Multi-robot collaborative task dynamic scheduling method and related equipment

Abstract

The invention provides a multi-robot collaborative task dynamic scheduling method and related equipment, which are applied to the technical field of robots, and the method comprises the following steps: collecting task scheduling scene information; the task scheduling scene information is used for constructing a multi-level graph structure, multi-dimensional input information is obtained based on the multi-level graph structure, the multi-level graph structure comprises a relation graph layer of multiple dimensions, and the relation graph layer comprises relation information between objects of the dimensions corresponding to the relation graph layer; determining a task allocation strategy based on the multi-dimensional input information; and based on the task allocation strategy, sending a task execution instruction to each target robot, and controlling the target robot to execute a task corresponding to the task execution instruction. Complicated relations among the robot, the environment and the tasks are expressed by using the multi-level graph structure, rich and complete relation information is provided for generating the task allocation strategy, the accuracy of the generated task allocation strategy is improved, and conflicts generated during task allocation are avoided.

Description

Technical Field

[0001] The present invention relates to the field of robot technology, and in particular to a multi-robot collaborative task dynamic scheduling method and related equipment. Background Art

[0002] With the rapid development of science and technology, multi-robots are increasingly widely used in manufacturing, logistics, home services, medical care and other fields. In manufacturing, multi-robots can greatly improve production efficiency and reduce costs; in the field of logistics, multi-robots can achieve fast and efficient cargo handling; in the service industry, multi-robots can provide consumers with more convenient and intelligent services.

[0003] However, due to the high degree of autonomy and complexity of multi-robot systems, their scheduling problem has become a challenging research field. Traditional multi-robot scheduling solutions often find it difficult to make effective scheduling decisions when faced with complex environments, diverse tasks, and complex interactions between robots, resulting in overall inefficiency and even conflicts. Summary of the invention

[0004] In view of this, an embodiment of the present application provides a method and related equipment for dynamic scheduling of tasks for multi-robot collaboration, which uses a multi-level graph structure to express the complex relationship between robots, environment and tasks, provides rich and complete relationship information for generating task allocation strategies, improves the accuracy of the generated task allocation strategies, and avoids conflicts when allocating tasks.

[0005] To achieve the above objectives, the present application provides the following technical solutions:

[0006] A multi-robot collaborative task dynamic scheduling method, comprising:

[0007] Collecting task scheduling scenario information, wherein the task scheduling scenario information includes robot information, environment information, and task information, wherein the robot information includes status information of each robot, and the task information includes information of each task to be executed in the task queue;

[0008] Applying the task scheduling scenario information to construct a multi-level graph structure, and obtaining multi-dimensional input information based on the multi-level graph structure, wherein the multi-level graph structure includes relationship layers of multiple dimensions, and the relationship layers include relationship information between objects of the dimensions corresponding to the relationship layers;

[0009] Determining a task allocation strategy based on the multi-dimensional input information;

[0010] Based on the task allocation strategy, a task execution instruction is sent to each target robot to control each target robot to execute the task corresponding to the task execution instruction.

[0011] In the above method, optionally, the relationship layers of multiple dimensions included in the multi-level graph structure include at least one of: a reachability layer, a task compatibility layer, a resource dependency layer, a timing dependency layer, a communication network layer, and an energy distribution layer;

[0012] The reachability layer includes execution path information of reachable paths between each robot and the execution point of each task;

[0013] The task compatibility layer includes task attribute information of executable task types of each robot;

[0014] The resource dependency layer includes resource dependency information of each task on various resources;

[0015] The timing dependency layer includes timing dependency information of the sequence between each task;

[0016] The communication network layer includes communication attribute information of the communication status between each robot and the system;

[0017] The energy distribution layer includes charging path information of accessible paths between the robot that needs to be charged and each charging station.

[0018] In the above method, optionally, the step of obtaining multi-dimensional input information based on the multi-level graph structure includes:

[0019] Using a pre-trained graph neural network model, multidimensional relationship expression data is extracted from the multi-level graph structure, wherein the multi-dimensional relationship expression data includes expression data of the relationship between each object in the multi-level graph structure; and multi-dimensional input information is generated based on the multi-dimensional relationship expression data.

[0020] In the above method, optionally, determining the task allocation strategy based on the multi-dimensional input information includes:

[0021] The multi-dimensional input information is input into a pre-trained decision model, and a task allocation strategy output by the decision model after processing the multi-dimensional input information is obtained.

[0022] The above method, optionally, inputting the multi-dimensional input information into a pre-trained decision model, and obtaining the task allocation strategy output by the decision model after processing the multi-dimensional input information, comprises:

[0023] The multidimensional input information is input into a pre-trained decision model, and the decision model obtains task requirement information from the multidimensional input information, wherein the task requirement information includes requirement feature information of the task to be performed; and based on the task requirement information and the multidimensional input information, a plurality of allocation combinations are generated, wherein the allocation combinations include selecting each robot participating in the execution of the task and the tasks assigned to each robot participating in the execution of the task; the decision model evaluates the expected return value of each allocation combination, and obtains the task allocation strategy generated by the decision model based on the allocation combination with the highest expected return value.

[0024] In the above method, optionally, the decision model is optimized in the following manner:

[0025] Collect feedback data of the task allocation strategy generated by the decision model, and use the feedback data to train the decision model to adjust the weight values ​​of various influencing factors of the reward function in the decision model, until the decision model meets the preset end training conditions, thereby completing the optimization of the decision model.

[0026] In the above method, optionally, sending a task execution instruction to each target robot based on the task allocation strategy includes:

[0027] Based on the task allocation strategy, determining each target robot and determining the task to be performed by each target robot;

[0028] Acquire a global environment map, and use the global environment map to plan a motion path for each of the target robots;

[0029] Based on the motion path and task of each target robot, generating a task execution instruction for each target robot;

[0030] Each of the task execution instructions is sent to the target robot corresponding to it.

