paths, apply MILP1 to redistribute all paths to all robots; the final output is the optimized non-dominated solution set.

[0028] Furthermore, the hierarchical coding structure includes a macro-scheduling layer and a micro-path layer. The macro-scheduling layer encodes the complete task allocation scheme into a multi-dimensional structure that includes a global task sequence, a task mapping sequence for each robot, a performance index set, and a charging position record. The global task sequence distinguishes different robots and different paths through a specific separator. The micro-path layer encodes each independent task path into a triple that includes a task sequence, execution time, and energy consumption.

[0029] Furthermore, in S32, TRRM performs any of the following operations with a preset probability:

[0030] Operation 1, Task Exchange: Randomly select two robots and exchange any task node in their task sequences;

[0031] Operation 2, Task Reassignment: Randomly select two robots and move any task node of the robot with the longer completion time to the robot with the shorter completion time.

[0032] Furthermore, the preset probability is an equal probability.

[0033] Furthermore, in S33, the specific steps of the CRRM are as follows:

[0034] S331, Task Extraction: Extract all tasks after the last charging operation of each robot to form a set of tasks to be optimized;

[0035] S332, Sequence optimization: Apply DRRM optimization to the set of tasks to be optimized;

[0036] S333, Task Reassignment: The optimized task sequence is reassigned to all robots using the second mixed integer linear programming model (MILP2), with the goal of minimizing the maximum completion time after introducing new tasks;

[0037] Furthermore, in S333, the specific definition of MILP2 is as follows:

[0038] Decision variables: binary variable zᵢⱼ indicates whether the i-th task to be assigned is assigned to the j-th robot, and binary variable wⱼ indicates whether the j-th robot accepts the new task;

[0039] Objective function: Minimize the maximum completion time C max ;

[0040] Constraints include task allocation integrity constraints, robot state update constraints, and maximum completion time calculation and limitation constraints. The task allocation integrity constraint assigns each task to one robot. The robot state update constraint activates when the robot receives at least one new task. The maximum completion time calculation and limitation constraint is that the robot's new completion time = initial task time before charging + execution time of the newly assigned task + possible battery swapping penalty time, and the new completion time ≤ C. max .

[0041] The beneficial effects of this invention are:

[0042] 1. By using a hierarchical coding structure, the complex task allocation problem is decoupled into two sub-problems: path construction and path allocation. This reduces the difficulty of optimization and supports targeted optimization for different levels.

[0043] 2. The variable load constraint initialization method effectively improves the quality and diversity of the initial population, laying a good foundation for subsequent global search.

[0044] 3. It integrates multiple specialized local optimization and reconstruction mechanisms (DRRM, TRRM, CRRM, SRRM), which can collaboratively and specifically handle core constraints and objectives such as task sequences, inter-robot load balancing, battery charging, and maximum completion time, significantly improving the overall performance of the algorithm.

[0045] 4. The method demonstrates superior performance and robustness on multiple test instances of varying scales and complexities, providing high-quality decision-making solutions for practical applications such as agricultural automation and intelligent logistics. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the overall framework of the present invention;

[0047] Figure 2 This is a schematic diagram of the layered coding structure of the present invention;

[0048] Figure 3 This is a schematic diagram of the 2-opt operation used in the DRRM mechanism of this invention. Detailed Implementation

[0049] 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.

[0050] like Figure 1 As shown, this invention is a multi-robot task allocation optimization method for orchards based on hierarchical path reconstruction, comprising the following steps:

[0051] S1, Problem Modeling and Initialization: Establish a multi-objective mathematical model that simultaneously optimizes the maximum completion time and total energy consumption. This model includes dynamic constraints on robot load, speed, energy consumption, and battery capacity. A two-stage initialization mechanism with variable load constraints is used to generate an initial solution population, and each solution adopts a hierarchical coding structure.

