Robot cluster individual behavior control method for space-time travel task

Abstract

The application relates to a robot cluster individual behavior control method for a space-time travel task, and belongs to the technical field of automatic control. The method comprises the following steps: acquiring the state of an i-th individual in a robot cluster when the individual executes a j-th task, inputting a pre-trained individual behavior control strategy model, outputting a predicted control strategy of the i-th individual executing the j-th task from the individual behavior control strategy model, determining a new control strategy of the i-th individual executing the j-th task according to a preset individual dynamic adjustment strategy, controlling the i-th individual to execute the j-th task according to the new control strategy, and collecting actual action space changes of the i-th individual executing the j-th task. The application provides a mode capable of automatically determining an individual behavior control strategy, the mode can avoid manual intervention, has the advantages of low cost, fast response speed, adaptability to complexity of large-scale cluster cooperation tasks and the like.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, and in particular to a method for controlling the behavior of individual robots in a swarm for spatiotemporal movement tasks. Background Technology

[0002] In the context of modern industrial and intelligent development, spatiotemporal navigation tasks, as a typical collaborative task, are widely found in application scenarios such as robot swarms, drone swarms, autonomous vehicle fleets, and collaborative robots in intelligent manufacturing systems. These tasks require each individual member of the swarm to cooperate with each other in a dynamically changing spatiotemporal environment to complete highly complex task objectives. For example, dynamic path planning by robot swarms on industrial production lines, precise coordinated flight of drone swarms in complex airspace, and even the coordination of automated equipment in large-scale logistics and distribution networks all require individuals to respond to task requirements and coordinate their actions in real time in both time and space.

[0003] During each task, each individual in the cluster performs actions according to the task objective. However, due to various factors such as control precision errors, power limitations, and environmental changes, some individuals in the cluster may fail to fully achieve the expected task results. This deviation may lead to the failure of spatiotemporal coordination of the cluster's overall task, thereby affecting the cluster's task completion efficiency and accuracy.

[0004] To compensate for these shortcomings, traditional methods rely on expert intervention. By analyzing deviations in task execution, adjustments are made to individuals that fail to meet the targets, such as correcting paths, optimizing task allocation, or replanning movement trajectories, to ensure they achieve the intended task results. While this method can guarantee task completion to some extent, it has significant drawbacks, such as high costs of human intervention, slow response times, and difficulty in adapting to the complexity of large-scale cluster collaborative tasks. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for controlling the behavior of individual robots in a swarm for spatiotemporal movement tasks. The technical solution of this invention is as follows:

[0006] A method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks, comprising:

[0007] S1, obtain the state of the i-th individual in the robot cluster when it performs the j-th task.

[0008] S2, will Input a pre-trained individual behavior control strategy model, and the model outputs a predictive control strategy for the i-th individual to perform the j-th task.

[0009] S3, adjusts the strategy dynamically based on preset individual settings and Determine a new control strategy for the i-th individual to perform the j-th task.

[0010] S4, control the i-th individual according to Perform the j-th task and collect the actual action space changes of the i-th individual performing the j-th task.

[0011] Optionally, S2 in the process of... Before inputting the pre-trained individual behavior control strategy model, the following steps are also included:

[0012] S01, Obtain the actual motion space change matrix of the robot cluster when it performs the target task in history;

[0013] S02, Construct the spatial weight vector of the i-th individual when performing the target task in history;

[0014] S03, Construct the task count weight vector for the j-th task in the history of the i-th individual when performing the target task;

[0015] S04, construct a spatiotemporal weight matrix centered on the i-th individual based on the spatial weight vector and the task count weight vector;

[0016] S05, determine the spatiotemporal actual action space change matrix centered on the i-th individual;

[0017] S06, Determine the behavioral trajectory of the i-th individual in the robot swarm when it performs the target task in history, based on the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix;

[0018] S07, Train the individual behavior control strategy model based on the behavior trajectory of the i-th individual until a well-trained individual behavior control strategy model is obtained.

[0019] Optionally, S01 includes:

[0020] S011, It is determined that the robot cluster has a total of n individuals, and each individual has performed the target task m times in the past.

[0021] S012, Determine the actual motion space changes of the i-th individual when performing the j-th task in the past when performing the target task. in This represents the changes in angle and distance of individual i in three-dimensional space;

[0022] S013, construct an actual action space variation matrix of the robot cluster according to the actual action space variation of each individual in the robot cluster when completing each task, expressed as:

[0023]

[0024] wherein, represents the actual action space variation when the i-th individual in the robot cluster historically executes the j-th task of the target task.