[0031] A multi-robot collaborative task dynamic scheduling system, comprising:

[0032] A collection unit, used to collect task scheduling scene information, the scene information includes robot information, environment information and task information, the robot information includes status information of each robot, and the task information includes information of each task to be assigned;

[0033] A construction unit, used to construct a multi-level graph structure by applying the task scheduling scenario information, and obtain multi-dimensional input information based on the multi-level graph structure, wherein the multi-level graph structure includes relationship layers of multiple dimensions, and the relationship layers include relationship information between objects of the dimensions corresponding to the relationship layers, wherein the object is one of the environment, the task, and the robot;

[0034] A decision-making unit, configured to determine a task allocation strategy based on the multi-dimensional input information;

[0035] The execution unit is used to send a task execution instruction to each target robot based on the task allocation strategy to control each target robot to execute the task corresponding to the task execution instruction.

[0036] In the above system, optionally, the relationship layers of multiple dimensions included in the multi-level graph structure include at least one of: a reachability layer, a task compatibility layer, a resource dependency layer, a timing dependency layer, a communication network layer, and an energy distribution layer;

[0037] The reachability layer includes execution path information of reachable paths between each robot and the execution point of each task;

[0038] The task compatibility layer includes task attribute information of executable task types of each robot;

[0039] The resource dependency layer includes resource dependency information of each task on various resources;

[0040] The timing dependency layer includes timing dependency information of the sequence between each task;

[0041] The communication network layer includes communication attribute information of the communication status between each robot and the system;

[0042] The energy distribution layer includes charging path information of accessible paths between the robot that needs to be charged and each charging station.

[0043] In the above system, optionally, the construction unit executes a process of acquiring multi-dimensional input information based on the multi-level graph structure, including:

[0044] Using a pre-trained graph neural network model, multidimensional relationship expression data is extracted from the multi-level graph structure, wherein the multi-dimensional relationship expression data includes expression data of the relationship between each object in the multi-level graph structure; and multi-dimensional input information is generated based on the multi-dimensional relationship expression data.

[0045] In the above system, optionally, the decision unit performs a process of determining a task allocation strategy based on the multi-dimensional input information, including:

[0046] The multi-dimensional input information is input into a pre-trained decision model, and a task allocation strategy output by the decision model after processing the multi-dimensional input information is obtained.

[0047] In the above system, optionally, the decision unit executes a process of inputting the multi-dimensional input information into a pre-trained decision model to obtain a task allocation strategy output after the decision model processes the multi-dimensional input information, including:

[0048] The multidimensional input information is input into a pre-trained decision model, the decision model obtains task requirement information from the multidimensional input information, and generates multiple allocation combinations based on the task requirement information and the multidimensional input information, the allocation combination including selecting each robot participating in the task and the task assigned to each robot participating in the task; the decision model evaluates the expected return value of each allocation combination, and obtains the task allocation strategy generated by the decision model based on the allocation combination with the highest expected return value.

[0049] In the above system, optionally, the decision model is optimized in the following manner:

[0050] Collect feedback data after the decision model generates a task allocation strategy, and use the feedback data to train the decision model to adjust the weight values ​​of various influencing factors of the reward function in the decision model, until the decision model meets the preset end training conditions, and the optimization of the decision model is completed.

[0051] In the above system, optionally, the execution unit executes a process of sending task execution instructions to each target robot based on the task allocation strategy, including:

[0052] Based on the task allocation strategy, determining each target robot and determining the task to be performed by each target robot;

[0053] Acquire a global environment map, and use the global environment map to plan a motion path for each of the target robots;

[0054] Based on the motion path and task of each target robot, generating a task execution instruction for each target robot;

[0055] Each of the task execution instructions is sent to the target robot corresponding to it.

[0056] A storage medium includes a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the above-mentioned multi-robot collaborative task dynamic scheduling method.

[0057] An electronic device, comprising:

[0058] one or more processors;

[0059] a memory having one or more programs thereon;

[0060] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for dynamic task scheduling of multi-robot collaboration.

[0061] Compared with the prior art, this application has the following advantages:

[0062] The present application provides a method and related equipment for dynamic scheduling of tasks for multi-robot collaboration, which uses a multi-level graph structure to express the complex relationship between robots, environment and tasks, provides rich and complete relationship information for generating task allocation strategies, improves the accuracy of the generated task allocation strategies, and avoids conflicts when allocating tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0064] Figure 1 A flowchart of a multi-robot collaborative task dynamic scheduling method provided in an embodiment of the present application;

[0065] Figure 2 A flowchart of sending task execution instructions to each target robot based on the task allocation strategy provided in an embodiment of the present application;

[0066] Figure 3 A schematic diagram of the structure of a multi-robot collaborative task dynamic scheduling system provided in an embodiment of the present application;

[0067] Figure 4 A schematic diagram of the structure of another multi-robot collaborative task dynamic scheduling system provided in an embodiment of the present application;

[0068] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0071] At present, for distributed multi-task execution scenarios with multiple robots (such as large-scale industrial assembly line operation tasks), robots are usually not only diverse in type, but also have huge differences in functions and performance. The existing task scheduling system still has the following shortcomings when facing complex, dynamic, and large-scale environments:

[0072] 1. Lack of adaptability: predefined rules and static algorithms are difficult to cope with complex and changing actual operating environments;

[0073] 2. Poor scalability: The computational complexity of centralized algorithms increases dramatically when the number of robots and the scale of tasks increase, making it difficult to respond quickly;

[0074] 3. Insufficient modeling of complex relationships: Existing methods have difficulty in fully expressing and utilizing the complex relationships between robots, tasks, and environments, resulting in poor decision-making results;

[0075] 4. Insufficient long-term planning capabilities: Most methods focus on short-term (one-time) task allocation and lack consideration of long-term benefits;

[0076] 5. Insufficient multi-dimensional collaborative optimization: It is difficult to take into account multiple optimization goals such as efficiency, energy consumption, and load balancing at the same time.