[0052] like Figure 2 As shown, the hierarchical coding structure includes a macro-scheduling layer and a micro-path layer. The macro-scheduling layer encodes the complete task allocation scheme into a multi-dimensional structure that includes a global task sequence, a task mapping sequence for each robot, a performance index set, and a charging position record. The global task sequence distinguishes different robots and different paths through a specific separator. The micro-path layer encodes each independent task path into a triplet that includes a task sequence, execution time, and energy consumption.

[0053] The aforementioned two-stage initialization mechanism for variable load limits includes the following steps:

[0054] S11, Path construction stage: An independent task path is constructed using a distance-based greedy strategy, while a linearly varying load limit strategy is introduced to make the upper limit of robot load different for different solutions in the population.

[0055] S12, Path-Robot Allocation Phase: If the number of constructed paths is greater than or equal to the number of robots, MILP1 allocation is used to minimize the maximum completion time;

[0056] MILP1 defines a binary decision variable zᵢⱼ; if zᵢⱼ=1, it means that the i-th prefabricated path is assigned to the j-th robot, otherwise it is 0; the objective function is to minimize the maximum completion time C. max The constraints include path allocation constraints, completion time calculation constraints (the robot's completion time is the sum of the execution times of all its allocated paths), and maximum completion time definition constraints.

[0057] The path assignment constraint is that each pre-built path is assigned to only one robot; the completion time calculation constraint is that the robot's completion time is the sum of the execution times of all its assigned paths; the maximum completion time definition constraint is C. max ≥ Completion time of any robot;

[0058] If the number of constructed paths is less than the number of robots, iteratively split the longest path until the number of paths equals the number of robots, and then perform balanced distribution.

[0059] Iterative splitting into loop execution: Identify and select the path with the longest execution time in the current path set, and use an iterative greedy strategy to split it into two sub-paths with approximately equal execution times. That is, transfer task nodes one by one from one end of the longest path to the new sub-path, and calculate the execution time difference between the two sub-paths in real time to find the split point that minimizes the time difference.

[0060] Balanced allocation is the process of assigning all paths after splitting to each robot;

[0061] S2, Initial Path Optimization: For each initial solution generated in S1, DRRM is applied at the micro-path layer to optimize the execution order of each independent task path; the specific steps of DRRM are as follows:

[0062] S21, Distance Reordering: Calculate the distance from each task point within the path to the warehouse, and reorder the task execution order in descending order of distance;

[0063] S22, 2-opt local search: Based on the reordered sequence, the neighborhood solution is explored by exchanging two disjoint edges using the 2-opt operator, such as... Figure 3 As shown, further optimization of path length and energy consumption is achieved;

[0064] S3, Iterative Optimization: Repeat the following steps before the preset termination condition is met:

[0065] S31, Robot-level Path Sequence Optimization: Apply DRRM at the macro-scheduling layer to optimize the complete task sequence assigned to each robot;

[0066] S32, Task redistribution among robots: Apply TRRM to optimize global task load balancing; wherein, TRRM performs any of the following operations with equal probability:

[0067] Operation 1, Task Exchange: Randomly select two robots and exchange any task node in their task sequences;

[0068] Operation 2, Task Reassignment: Randomly select two robots and move any task node of the robot with the longer completion time to the robot with the shorter completion time.

[0069] S33, Charge-Based Path Reconstruction: CRRM is performed on non-dominated solutions in the current population. This mechanism extracts the task sequence after each robot's last charging operation, performs local optimization, and then redistributes the tasks to reduce the interference of charging behavior on the path structure. The specific steps of CRRM are as follows:

[0070] S331, Task Extraction: Extract all tasks after the last charging operation of each robot to form a set of tasks to be optimized;

[0071] S332, Sequence Optimization: Apply DRRM optimization to the set of tasks to be optimized;

[0072] S333, Task Reassignment: MILP2 is used to redistribute the optimized task sequence to all robots, with the objective of minimizing the maximum completion time after introducing new tasks; the specific definition of MILP2 is:

[0073] Decision variables: binary variable zᵢⱼ indicates whether the i-th task to be assigned is assigned to the j-th robot, and binary variable wⱼ indicates whether the j-th robot accepts the new task;

[0074] Objective function: Minimize the maximum completion time C max ;

[0075] Constraints include task allocation integrity constraints, robot state update constraints, and maximum completion time calculation and limit constraints. The task allocation integrity constraint assigns each task to one robot. The robot state update constraint activates when the robot receives at least one new task. The maximum completion time calculation and limit constraint is that the robot's new completion time = initial task time before charging + execution time of the newly assigned task + possible battery swapping penalty time, and the new completion time ≤ C.max ;

[0076] S34, Environment Selection: The optimized new solution is merged with the original population, and the next generation population is selected by non-dominated sorting and crowding calculation.

