Multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm

Abstract

This invention relates to the field of multi-robot cooperative control and discloses a multi-robot cooperative task allocation method based on a hybrid particle swarm optimization algorithm. It aims to solve the problems of low efficiency in heterogeneous task allocation under dynamic environments and long global replanning time due to sudden events in traditional algorithms. The method includes: acquiring cluster status and task information; constructing a discretized particle swarm model based on a capability matching matrix to eliminate invalid solutions; introducing a dynamic energy penalty factor to construct a multi-objective fitness function; performing particle iterative optimization through a discrete crossover and mutation mechanism; issuing the optimal solution and monitoring sudden events, freezing the normal sequence when triggered, and extracting only local subgroups for rapid rescheduling. This invention effectively improves the optimization efficiency of clusters and the dynamic fault tolerance robustness of the system in complex industrial scenarios.

Description

Technical Field

[0001] This invention relates to the field of multi-robot cooperative control and discloses a multi-robot cooperative task allocation method based on a hybrid particle swarm optimization algorithm. Background Technology

[0002] With the development of intelligent manufacturing and warehousing logistics, multi-robot collaborative task allocation has become a core underlying control technology. Existing technologies typically employ swarm intelligence algorithms for solving this problem. For example, Chinese patent document CN107168054A discloses a "Multi-Robot Task Allocation and Path Planning Method," which allocates tasks by establishing a comprehensive cost evaluation function that considers factors such as robot battery power and working time. However, this method is primarily designed for static environments and represents a centralized, one-time global allocation. In complex industrial settings, when faced with unexpected events, this method must interrupt all unfinished tasks of all robots, re-initialize the global system, and perform a complete PSO iterative optimization. This global replanning mechanism incurs enormous computational overhead and extremely high system reconfiguration delays, completely failing to meet the real-time scheduling and fault-tolerant control requirements of dynamic and complex environments.

[0003] On the other hand, the standard PSO algorithm is essentially a continuous space optimization algorithm, which suffers from serious mapping defects when directly applied to highly discretized task allocation problems. For example, Chinese patent document CN105069530A discloses a "Multi-robot Task Allocation Method Based on Multi-Objective Optimization," which uses an improved particle swarm optimization algorithm to solve for time utility and energy utility. However, this prior art lacks deep physical constraint filtering for the "heterogeneous capabilities" of the cluster when updating the position of particles and discretizing the mapping, resulting in a large number of "invalid dead solutions" that assign tasks to robots without execution capabilities during the iteration process. In addition, its objective function fails to construct nonlinear constraints for extreme hardware states, which can easily lead to excessive power consumption and mid-process crashes of individual underlying devices when facing large-scale scheduling.

[0004] In summary, existing multi-robot task allocation technologies based on particle swarm optimization generally suffer from a disconnect between the underlying algorithm architecture and the complex physical control constraints of reality. Their core technical bottlenecks are: first, the lack of an efficient dynamic event-based local rescheduling mechanism, resulting in poor system robustness in handling unexpected situations and excessively long replanning times; second, the lack of effective discretization filtering and dynamic hardware state protection mechanisms for heterogeneous clusters, leading to a large amount of invalid computation during algorithm optimization, slow convergence speed, and difficulty in directly meeting the standards of practical engineering applications with multiple objectives and constraints. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-robot collaborative task allocation method based on a hybrid particle swarm optimization algorithm. Its core lies in constructing a global-local collaborative scheduling architecture that deeply couples actual physical constraints with a dynamic fault-tolerant mechanism. First, by constructing a capability matching matrix that integrates the heterogeneous characteristics of robots, the discrete particle swarm model is validated and filtered for legality, eliminating invalid allocation solutions from the bottom layer to significantly accelerate algorithm convergence. Second, in the fitness function with time, energy consumption, and load balancing as multiple objectives, a dynamic penalty factor based on the power safety boundary is innovatively introduced to achieve proactive protection of the underlying device hardware lifespan. Discrete crossover and mutation mechanisms are used to replace continuous speed updates for hybrid iterative optimization. Finally, an event-triggered local rescheduling mechanism is designed. When faced with sudden events such as emergency order insertion or equipment failure, the system does not need to perform high-latency global replanning. Instead, it freezes the unaffected task execution sequence and extracts only the affected local tasks and associated robot reconstructed subgroups for rapid fine-tuning. This ensures high-quality global scheduling while greatly improving the real-time response speed and fault-tolerant robustness of multi-robot clusters in complex and dynamic industrial environments.