[0077] In order to solve the above problems and the problems in the background technology, the present application provides a multi-robot collaborative task dynamic scheduling method, which can be used in many general or special computing device environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor devices, distributed computing environments including any of the above devices or devices, etc.

[0078] Reference Figure 1 , is a flow chart of a multi-robot collaborative task dynamic scheduling method provided in an embodiment of the present application, and is specifically described as follows:

[0079] S101. Collect task scheduling scenario information, where the task scheduling scenario information includes robot information, environment information, and task information. The robot information includes status information of each robot, and the task information includes information of each task to be executed in the task queue.

[0080] In the method provided in the embodiment of the present application, task scheduling scenario information can be collected in real time, wherein: the robot's status information includes but is not limited to whether the robot is currently executing a task, whether it is in a schedulable state, whether it is charging, whether there is a fault, battery power, location information, load conditions and other information.

[0081] Task information includes, but is not limited to, task type, whether it has been assigned, task content, task priority, estimated execution time, resource requirements, and other information.

[0082] Environmental information includes but is not limited to information on the execution point of the task, obstacle distribution, network communication status, etc.

[0083] S102, applying task scheduling scenario information to construct a multi-level graph structure, and obtaining multi-dimensional input information based on the multi-level graph structure, wherein the multi-level graph structure includes relationship layers of multiple dimensions, and the relationship layers include relationship information between objects of the dimensions corresponding to the relationship layers.

[0084] The multi-level graph structure is used to represent the complex relationships between various objects in the task scheduling scenario, such as tasks, robots, resources, and task execution points.

[0085] The multi-level structure includes relationship layers of multiple dimensions. Different relationship layers correspond to different dimensions and different relationship layers correspond to different objects.

[0086] Preferably, each relationship layer includes at least one relationship layer of a reachability layer, a task compatibility layer, a resource dependency layer, a timing dependency layer, a communication network layer, and an energy distribution layer, wherein:

[0087] The objects corresponding to the reachability layer are the robots and the execution points of the tasks; the reachability layer includes the execution path information of the reachable paths between each robot and the execution points of each task; the execution path information includes the path length and the estimated travel time of the reachable path; preferably, the end point of the reachable path here is the execution point of the task, the starting point is the current position of the robot, and the estimated travel time is the time required for the robot to start from the current position and travel along the reachable path to reach the end point of the reachable path.

[0088] The objects corresponding to the task compatibility layer are robots and tasks; the task compatibility layer includes the task attribute information of the executable task types of each robot; the task attribute information includes the execution efficiency of the robot when executing tasks of the executable task types.

[0089] The objects corresponding to the resource dependency layer are tasks and resources; the resource dependency layer includes the resource dependency information of each task on various resources; the resource dependency information includes the information of various resources required for the execution of each task, specifically including the various resources required for the execution of the task, the quantity of each resource required, etc.; further, the various resources include but are not limited to CPU and memory space.

[0090] The object corresponding to the timing dependency layer is the task; the timing dependency layer includes the timing dependency information of the sequence between each task; the timing dependency information includes the sequence between each task, and the critical path mark of each task and other information.

[0091] The objects corresponding to the communication network layer are robots and task dynamic scheduling systems; the communication network layer includes the communication attribute information of the communication status between each robot and the system; the communication attribute information includes the signal strength and bandwidth of the robot, the signal strength and bandwidth of the system, the signal strength and bandwidth of each robot communicating with the system, etc.; the system here can be understood as a task dynamic scheduling system for multi-robot collaboration.

[0092] The objects corresponding to the energy distribution layer are robots and charging stations; the energy distribution layer includes the charging path information of the reachable paths between the robots that need to be charged and each charging station; the charging path information includes the reachable paths between each robot that needs to be charged and each charging station, as well as the charging efficiency and other information.

[0093] The multi-level graph structure can represent the complex relationships between objects and effectively realize the deep reasoning and representation of multiple graph relationships. After obtaining the content of each relationship layer in the multi-level graph structure, it can be updated according to the information collected subsequently.

[0094] In the solution provided in the embodiment of the present application, a pre-trained graph neural network model can be used to extract multidimensional relationship expression data from a multi-level graph structure, and generate multidimensional input information based on the multidimensional relationship expression data.

[0095] Preferably, the relationship feature information of multiple dimensions is first extracted, and then the relationship feature information of multiple dimensions is converted into multidimensional relationship expression data, wherein the multidimensional relationship expression data includes the complex relationship between the robot, the environment and the task in the multi-level graph structure, and the multidimensional relationship expression data can be a multidimensional array. The multidimensional relationship expression data includes the expression data of the relationship between each object in the multi-level graph structure. In other words, the multidimensional relationship expression data includes data mined from the multi-level graph structure for expressing the relationship between each object. The multidimensional relationship expression data includes, but is not limited to, data for expressing the relationship between the resources on which the task depends when executing, data for expressing the execution order between each task, data for expressing the type of task that the robot can perform, data for expressing the relationship between the power of the robot and each task that the robot can perform, etc., and examples are not given here.

[0096] Exemplarily, for data used to express the relationship between the robot's power and the various tasks that the robot can perform, when expressing the relationship between the robot's power and the time it takes to complete each task, whether the robot can perform the task can be determined by judging whether the robot's power is sufficient to support the robot's operation to the execution point of a task, thereby expressing the tasks that the robot can and cannot perform based on the power of the robot.

[0097] The generated multidimensional input information includes multidimensional relationship expression data, task status list, robot list and time information; the task status list includes the status information of each task to be executed, and the task status list can be represented as an array; the robot list includes all currently available robots. The data type of the multidimensional input information can be a data type that can be used and recognized by the decision model.