[0077] S4, Final Refinement and Output: In the final stage of iterative optimization, a split-based path reconstruction mechanism is executed, including:

[0078] S41, Identify the longest path: Identify the path with the longest execution time among all independent task paths;

[0079] S42, Path Splitting: The longest path is split into two sub-paths with approximately equal execution times using an iterative greedy strategy;

[0080] S43, Global Reassignment: Merge the split sub-paths with other unchanged paths, apply MILP1 to redistribute all paths to all robots to further reduce the maximum completion time; the final output is the optimized non-dominated solution set.

[0081] 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 multi-robot task allocation algorithms, namely Multi-objective Discrete ArtificialBee Colony (MODABC), Advanced Multi-objective Evolutionary Algorithm (AMOEA), and Non-dominated Sorting Genetic Algorithm-Ⅱ (NSGA-Ⅱ), on the generated orchard multi-robot task allocation test set.

[0082] Table 1 Comparison of the optimal number of times achieved in the dataset experiments.

[0083] Methods\Test Set 50×50 60×60 70×70 MODABC 1 2 0 AMOEA 0 0 1 NSGA-II 0 0 0 This invention 9 8 9

[0084] 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 this invention 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.

[0085] In summary, this invention can effectively handle multi-objective, multi-robot task allocation problems including those with power constraints, providing decision-makers with a series of ideal compromise solutions.

[0086] 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 multi-robot task allocation optimization method for orchards based on hierarchical path reconstruction, characterized in that, Includes the following steps: S1, Problem Modeling and Initialization: Establish a multi-objective mathematical model that simultaneously optimizes the maximum completion time and total energy consumption. The model includes dynamic constraints on robot load, speed, energy consumption, and battery capacity. A two-stage initialization mechanism with variable load constraints is used to generate an initial solution population, and each solution adopts a hierarchical coding structure. The two-stage initialization mechanism for the variable load limit includes the following steps: S11, Path construction stage: An independent task path is constructed using a distance-based greedy strategy, while a linearly varying load limit strategy is introduced to make the upper limit of robot load different for different solutions in the population. S12, Path-Robot Allocation Stage: If the number of constructed paths is greater than or equal to the number of robots, the first mixed-integer linear programming model is used for allocation to minimize the maximum completion time; The first mixed-integer linear programming model defines binary decision variables zᵢⱼ; If zᵢⱼ=1, it means that the i-th prefabricated path is assigned to the j-th robot; otherwise, it is 0. The objective function is to minimize the maximum completion time C. max The constraints include path allocation constraints, completion time calculation constraints (the robot's completion time is the sum of the execution times of all its allocated paths), and maximum completion time definition constraints. The path allocation constraint is that each pre-built path is assigned to only one robot; the completion time calculation constraint is that the robot's completion time is the sum of the execution times of all its assigned paths; the maximum completion time definition constraint is C. max ≥ Completion time of any robot; If the number of constructed paths is less than the number of robots, iteratively split the longest path until the number of paths equals the number of robots, and then perform balanced distribution. The iterative splitting is executed in a loop: the path with the longest execution time in the current path set is identified and selected, and an iterative greedy strategy is used to split it into two sub-paths with approximately equal execution times. That is, task nodes are transferred one by one from one end of the longest path to the new sub-path, and the execution time difference between the two sub-paths is calculated in real time to find the splitting point that minimizes the time difference. The balanced allocation is to distribute all paths after the split to each robot; S2, Initial Path Optimization: For each initial solution generated in S1, a distance-based path reordering mechanism is applied at the micro-path layer to optimize the execution order of each independent task path; the specific steps of the distance-based path reordering mechanism are as follows: S21, Distance Reordering: Calculate the distance from each task point within the path to the warehouse, and reorder the task execution order in descending order of distance; S22, 2-opt local search: Based on the reordered sequence, explore neighborhood solutions by exchanging two disjoint edges using the 2-opt operator; S3, Iterative Optimization: Repeat the following steps before the preset termination condition is met: S31, Robot-level path sequence optimization: Apply a distance-based path reordering mechanism at the macro scheduling layer to optimize the complete task sequence assigned to each robot; S32, Inter-robot task redistribution: Optimize global task load balancing by applying a task-based path redistribution mechanism; S33, Charge-based path reconstruction: Perform a charge-based path reconstruction mechanism on non-dominated solutions in the current population; S34, Environment Selection: The optimized new solution is merged with the original population, and the next generation population is selected by non-dominated sorting and crowding calculation. S4, Final Refinement and Output: Execute a split-based path reconstruction mechanism, which includes: S41, Identify the longest path: Identify the path with the longest execution time among all independent task paths; S42, Path Splitting: The longest path is split into two sub-paths with approximately equal execution times using an iterative greedy strategy; S43, Global Reassignment: Merge the split sub-paths with other unchanged paths, apply the first mixed-integer linear programming model to redistribute all paths to all robots; finally output the optimized non-dominated solution set.