[0006] The objective of this invention can be achieved through the following technical solutions: A multi-robot cooperative task allocation method based on a hybrid particle swarm optimization algorithm includes the following steps: S1. Obtain real-time status information of multiple robot clusters and environmental information of the task set to be assigned; the real-time status information includes robot coordinates, remaining battery power, and heterogeneous capability tags; the environmental information includes task coordinates, priority, and capability requirements. S2. Construct a discretized particle swarm model, mapping each particle to an allocation scheme. Use integer encoding to generate an initial particle swarm, with particle dimensions corresponding to tasks and dimension values ​​corresponding to robot numbers. Establish a capability matching matrix based on the labels and capability requirements, filter particle position variables, and eliminate invalid allocation solutions. S3. Construct a multi-objective fitness function with the objectives of minimizing the maximum completion time, minimizing the total energy consumption, and maximizing the load balance, and introduce a dynamic penalty factor based on the power consumption: when the scheme causes the expected remaining power of a robot to be lower than a preset threshold, the penalty factor increases exponentially, reducing the fitness value of the particle. S4. Calculate the particle fitness value and update the individual and global optimal positions. Introduce a discrete mixing mechanism in the update: convert the continuous velocity update into discrete crossover and mutation operations. Particles crossover and recombine with the individual and global optimal positions with a set probability, and perform mutation with a set probability. Iterate until the convergence condition is met and output the optimal solution. S5. Parse the optimal solution into a task sequence and send it to the robot for execution; monitor for sudden events during execution, and when a threshold is triggered, freeze the unaffected task sequence, extract the affected tasks and associated robots to form a local subgroup, and return to S4 for local rescheduling.

[0007] Preferably, in step S1, the heterogeneous capability label includes the robot's load level, end effector type, or sensor configuration; the capability requirements include the minimum load threshold or specific hardware matching type required to complete the task.

[0008] Preferably, in step S2, the capability matching matrix is ​​a two-dimensional Boolean matrix, where the rows of the matrix correspond to the tasks in the task set and the columns correspond to the robots in the cluster; if the heterogeneous capability tag of a robot meets the capability requirements of a task, then the corresponding element of the capability matching matrix is ​​1, otherwise it is 0. The specific method for eliminating invalid allocation solutions is as follows: when generating the initial particle swarm, if the robot corresponding to a certain dimension of the particle and the element of the task in that dimension in the capability matching matrix are 0, then a robot number with a matrix element of 1 is randomly assigned again until all dimensions of the particle meet the matching requirements.

[0009] Preferably, in step S3, the method for calculating the minimum maximum completion time is as follows: The total time taken for each robot in the cluster to complete all tasks in its assigned sequence is calculated, and the maximum value among the total times of all robots is taken as the maximum completion time. The total time is composed of the robot's movement time from the current coordinate to the task coordinate, the waiting time after arriving at the task point, and the physical execution time of the task.

[0010] Preferably, in step S3, the robot's expected remaining power is: the robot's current remaining power minus the predicted total energy consumption required to execute its assigned sequence of tasks; The dynamic penalty factor based on battery power is calculated using an exponential penalty function: when the expected remaining battery power of a robot is greater than or equal to a preset threshold, the penalty factor is zero; when the expected remaining battery power is less than the preset threshold, the penalty factor increases exponentially by the difference between the preset threshold and the expected remaining battery power, and is added as a penalty term to the fitness value of the particle.

[0011] Preferably, in step S4, the specific method for converting continuous velocity updates into discrete crossover and mutation operations is as follows: A random number is generated. If the random number is less than a set first probability, a random dimension segment of the particle's current position is selected and cross-substituted with the individual's optimal position. If the random number is less than a set second probability, a random dimension segment of the particle's current position is selected and cross-substituted with the global optimal position. The new particle position after each cross-substitution needs to be validated again by the capability matching matrix.