[0098] S103: Determine a task allocation strategy based on the multi-dimensional input information.

[0099] In the solution provided in the embodiment of the present application, multi-dimensional input information can be input into a pre-trained decision model to obtain a task allocation strategy output by the decision model after processing the multi-dimensional input information.

[0100] The decision model is built using an agent-based reinforcement learning algorithm.

[0101] The multi-dimensional input information can be used as the state space of the decision model; the multi-dimensional input information is input into the pre-trained decision model, and the decision model processes the multi-dimensional input information to output a task allocation strategy. The task allocation strategy includes the task execution information of the robots involved in the task execution, and the task execution information includes the execution point of the task assigned to the robot and the task executed at the execution point.

[0102] Furthermore, the process of inputting the multi-dimensional input information into a pre-trained decision model and obtaining the task allocation strategy output by the decision model after processing the multi-dimensional input information is as follows: the multi-dimensional input information is input into a pre-trained decision model, and the decision model obtains task requirement information from the multi-dimensional input information, and the task requirement information includes the requirement feature information of the task to be performed; the decision model generates multiple allocation combinations based on the task requirement information and the multi-dimensional input information, and the allocation combination includes selecting each robot participating in the task and assigning tasks to each robot participating in the task; the decision model evaluates the expected return value of each allocation combination, and generates a task allocation strategy based on the allocation combination with the highest expected return value. Furthermore, the pairing schemes of robots and tasks in different allocation combinations are different; illustratively, different robots participating in the execution of tasks form different allocation combinations, for example, allocation combination 1 includes robots A, B, and C, and allocation combination 2 includes robots A, B, and D; for allocation combinations in which the robots participating in the execution of tasks are exactly the same, the tasks assigned to the robots participating in the execution of tasks are different, for example, the robots participating in the execution of tasks in allocation combination 1 and allocation combination 3 are both A, B, and C, wherein the task assigned to robot A participating in the execution of tasks in allocation combination 1 is a, the task assigned to robot B participating in the execution of tasks is b, and the task assigned to robot C participating in the execution of tasks is c; the task assigned to robot A participating in the execution of tasks in allocation combination 3 is b, the task assigned to robot B participating in the execution of tasks is c, and the task assigned to robot C participating in the execution of tasks is a.

[0103] Furthermore, the task requirement information includes requirement characteristic information of each task to be executed, and the requirement characteristic information includes the type of task, resources and time required for task execution, and the like.

[0104] The expected return value can be understood as the score of the allocation combination. The larger the expected return value, the more reasonable the allocation combination is. The task allocation strategy obtained based on the allocation combination can better meet the needs. The task allocation strategy obtained based on the allocation combination is more reasonable and efficient.

[0105] When evaluating the expected return value of each allocation combination, various aspects can be evaluated from the working performance of each robot in the allocation combination, the degree of coordination between the robots, the matching degree between each robot and the task assigned to it, the efficiency of each robot in performing the task assigned to it, etc., so as to comprehensively evaluate the allocation combination and subsequently select a more reasonable allocation combination based on the expected return value.

[0106] S104. Sending a task execution instruction to each target robot based on the task allocation strategy to control each target robot to execute the task corresponding to the task execution instruction.

[0107] According to the task allocation strategy, each target robot is determined, and the target robot is the robot participating in the task execution. A task execution instruction is sent to each target robot, so as to control the target robot to run to the target location and then execute the task. It should be noted that the target location here is the execution point of the task that the robot needs to perform.

[0108] In the method provided in the embodiment of the present application, a multi-level graph structure is used to express the complex relationship between the robot, the environment and the task, and multi-dimensional feature information is extracted from the multi-level graph structure to generate multi-dimensional input information, thereby improving the accuracy of the task allocation strategy, thereby avoiding conflicts when allocating tasks and improving the efficiency of system scheduling.

[0109] In another embodiment provided in the present application, the process of sending task execution instructions to each target robot based on the task allocation strategy is described. Figure 2 , the relevant contents are as follows:

[0110] S201. Based on the task allocation strategy, determine each target robot and determine the task to be performed by each target robot.

[0111] The task allocation strategy includes the identity of the robot participating in the task execution, at least one identity, and each target robot is determined based on each identity. The task allocation strategy also includes a task corresponding to each identity, and the task corresponding to the identity is determined as a task to be executed by the target robot corresponding to the identity.

[0112] S202: Acquire a global environment map, and use the global environment map to plan a motion path for each target robot.

[0113] The global environment map includes obstacle information. The motion path is planned for the target robot in the global environment map based on factors such as the position of the target robot, the positions of other robots, the execution point of the target robot's task to be performed, the robot's power level, and obstacle information.

[0114] When planning a motion path for the target robot, many aspects need to be considered, such as the length of the path that the target robot can travel with its power, to avoid the planned motion path being too long, resulting in the target robot's power being unable to support the target robot's operation to the task execution point; consider whether the estimated time for the target robot to travel from its current position to the task execution point meets the task's time conditions, to avoid the situation where the target robot's time to reach the execution point exceeds the task execution time and cannot effectively execute the task; consider the obstacles that the target robot needs to avoid; consider whether the target robot needs to be charged, and if it needs to be charged, consider the charging time and other aspects.

[0115] In the embodiments provided in the present application, multiple aspects are taken into consideration when planning the motion path of the target robot, and the generated motion path can meet the needs of multiple aspects, so that the target robot can smoothly perform the task, thereby avoiding the situation where the generated motion path cannot meet the needs of multiple aspects due to too single consideration, resulting in the target robot being unable to smoothly perform the task.

[0116] S203: Generate a task execution instruction for each target robot based on the motion path and task of each target robot.