2. The orchard multi-robot task allocation optimization method based on hierarchical path reconstruction according to claim 1, characterized in that, The hierarchical coding structure includes a macro-scheduling layer and a micro-path layer. The macro-scheduling layer encodes the complete task allocation scheme into a multi-dimensional structure that includes a global task sequence, a task mapping sequence for each robot, a performance index set, and a charging position record. The global task sequence distinguishes different robots and different paths through a specific separator. The micro-path layer encodes each independent task path into a triple that includes a task sequence, execution time, and energy consumption.

3. The orchard multi-robot task allocation optimization method based on hierarchical path reconstruction according to claim 1, characterized in that, In S32, the task-based path reallocation mechanism performs any of the following operations based on a preset probability (equal probability is used here): Operation 1, Task Exchange: Randomly select two robots and exchange any task node in their task sequences; Operation 2, Task Reassignment: Randomly select two robots and move any task node of the robot with the longer completion time to the robot with the shorter completion time.

4. The orchard multi-robot task allocation optimization method based on hierarchical path reconstruction according to claim 3, characterized in that, The preset probability is an equal probability.

5. The orchard multi-robot task allocation optimization method based on hierarchical path reconstruction according to claim 1, characterized in that, In S33, the specific steps of the charging-based path reconstruction mechanism are as follows: S331, Task Extraction: Extract all tasks after the last charging operation of each robot to form a set of tasks to be optimized; S332, Sequence optimization: Apply a distance-based path reordering mechanism to the set of tasks to be optimized; S333, Task Reassignment: The optimized task sequence is reassigned to all robots using the second mixed-integer linear programming model, with the goal of minimizing the maximum completion time after introducing new tasks.

6. The orchard multi-robot task allocation optimization method based on hierarchical path reconstruction according to claim 1, characterized in that, In S333, the specific definition of the second mixed-integer linear programming model is: Decision variables: binary variable zᵢⱼ indicates whether the i-th task to be assigned is assigned to the j-th robot, and binary variable wⱼ indicates whether the j-th robot accepts the new task; Objective function: Minimize the maximum completion time C max ; Constraints include task allocation integrity constraints, robot state update constraints, and maximum completion time calculation and limitation constraints. The task allocation integrity constraint assigns each task to one robot. The robot state update constraint activates when the robot receives at least one new task. The maximum completion time calculation and limitation constraint is that the robot's new completion time = initial task time before charging + execution time of the newly assigned task + possible battery swapping penalty time, and the new completion time ≤ C. max .

Images