[0012] Preferably, in step S4, the specific method for performing the mutation operation is as follows: Generate a random number. If the random number is less than the set third probability, randomly select one or more dimensions of the particle's current position, query the capability matching matrix, and randomly replace the value of the selected dimension with other valid robot numbers that meet the task capability requirements of that dimension.

[0013] Preferably, in step S4, the specific method for outputting the optimal solution is as follows: The non-dominated sorting algorithm is used to sort the front of the particle swarm that meets the convergence condition, calculate the crowding distance of each particle, and construct and output the Pareto optimal solution set. The convergence condition is: the Pareto optimal solution set is not updated within a preset number of consecutive iterations, or the algorithm reaches the preset maximum number of iterations.

[0014] Preferably, in step S5, the emergency events include adding an emergency insertion task, a mechanical failure of the robot, or a loss of robot communication; the trigger thresholds include the priority of the added task being higher than the highest priority of the currently executing task, or the downtime caused by the robot failure exceeding a preset safety tolerance period.

[0015] As a preferred embodiment, in step S5, the specific method for extracting the local subgroups formed by the affected tasks and associated robots is as follows: Extract all unexecuted tasks within a preset radius of the spatial coordinates of the sudden event, as well as all unexecuted tasks originally planned to be assigned to the robot that experienced the sudden event, and merge them as affected tasks; extract the robots responsible for the affected tasks and the robots in an idle state as associated robots; input the affected tasks and associated robots into step S4 to reconstruct the local particle swarm for iterative optimization.

[0016] The beneficial effects of this invention are: This invention introduces a capability matching matrix based on heterogeneous capability tags to rigorously verify the physical validity of particle dimensions during the initialization and subsequent crossover and mutation iteration stages of the discretized particle swarm model. This mechanism directly eliminates invalid allocation solutions that cannot be executed by actual robots from the underlying structure, avoiding the waste of computational power on meaningless dead solutions in traditional algorithms. This significantly reduces the overall computational complexity of the system, improves the global convergence speed of the algorithm, and ensures the high adaptability and reliable implementation of the final solution across complex heterogeneous devices.

[0017] This invention innovatively designs a dynamic penalty factor based on a power safety boundary in the multi-objective fitness function. By actively avoiding high-energy-consuming tasks by using nonlinear exponential penalties during the optimization process, high-energy-consuming tasks are assigned to robots approaching their power threshold, thus achieving proactive intervention in hardware extreme states. This design effectively prevents task interruption or even system crashes caused by excessive load or power depletion in a single device. While achieving overall load balancing of the cluster, it greatly extends the physical lifespan and fault-free operation cycle of the multi-robot system.

[0018] This invention addresses the unpredictable and unforeseen situations in industrial settings by designing an event-triggered local rescheduling mechanism. When faced with urgent order insertions or local equipment failures, the system breaks away from the limitations of traditional algorithms that require global rescheduling. Instead, it selectively freezes unaffected normal execution sequences, extracting only local tasks within the spatial intersection range and reconstructing subgroups of associated robots for rapid iteration. This "local fine-tuning" strategy preserves the original optimal allocation scheme to the greatest extent, enabling the system to quickly resolve conflicts with extremely low computational latency, thus endowing multi-robot collaborative scheduling with strong dynamic fault tolerance and system robustness. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the multi-robot collaborative task allocation method based on the hybrid particle swarm optimization algorithm of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] Example: The system on which this invention is based typically includes: a central scheduling server, a communication network, and a physical execution cluster composed of multiple heterogeneous robots.

[0022] The central scheduling server is equipped with a high-performance processor and memory to run the hybrid particle swarm scheduling algorithm described in this invention. Each robot at the bottom layer is equipped with a corresponding environmental perception and localization module, as well as an internal state perception module, in the physical space. The robots and the central scheduling server exchange data frequently and in real-time via industrial IoT protocols such as MQTT or TCP / IP.