[0117] The task execution instruction includes the motion path and task of the target robot. Furthermore, the task instruction also includes a command for controlling the target robot to move along the motion path and a command for controlling the target robot to execute the task when the target robot reaches an execution point.

[0118] S204: Send each task execution instruction to its corresponding target robot.

[0119] S205 . For each target robot, control the target robot to move according to the motion path in the task execution instruction, and execute the task in the task execution instruction when reaching the execution point in the task execution instruction.

[0120] In the method provided in the embodiment of the present application, after obtaining the task allocation strategy, a motion path is planned for each target robot. When planning the motion path for each target robot, multiple aspects are considered so that the planned motion path can meet multiple needs, take into account multiple planned optimization directions, effectively optimize resource utilization, improve the safety of robot task execution, and promote multiple robots to collaborate to complete multiple tasks, thereby improving the overall task execution performance.

[0121] In another embodiment provided by the present application, the decision model provided by the present application can also be optimized after training is completed. The optimization process is as follows: collecting feedback data of the decision model to generate a task allocation strategy, and using the feedback data to train the decision model to adjust the weight values ​​of various influencing factors of the reward function in the decision model, until the decision model meets the preset end training conditions, the optimization of the decision model is completed. Preferably, the feedback data includes the experience data for generating the task allocation strategy and the feedback information after using the task allocation strategy to control the robot to perform the task.

[0122] Preferably, each evaluation factor in the reward function can be understood as the various directions that the decision model pays attention to when generating the task allocation strategy. Different influencing factors correspond to different directions. The weight value can be understood as priority or attention. The larger the weight value of the influencing factor, the higher its priority or the greater the attention. When generating the task allocation strategy, the focus is on the direction of the influencing factor, so that the generated task allocation strategy can better meet the needs of the evaluation factor. The various influencing factors are task completion, operational feasibility, scheduling stability, energy efficiency and load balancing; illustratively, when the weight value of task completion is the largest, when generating the task allocation strategy, the focus is on the completion of the task, so that when the robot is controlled to perform the task according to the generated task allocation strategy, the final task completion can reach the expected value; for example, when the weight value of load balancing is larger, the load balancing is considered more when generating the task allocation strategy, so that when the robot is controlled to perform the task according to the generated task allocation strategy, the load balancing of the system can reach the expected value.

[0123] When optimizing the decision model, the weight values ​​of various evaluation factors in the reward function of the decision model are adjusted so that when the decision model faces various evaluation factors such as task completion, operational feasibility, scheduling stability, energy efficiency and load balancing, a task allocation strategy is generated according to the weight values ​​of various evaluation factors, so that the generated task allocation strategy fully considers various evaluation factors to meet different task requirements and environmental conditions, so that the task allocation strategy generated by the decision model is more targeted, the generation process is more efficient and accurate, and unnecessary waste of resources and time consumption is reduced. It can also better adapt to changes in the environment and needs, and can generate reliable and accurate task allocation strategies in various environments.

[0124] The present application also provides a multi-robot collaborative task dynamic scheduling system, referring to Figure 3 , is a structural diagram of a multi-robot collaborative task dynamic scheduling system provided in an embodiment of the present application, and the relevant description is as follows:

[0125] A collection unit 301 is used to collect task scheduling scene information, where the scene information includes robot information, environment information, and task information. The robot information includes status information of each robot, and the task information includes information of each task to be assigned.

[0126] A construction unit 302 is used to construct a multi-level graph structure using the task scheduling scenario information, and obtain multi-dimensional input information based on the multi-level graph structure, wherein the multi-level graph structure includes a relationship layer of multiple dimensions, and the relationship layer includes relationship information between objects of the dimension corresponding to the relationship layer, and the object is one of the environment, the task, and the robot;

[0127] A decision unit 303, configured to determine a task allocation strategy based on multi-dimensional input information;

[0128] The execution unit 304 is used to send a task execution instruction to each target robot based on the task allocation strategy, so as to control each target robot to execute the task corresponding to the task execution instruction.

[0129] In another embodiment provided by the present application, the relationship layers of multiple dimensions included in the multi-level graph structure of the system include: at least one of a reachability layer, a task compatibility layer, a resource dependency layer, a timing dependency layer, a communication network layer, and an energy distribution layer;

[0130] The reachability layer includes the execution path information of the reachable paths between each robot and the execution point of each task;

[0131] The task compatibility layer includes task attribute information of executable task types of each robot;

[0132] The resource dependency layer includes the resource dependency information of each task on various resources;

[0133] The timing dependency layer includes the timing dependency information of the order of each task;

[0134] The communication network layer includes the communication attribute information of the communication status between each robot and the system;

[0135] The energy distribution layer includes the charging path information of the reachable paths between the robots that need to be charged and each charging station.

[0136] In another embodiment provided by the present application, the construction unit 302 of the system executes a process of acquiring multi-dimensional input information based on a multi-level graph structure, including:

[0137] Using a pre-trained graph neural network model, multidimensional relationship expression data is extracted from the multi-level graph structure, where the multi-dimensional relationship expression data includes the expression data of the relationship between each object in the multi-level graph structure; and multi-dimensional input information is generated based on the multi-dimensional relationship expression data.

[0138] In another embodiment provided by the present application, the decision unit 303 of the system performs a process of determining a task allocation strategy based on multi-dimensional input information, including:

[0139] The multi-dimensional input information is input into a pre-trained decision model, and the task allocation strategy output by the decision model after processing the multi-dimensional input information is obtained.

[0140] In another embodiment provided by the present application, the decision unit 303 of the system executes a process of inputting multi-dimensional input information into a pre-trained decision model to obtain a task allocation strategy output after the decision model processes the multi-dimensional input information, including:

[0141] The multi-dimensional input information is input into a pre-trained decision model, the decision model obtains task requirement information from the multi-dimensional input information, and generates multiple allocation combinations based on the task requirement information and the multi-dimensional input information, the allocation combination including each robot selected to participate in the task execution and the task assigned to each robot participating in the task execution; the decision model evaluates the expected return value of each allocation combination, and obtains the task allocation strategy generated by the decision model based on the allocation combination with the highest expected return value.

[0142] In another embodiment provided in the present application, the decision model is optimized in the following manner:

[0143] Collect feedback data after the decision model generates a task allocation strategy, and use the feedback data to train the decision model to adjust the weight values ​​of various influencing factors of the reward function in the decision model until the decision model meets the preset end training conditions, completing the optimization of the decision model.

[0144] In another embodiment provided by the present application, the execution unit 304 of the system executes a process of sending task execution instructions to each target robot based on the task allocation strategy, including:

[0145] Based on the task allocation strategy, determine each target robot and determine the task to be performed by each target robot;

[0146] Obtain a global environment map and use the global environment map to plan a motion path for each target robot;

[0147] Based on the motion path and task of each target robot, generate a task execution instruction for each target robot;

[0148] Each task execution instruction is sent to the corresponding target robot.

[0149] The relevant descriptions of the steps performed by each unit provided in this application can be referred to in Figure 1 The description will not be repeated here.

[0150] Further, see Figure 4 , which is a structural diagram of another multi-robot collaborative task dynamic scheduling system provided by an embodiment of the present application. As shown in the figure, the system may include a support layer, a presentation layer, a planning layer and an execution layer. The aforementioned acquisition unit is set in the support layer, the construction unit is set in the presentation layer, the decision unit is set in the planning layer, and the execution unit is set in the execution layer. The specific functions implemented by each layer are described in detail below.

[0151] The support layer has the ability to provide the necessary relevant contextual data foundation and management tasks for upper-level decision-making, specifically including acquisition units and task management modules.

[0152] The collection unit is used to collect task scheduling scene information including robot information, task information and environment information. Further, in the process of collecting data, the collected task scheduling scene information is reported to the construction unit so that the construction unit can construct a multi-level graph structure. The collection unit can adaptively adjust the data reporting frequency according to the degree of dynamic change of the environment (i.e., the fluctuation of the collected data), that is, the collection unit can collect and report in real time, or collect and report in real time at a fixed time, such as collecting data in real time, reporting data according to a dynamic frequency, and the frequency can be adjusted according to data fluctuations; when the data fluctuations are large, the frequency is increased, thereby increasing the number of data uploads and timely updating the multi-level graph structure; when the data fluctuations are small, the frequency is reduced, reducing the number of data reports, avoiding frequent adjustments to the allocation strategy, and affecting the stability of the overall operation of the system. Further, the collection unit can also provide a higher sampling frequency for key nodes (such as robots that are about to complete tasks) so as to update new data in a timely manner, and the unit can also detect and process abnormal values ​​of the collected data to ensure the validity of the data.

[0153] The task management module is used to evaluate the priority of the tasks to be queued, and insert the tasks to be queued into the task queue based on the priority of the tasks to be queued, and cancel the completed tasks when the tasks in the task queue are executed. Further, the task management module is mainly responsible for the life cycle management and priority scheduling of tasks. Specific functions include task deployment, task queue management, priority evaluation and task cancellation. When evaluating the priority of a task, it can be evaluated according to information such as the execution time and task type of the task. The earlier the execution time, the higher the priority. By evaluating the priority of the task, the task can be inserted into the task queue according to the priority of the task, and the task with high priority can be executed first. The tasks in the task queue are arranged in order from high to low according to the priority. For the tasks with high restrictive priority, when the tasks in the task queue have not been executed for a long time, the priority of the task will be updated again, thereby adjusting the order of each task in the task queue, which can avoid the situation where the task has not been executed for a long time, resulting in the failure of the task. Further, the task management module accepts the task instructions and configurations from the user, and performs corresponding task deployment and management in the system. When the task is successfully executed, the task will be cancelled.

[0154] The representation layer is responsible for modeling and representing complex environments, and uses graph neural network models to infer various relationships in the environment to provide richer contextual information for task allocation decisions. The representation layer specifically includes construction units and graph network training modules, and the construction units include graph neural network models.

[0155] The graph network training module is used to train and optimize the graph neural network model in the construction unit. The main functions of this module include: collecting and preprocessing training data, performing model training, and evaluating model performance. By regularly training and fine-tuning the graph neural network model, it ensures that the graph neural network module can accurately represent and analyze the current environmental status, thereby providing reliable input for the decision-making model.

[0156] The construction unit is used to construct a multi-level relationship graph structure, which is used to represent the complex relationship between objects in the environment, and uses a graph neural network for feature extraction and relationship reasoning, thereby obtaining multi-dimensional input information, which includes multi-dimensional relationship expressions and can fully express the multi-dimensional relationship between robots, tasks, and environments. At the same time, the graph neural network model is a graph neural network architecture with a shared backbone network plus multi-task specific heads, which will effectively realize the deep reasoning and representation of multiple graph relationships; the multi-level relationship graph structure constructed by this unit is updated based on the data uploaded by the acquisition unit of the support layer.

[0157] Furthermore, each relationship layer in the multi-level relationship graph structure can refer to the above description, which will not be repeated here.

[0158] The planning layer is the decision-making center of the system. It dynamically generates corresponding task allocation strategies through the decision model built using the reinforcement learning algorithm. The planning layer specifically includes decision-making units and optimization modules.

[0159] The decision unit includes a decision model. The decision unit generates a task allocation strategy for the system based on multi-dimensional input information and the decision model constructed by using an application intelligent agent reinforcement learning algorithm.