[0023] like Figure 1 As shown in the figure, this embodiment provides a multi-robot cooperative task allocation method based on a hybrid particle swarm optimization algorithm. The core execution flow is detailed below: In this embodiment, step S1 includes acquiring physical environment data and cluster status parameters, and the specific implementation method is as follows: In the initial stage of system operation, or when a rescheduling event is triggered, the data acquisition module of the central scheduling server synchronously pulls the real-time status information of the underlying multi-robot cluster and the set of tasks to be assigned issued by the upstream business system through the wireless communication network.

[0024] Robot state modeling: The system defines a multi-robot cluster as... , where m is the total number of robots currently online in the cluster. For any online robot The system encapsulates its state parameters into a tuple containing dynamic data and static features. .

[0025] coordinate parameters : Represents robot The current absolute coordinates in three-dimensional physical space. If it is a planar AGV, then... Constant; if it is a drone or a 3D stacker, then consider the height dimension.

[0026] Motion parameters : Indicates the robot's rated average moving speed or current instantaneous speed.

[0027] Battery parameters This indicates the current remaining battery power (SOC) reported by the robot through the BMS, and the value range is usually from 0% to 100%.

[0028] Heterogeneous capability tags This is a Boolean or enumerated data set used to describe the robot's physical hardware attributes. For example, the label dictionary could include [load capacity, lifting capacity, vision detection capacity, robotic arm grasping capacity]. If... For a heavy-duty AGV with lifting capabilities, its capability tag can be coded as follows: .

[0029] Task environment modeling: The system defines the set of tasks to be assigned as follows: , where n is the total number of tasks in the current scheduling period. For any task Its environmental parameters are defined as tuples. .

[0030] coordinate parameters : Indicates task The physical point of occurrence.

[0031] Time consumption parameters : Indicates the baseline physical operation time required to complete the task.

[0032] Priority weight : Indicates the urgency level of the task, which is preset through the business system, with values ​​such as 1 (normal), 2 (urgent), and 3 (critical).

[0033] Capability requirement matching matrix : Capability tags for robots Correspondingly, the underlying hardware requirements for performing this task are defined. If a material handling task requires the equipment to have heavy-duty and lifting capabilities, then... .

[0034] In this embodiment, step S2: Constructing a discretized particle swarm model based on heterogeneous capability matching is specifically implemented as follows: Standard particle swarm optimization (PSO) seeks the optimal solution in a continuous real space, while multi-robot task assignment is a typical NP-hard discrete combinatorial optimization problem. Using conventional sorting methods would result in a large number of invalid mappings. This invention innovatively employs an integer encoding mechanism that incorporates physical constraints on capabilities.

[0035] Particle swarm coding mechanism: Let the particle swarm size be PopSize, and the k-th particle be denoted as... Its dimensional length is strictly equal to the total number of tasks, n. The value of the particle in the j-th dimension... It is an integer, and Its physical meaning is: Under this scheme, the task It was explicitly assigned to the number The robot performs the task.

[0036] Establishment and filtering validation of the capability matching matrix: To avoid assigning physically infeasible "dead solutions" such as heavy-load tasks to ordinary AGVs, the system initializes the particle population based on the tags in step S1. With demand Automatically generate a dimension as Two-dimensional Boolean capability matching matrix .

[0037] Its elements The computational logic is as follows: compare the robot's data through a logical AND operation. tags and tasks demand .if Include If all the required attributes are met, the robot is deemed to possess the physical capability to perform the task. If any one of the hard requirements is not met, then .

[0038] When the system generates the initial particle swarm using a pseudo-random number generator or produces new particles in subsequent mutation operations, it must undergo mandatory filtering by this matrix. The specific verification process is as follows: Traversing particles All dimensions Check the corresponding elements in the matrix. The value of . If If the dimension is valid, then that dimension is valid; if This indicates that the system randomly generated an invalid assignment. The system triggers the spinlock reset mechanism, immediately re-randomly selecting a number from the set of robot numbers with a value of 1 in the j-th column of the matrix to replace the current one. This mechanism ensures that 100% of the solutions in the initial solution space of the algorithm are physically executable, greatly reducing the invalid search space.