[0160] Preferably, the state space in the decision model is composed of four parts: a multidimensional relationship expression generated by a graph neural network model, a task state list, a robot list, and time information; the action space in the decision model is represented by a two-dimensional matrix of length K, where K is the number of all robots participating in the task execution in the scene. For each robot, the execution point of the task to which it needs to be assigned and the task to be performed at that point are specified in the decision action; the reward function in the decision model will comprehensively consider factors such as task completion, operational feasibility, scheduling stability, energy efficiency, and load balancing, and dynamically adjust the weights of each evaluation factor to adapt to different task requirements and environmental conditions, making this solution more flexible and applicable to various scenarios.

[0161] The optimization module is responsible for training, optimizing and updating the decision model used in the system. The main functions of this module include collecting and processing experience data of the interaction between the decision model and the environment. The experience data can also be understood as feedback data for generating task allocation strategies, and based on the collected experience information, the decision model is trained with the help of reinforcement learning algorithms. Through continuous training and iteration, the decision model can gradually improve its decision-making level and adapt to complex and changing task environments. Furthermore, the feedback data includes feedback information after using the task allocation strategy to control each robot to perform tasks. At the same time, this module is also responsible for managing the experience pool, which is used to save experience data, adjust the reward function, and handle the balance between exploration and utilization, thereby improving learning efficiency and model performance.

[0162] The execution layer is responsible for converting the decisions of the planning layer into actual actions, issuing specific task strategies, planning the scheduling paths of each robot, and tracking and evaluating the execution of tasks. The execution layer includes the execution unit and the monitoring module. Furthermore, the execution unit includes the task execution module, the path planning module, and the communication module.

[0163] The task execution module first converts the abstract task allocation strategy output by reinforcement learning into specific task allocation instructions through action mapping, and then calls the path planning module to perform unified global path planning for all robots that need to move in the current task allocation strategy, and sends motion execution instructions to each robot through the communication module to start coordinating task execution.

[0164] The path planning module plans the optimal movement path for the robot, and can simultaneously consider factors such as environmental obstacles and the positions of other robots. Among them, the environmental map builder in the path planning module can directly generate a global map by directly importing the environmental CAD model or other environmental model files, or dynamically build the environmental map through sensors deployed in the environment (such as depth cameras, radars); dynamic path planning can achieve dynamic obstacle avoidance when the robot moves between the execution points of the task through prediction.

[0165] The monitoring module is used to obtain monitoring data in real time. The monitoring data includes various system performance indicators and status information of each executed task. The various system performance indicators include system resource utilization, load balancing, response time, network throughput, etc.; the status information of each executed task includes but is not limited to the progress of task execution, resources occupied by execution, and other information. The monitoring module is responsible for tracking the execution status of tasks in real time. The module identifies system bottlenecks and performance shortcomings through monitoring data, thereby triggering necessary adjustments. For example, when the preset task reallocation conditions are met, each task in the task queue is reallocated. The specific process of the reallocation operation can be referred to in this application. Figure 1The process shown. In terms of real-time task tracking, visual data can be used to determine abnormal situations; at the same time, through the analysis and regression of historical data on task execution status, it is also possible to determine task anomalies and analyze bottleneck processes. In addition to real-time tracking of task execution status, the monitoring module can also perform execution quality assessment, that is, assess the execution quality of tasks. It can evaluate the current task allocation strategy through a comprehensive analysis of multiple indicators such as task completion efficiency and execution abnormality rate, and promptly feed back the results to the planning layer, which can use the results to optimize the decision model.

[0166] Preferably, the communication module is responsible for managing the communication between robots and between the robots and the scheduling system, and also supports real-time monitoring of the status of the communication link.

[0167] In another embodiment provided by the present application, the above system implements a multi-robot collaborative task dynamic scheduling process such as:

[0168] 1. System initialization: load the pre-trained graph neural network model and decision model, and initialize the parameters and data structure of each module;

[0169] 2. Real-time data collection: The collection unit continuously collects task scheduling scenario information including data such as environment, robot and task status, and provides the task scheduling scenario information as the context data of the scenario to the construction unit; further, the collection unit can use an adaptive sampling strategy to dynamically adjust the sampling frequency of the task scheduling scenario information according to the environment;

[0170] 3. Graph model update: The construction unit updates the multi-level graph structure according to the task scheduling scenario information of the latest scenario, and obtains the multi-dimensional relationship expression through reasoning. The multi-dimensional relationship expression is the embedded representation reflecting the current scene relationship state;

[0171] 4. Strategy generation: The decision model generates a task allocation strategy based on the multidimensional relationship expression provided by the building block;

[0172] 5. Task allocation and execution: The task execution module converts the abstract task allocation strategy into a specific task allocation plan, and issues task execution instructions to the corresponding robots to control the robots to execute the assigned tasks.

[0173] After controlling the robot to perform the task, at least one of the following steps is also included: Step a, execution monitoring: the monitoring module tracks the task execution progress and system performance indicators in real time; Step b, dynamic optimization: identify system bottlenecks and anomalies based on the monitoring results, and trigger necessary adjustments, such as task reallocation, etc.; Step c, experience accumulation and learning: store the feedback data during the task execution process into the agent reinforcement learning experience pool, so as to regularly update the reinforcement learning model. It should be noted that steps ac are independent of each other.

[0174] This application implements a general and scalable intelligent task dynamic scheduling system for multi-robot collaboration. The system can effectively solve the task scheduling problem under the conditions of multi-robot, multi-task, and dynamic scene changes, and can achieve rapid migration and application in similar scenes.