[0039] In this embodiment, step S3 includes constructing a multi-objective fitness function constrained by a dynamic penalty factor, specifically implemented as follows: To address the multidimensional requirements of time efficiency, energy consumption, and hardware lifespan in real-world industrial scenarios, this invention constructs a nonlinear, multi-objective fitness evaluation system.

[0040] Time dimension: Minimize the maximum completion time, denoted as .

[0041] Since the bottleneck of system scheduling often depends on the slowest robot in the cluster or the robot with the heaviest assigned tasks, the system iterates through the allocation schemes corresponding to particles and assigns them to the same robot. All tasks are sorted according to proximity or pre-defined business logic to form an execution sequence set. .

[0042] robot Total time required to physically complete this sequence The calculation formula is as follows: In the formula, the first term represents the path time of spatial movement ( These are the coordinates of the previous task node; for the first task in the sequence, they are the robot's current physical coordinates. ), This represents the time taken for physical operations at the task node.

[0043] The objective function for the maximum completion time of the entire cluster is: .

[0044] Energy consumption dimension: Minimize the total energy consumption of the cluster, denoted as .

[0045] Significant differences exist in the energy consumption per 100 kilometers and standby energy consumption of different heterogeneous robots. (The text then abruptly shifts to a seemingly unrelated topic: setting up robots.) The power consumption per unit distance traveled is The power consumption per unit time of operation is The system calculates the total predicted energy consumption of all robots under this allocation scheme using the following formula: .

[0046] Lifetime dimension: Dynamic penalty mechanism based on power safety boundary.

[0047] In order to completely solve the industry problem of "a machine being over-assigned tasks, resulting in a power outage and paralysis" in the existing technology, this embodiment introduces an exponential hardware protection penalty factor.

[0048] Before each particle evaluation, the system uses the aforementioned energy consumption formula to calculate the robot's energy consumption. Expected remaining battery power after completing the assigned task Let the factory's required safe power threshold be... It is usually set to 20% to ensure that it has enough power to return to the charging station.

[0049] Define penalty items as follows: Where the expected battery level is greater than the safety threshold, the penalty value is 0, and the system can freely schedule power allocation; however, if the allocation scheme causes a robot's battery level to fall below the red line, the penalty will increase exponentially to the base e. Based on the penalty constant, The steepness coefficient.

[0050] Finally, a comprehensive fitness function for a single particle is constructed through a linear weighted fusion mechanism. : ,in , These are the weighting coefficients. To penalize the amplification factor, a maximum value is usually taken to achieve a veto effect.

[0051] In this embodiment, step S4 includes performing particle swarm optimization based on a hybrid strategy, specifically implemented as follows: In the traditional PSO algorithm, the update formula for velocity V is... This invention is only applicable to continuous floating-point numbers. It utilizes the operator concept from genetic algorithms to discretize and reconstruct the evolution equation of PSO.

[0052] For the k-th particle in the population, its position at the (t+1)-th iteration is... The update process is broken down into a hybrid operation based on probability intervals. The system generates a uniformly distributed random number in the range [0,1). And determine the direction of particle evolution based on a set probability threshold: Detailed analysis of the hybrid optimization mechanism: Cognitive learning and individual optimal crossover (probability) When a random number falls within this interval, the particle's historical best solution is... Crossover and recombination occurred. The system randomly selected... The sequence of several dimensions in the algorithm directly covers the current particle, which means that the algorithm uses local historical experience to converge.

[0053] Social learning and global optimal intersection (probability) When a random number falls within this interval, the particle is attracted by the shared information of the entire population and replaces part of its own dimension with the global optimal solution of the current population. The corresponding dimension sequence accelerates the algorithm's approach to the global extremum.

[0054] Physical constraint variation (probability) To prevent the algorithm from getting stuck in local optima, a mutation operation is introduced. Mutation function. The execution process is as follows: discard the current task allocation result in the j-th dimension, and instead query the capability matching matrix. The j-th column is randomly selected from the group of machines capable of execution to replace the current value. This operation greatly maintains the genetic diversity of the population.