[0175] By constructing a multi-level graph relationship structure, the system in this application can more comprehensively model complex working environments, thereby adapting to various different task scenarios; at the same time, the adaptive information sampling strategy enables the system to respond to changes in the task environment more quickly, improve scheduling flexibility, and effectively overcome the problem of insufficient adaptability of rule-based static allocation methods in the prior art; and the multi-level graph structure fully expresses the complex relationship between robots, tasks, and environments, and uses graph neural networks to further deeply explore the potential relationship between various factors, which enables the system to fully express and utilize the complex relationships in the environment, and provides a rich contextual information basis for task allocation decisions, thereby effectively solving the shortcomings of existing methods in insufficient modeling of complex relationships; in addition, in the decision-making process, the system can take into account multiple optimization goals such as efficiency, energy consumption, and load balancing through the design of multi-objective reward functions, thereby maximizing system efficiency.

[0176] An embodiment of the present application further provides a storage medium, comprising a stored program, wherein when the program is running, a device where the storage medium is located is controlled to execute the above-mentioned multi-robot collaborative task dynamic scheduling method.

[0177] The present application also provides an electronic device, whose structural diagram is as follows: Figure 5 As shown, one or more processors 503; a memory 501, on which there are one or more programs 502; when the one or more programs 502 are executed by one or more processors 503, the one or more processors 503 implement the dynamic task scheduling method for multi-robot collaboration as described above.

[0178] Although the present invention depicts operations in a particular order, this should not be understood as requiring that the operations be performed in the particular order shown or in a sequential order.Multitasking and parallel processing may be advantageous under certain circumstances.

[0179] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0180] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards in the relevant regions.

[0181] The specific implementation processes and derivative methods of the above-mentioned embodiments are all within the protection scope of the present invention.

[0182] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0183] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0184] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-robot collaborative task dynamic scheduling method, characterized in that: include: Collecting task scheduling scenario information, wherein the task scheduling scenario information includes robot information, environment information, and task information, wherein the robot information includes status information of each robot, and the task information includes information of each task to be executed in the task queue; Applying the task scheduling scenario information to construct a multi-level graph structure, and obtaining multi-dimensional input information based on the multi-level graph structure, wherein the multi-level graph structure includes relationship layers of multiple dimensions, and the relationship layers include relationship information between various objects of the dimensions corresponding to the relationship layers; Determining a task allocation strategy based on the multi-dimensional input information; Based on the task allocation strategy, a task execution instruction is sent to each target robot to control each target robot to execute the task corresponding to the task execution instruction.

2. The method according to claim 1, characterized in that The multi-dimensional relationship layers included in the multi-level graph structure include at least one of: a reachability layer, a task compatibility layer, a resource dependency layer, a timing dependency layer, a communication network layer, and an energy distribution layer; The reachability layer includes execution path information of reachable paths between each robot and the execution point of each task; The task compatibility layer includes task attribute information of executable task types of each robot; The resource dependency layer includes resource dependency information of each task on various resources; The timing dependency layer includes timing dependency information of the sequence between each task; The communication network layer includes communication attribute information of the communication status between each robot and the system; The energy distribution layer includes charging path information of accessible paths between the robot that needs to be charged and each charging station.

3. The method according to claim 1, characterized in that The acquiring of multi-dimensional input information based on the multi-level graph structure includes: Using a pre-trained graph neural network model, multidimensional relationship expression data is extracted from the multi-level graph structure, wherein the multi-dimensional relationship expression data includes expression data of the relationship between each object in the multi-level graph structure; and multi-dimensional input information is generated based on the multi-dimensional relationship expression data.

4. The method according to claim 1, characterized in that: The determining of the task allocation strategy based on the multi-dimensional input information comprises: The multi-dimensional input information is input into a pre-trained decision model, and a task allocation strategy output by the decision model after processing the multi-dimensional input information is obtained.

5. The method according to claim 4, characterized in that The step of inputting the multi-dimensional input information into a pre-trained decision model and obtaining a task allocation strategy output by the decision model after processing the multi-dimensional input information comprises: The multidimensional input information is input into a pre-trained decision model, the decision model obtains task requirement information from the multidimensional input information, and generates multiple allocation combinations based on the task requirement information and the multidimensional input information, the allocation combination including selecting each robot participating in the task and the task assigned to each robot participating in the task; the decision model evaluates the expected return value of each allocation combination, and obtains the task allocation strategy generated by the decision model based on the allocation combination with the highest expected return value.

6. The method according to claim 4, characterized in that The decision model is optimized in the following way: Collect feedback data after the decision model generates a task allocation strategy, and use the feedback data to train the decision model to adjust the weight values ​​of various influencing factors of the reward function in the decision model, until the decision model meets the preset end training conditions, and the optimization of the decision model is completed.

7. The method according to any one of claims 1 to 6, characterized in that: The sending of task execution instructions to each target robot based on the task allocation strategy includes: Based on the task allocation strategy, determining each target robot and determining the task to be performed by each target robot; Acquire a global environment map, and use the global environment map to plan a motion path for each of the target robots; Based on the motion path and task of each target robot, generating a task execution instruction for each target robot; Each of the task execution instructions is sent to the target robot corresponding to it.

8. A multi-robot collaborative task dynamic scheduling system, characterized in that: include: A collection unit, used to collect task scheduling scene information, the scene information includes robot information, environment information and task information, the robot information includes status information of each robot, and the task information includes information of each task to be assigned; A construction unit, used to construct a multi-level graph structure by applying the task scheduling scenario information, and obtain multi-dimensional input information based on the multi-level graph structure, wherein the multi-level graph structure includes relationship layers of multiple dimensions, and the relationship layers include relationship information between objects of the dimensions corresponding to the relationship layers, wherein the object is one of the environment, the task, and the robot; A decision-making unit, configured to determine a task allocation strategy based on the multi-dimensional input information; The execution unit is used to send a task execution instruction to each target robot based on the task allocation strategy to control each target robot to execute the task corresponding to the task execution instruction.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the multi-robot collaborative task dynamic scheduling method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: one or more processors; a memory having one or more programs thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-robot collaborative task dynamic scheduling method as described in any one of claims 1 to 7.

Images