[0055] Remain unchanged: Within the remaining probability interval, the particle retains its previous generation's allocation scheme.

[0056] It is worth noting that, regardless of whether crossover or mutation occurs, the new particles generated after each operation must be immediately put into the capability matching matrix described in step S2 for validity verification.

[0057] The system iteratively performs the above evaluation and update, while maintaining a Pareto optimal solution set using a non-dominated sorting algorithm. The algorithm continues until it reaches a preset maximum number of iterations. When the time is reached, the loop terminates and the optimization result is output.

[0058] In this embodiment, step S5 includes task plan issuance and local rescheduling triggered by dynamic events, and the specific implementation method is as follows: Based on the current management preferences of the workshop, the central server extracts the unique globally optimal allocation scheme from the Pareto solution set, parses it into a specific sequence of action instructions, and sends it to the PLC or ROS control nodes of each robot at the bottom layer via the MQTT protocol.

[0059] Dynamic event monitoring and local robust refactoring: Existing task allocation methods often lead to workshop shutdowns and waiting for recalculation due to rigid algorithms in the event of unforeseen circumstances. This embodiment constructs a highly fault-tolerant local rescheduling mechanism.

[0060] On the parallel timeline of the robot's task execution, the central server establishes an independent watchdog thread to frequently poll the environment status through heartbeat packets and business interfaces.

[0061] The preset emergency trigger conditions include: The system received an urgent order insertion task with a priority of 2 or higher. ); A certain robot If packet loss continues for more than the set time, or if the BMS / servo motor reports an abnormal error code.

[0062] Local rescheduling execution logic: Once any of the above events is triggered, the system immediately intercepts the global scheduling process and performs the following actions: Step 1: Freeze the normal sequence. For the remaining robots that are unaffected by the event and are operating normally, their underlying controllers lock the current short-term task execution sequence to ensure that the basic production rhythm of the workshop is not disrupted.

[0063] Step 2: Extract the local influence subgroup. Set the coordinates of the event occurrence center as... The spatial influence radius threshold is The system scans the global task queue and extracts tasks that meet the following criteria to form a local planning pool. : Originally assigned to the faulty robot All unstarted tasks; Spatial location satisfies the formula Other tasks to be performed, namely tasks that are physically close to the location of the accident or the location of the new order.

[0064] Step 3: Local low-dimensional iteration. The system will... The tasks and the nearest available robots in the region are extracted. By eliminating a large number of irrelevant remote tasks and robots, a very low-dimensional local mapping space is formed. The system uses this local data as initial input and quickly redirects back to step S4. At this point, the hybrid PSO algorithm only optimizes this low-dimensional subgroup.

[0065] Because the problem size is reduced geometrically, the algorithm can quickly converge and output a local remedial solution within tens of milliseconds. This rescheduling based on spatial isolation mechanism perfectly resolves sudden conflicts with minimal system computing power overhead while maintaining most of the original high-quality allocations globally, demonstrating extremely high industrial practical value.

[0066] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-robot cooperative task allocation method based on a hybrid particle swarm optimization algorithm, characterized in that, Includes the following steps: S1. Obtain real-time status information of multiple robot clusters and environmental information of the task set to be assigned; the real-time status information includes robot coordinates, remaining battery power, and heterogeneous capability tags; the environmental information includes task coordinates, priority, and capability requirements. S2. Construct a discretized particle swarm model, mapping each particle to an allocation scheme. Use integer encoding to generate an initial particle swarm, with particle dimensions corresponding to tasks and dimension values ​​corresponding to robot numbers. Establish a capability matching matrix based on the labels and capability requirements, filter particle position variables, and eliminate invalid allocation solutions. S3. Construct a multi-objective fitness function with the objectives of minimizing the maximum completion time, minimizing the total energy consumption, and maximizing the load balance, and introduce a dynamic penalty factor based on the power consumption: when the scheme causes the expected remaining power of a robot to be lower than a preset threshold, the penalty factor increases exponentially, reducing the fitness value of the particle. S4. Calculate the particle fitness value and update the individual and global optimal positions. Introduce a discrete mixing mechanism in the update: convert the continuous velocity update into discrete crossover and mutation operations. Particles crossover and recombine with the individual and global optimal positions with a set probability, and perform mutation with a set probability. Iterate until the convergence condition is met and output the optimal solution. S5. Parse the optimal solution into a task sequence and send it to the robot for execution; monitor for sudden events during execution, and when a threshold is triggered, freeze the unaffected task sequence, extract the affected tasks and associated robots to form a local subgroup, and return to S4 for local rescheduling.

2. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S1, the heterogeneous capability label includes the robot's load capacity level, end effector type, or sensor configuration; the capability requirements include the minimum load capacity threshold or specific hardware matching type required to complete the task.

3. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S2, the capability matching matrix is ​​a two-dimensional Boolean matrix, where the rows of the matrix correspond to the tasks in the task set and the columns correspond to the robots in the cluster. If a robot's heterogeneous capability tag meets the capability requirements of a certain task, then the corresponding element of the capability matching matrix is ​​1; otherwise, it is 0. The specific method for eliminating invalid allocation solutions is as follows: when generating the initial particle swarm, if the robot corresponding to a certain dimension of the particle and the element of the task in that dimension in the capability matching matrix are 0, then a robot number with a matrix element of 1 is randomly assigned again until all dimensions of the particle meet the matching requirements.

4. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S3, the method for minimizing the maximum completion time is as follows: The total time taken for each robot in the cluster to complete all tasks in its assigned sequence is calculated, and the maximum value among the total times of all robots is taken as the maximum completion time. The total time is composed of the robot's movement time from the current coordinate to the task coordinate, the waiting time after arriving at the task point, and the physical execution time of the task.

5. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S3, the robot's expected remaining power is: the robot's current remaining power minus the predicted total energy consumption required to execute its assigned sequence of tasks; The dynamic penalty factor based on battery power is calculated using an exponential penalty function: when the expected remaining battery power of a robot is greater than or equal to a preset threshold, the penalty factor is zero; when the expected remaining battery power is less than the preset threshold, the penalty factor increases exponentially by the difference between the preset threshold and the expected remaining battery power, and is added as a penalty term to the fitness value of the particle.

6. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S4, the specific method for converting continuous velocity updates into discrete crossover and mutation operations is as follows: A random number is generated. If the random number is less than a set first probability, a random dimension segment of the particle's current position is selected and cross-substituted with the individual's optimal position. If the random number is less than a set second probability, a random dimension segment of the particle's current position is selected and cross-substituted with the global optimal position. The new particle position after each cross-substitution needs to be validated again by the capability matching matrix.

7. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S4, the specific method for performing the mutation operation is as follows: Generate a random number. If the random number is less than the set third probability, randomly select one or more dimensions of the particle's current position, query the capability matching matrix, and randomly replace the value of the selected dimension with other valid robot numbers that meet the task capability requirements of that dimension.

8. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S4, the specific method for outputting the optimal solution is as follows: The non-dominated sorting algorithm is used to sort the front of the particle swarm that meets the convergence condition, calculate the crowding distance of each particle, and construct and output the Pareto optimal solution set. The convergence condition is: the Pareto optimal solution set is not updated within a preset number of consecutive iterations, or the algorithm reaches the preset maximum number of iterations.

9. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S5, the sudden events include adding an emergency insertion task, mechanical failure of the robot, or loss of robot communication; the trigger thresholds include the priority of the newly added task being higher than the highest priority of the currently executing task, or the downtime caused by the robot failure exceeding the preset safety tolerance time.

10. The multi-robot cooperative task allocation method based on hybrid particle swarm optimization algorithm according to claim 1, characterized in that, In step S5, the specific method for extracting the local subgroups formed by the affected tasks and associated robots is as follows: Extract all unexecuted tasks within a preset radius of the spatial coordinates of the sudden event, as well as all unexecuted tasks originally planned to be assigned to the robot that experienced the sudden event, and merge them as affected tasks; extract the robots responsible for the affected tasks and the robots in an idle state as associated robots; input the affected tasks and associated robots into step S4 to reconstruct the local particle swarm for iterative optimization.

Images