[0025] optionally, said S02 comprises:

[0026] S021, determine all individuals whose distance to the i-th individual is less than the Euclidean distance limit value , the number of such individuals is x;

[0027] S022, calculate the spatial position weight of each of the x individuals for the i-th individual according to the Euclidean distance between each of the x individuals and the i-th individual and the sum of the Euclidean distances between them and the i-th individual;

[0028] S023, re-sort and number the x individuals according to the spatial position weight of each individual for the i-th individual to obtain a spatial weight vector, expressed as: and wherein the principle of re-sorting and numbering is: taking the i-th individual as the center, if compared with the i-th individual, the serial number of the l-th individual satisfies l>i, the l-th individual is arranged on one side of the i-th individual, and the larger the spatial position weight is, the closer it is to the i-th individual; if l<i, the l-th individual is arranged on the other side of the i-th individual, and the larger the spatial position weight is, the closer it is to the i-th individual.

[0029] optionally, said S03 comprises:

[0030] S031, determine the number of tasks whose task number difference from the j-th task is less than the task number difference limit value before the i-th individual historically executes the j-th task of the target task , the total number of such tasks is y;

[0031] S032, assign weights to the y tasks to obtain a task number weight vector when the i-th individual historically executes the j-th task of the target task, expressed as:

[0032] optionally, said S04 comprises:

[0033] S041, determine that the weight of the row where the i-th individual is located in the spatio-temporal weight matrix is

[0034] S042, according to Determine the weight of the row containing the x-th individual that is related to the space of the i-th individual;

[0035] S403, combining the weights of the row containing the i-th individual and the weights of the row containing the x-th individual spatially related to the i-th individual, yields the spatiotemporal weight matrix, expressed as:

[0036]

[0037] Optionally, S05 includes:

[0038] Based on the actual motion space change matrix of the robot swarm, the spatiotemporal weight matrix, and the correspondence between the x individuals related to the i-th individual and the previous and subsequent reordering and numbering, the spatiotemporal actual motion space change matrix is ​​determined, and is expressed as:

[0039]

[0040] Where, f(l) = x; l = f -1 (x), where f represents a function that reorders and numbers the elements.

[0041] Optionally, S06 includes:

[0042] S061, the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix are used together as the state of the i-th individual executing the j-th task. Alternatively, the product of the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix can be used as the state of the i-th individual executing the j-th task.

[0043] S062, using the spatiotemporal actual action space change of the i-th individual performing the j-th task in the spatiotemporal actual action space change matrix as the control strategy.

[0044] S063, combining the states and control strategies of the i-th individual during multiple task executions, yields the behavioral trajectory of the i-th individual in the history of the robot swarm when performing the target task, denoted as...

[0045] Optionally, S3 includes:

[0046] S31, obtain Control strategy when the i-th individual performs the (j-1)-th task

[0047] S32, according to and Dynamic adjustment get

[0048] Optionally, S32 includes:

[0049] if equal Then control equal

[0050] if Greater than Then calculate and The difference is used as an adjustment difference and controlled. equal The sum of the adjustment difference;

[0051] if Less than Then calculate and The difference is used as an adjustment difference and controlled. equal The difference between the adjusted difference and the actual difference.

[0052] All of the above-mentioned optional technical solutions can be combined arbitrarily, and the present invention will not provide a detailed description of the structure after each combination.

[0053] By means of the above solution, the beneficial effects of the present invention are as follows:

[0054] By training an individual behavior control strategy model and individual behavior trajectory based on actual data obtained from the historical execution of target tasks by a robot swarm, an individual behavior control strategy model and individual behavior trajectory are obtained. This allows the i-th individual in the robot swarm to obtain its j-th task execution state based on its individual behavior trajectory and input this state into the individual behavior control strategy model. The individual behavior control strategy model outputs a predictive control strategy and dynamically adjusts the predictive control strategy based on a preset individual dynamic adjustment strategy. This provides a way to automatically determine the individual behavior control strategy. This method of determining the individual behavior control strategy can avoid manual intervention and has the advantages of low cost, fast response speed, and adaptability to the complexity of large-scale swarm collaborative tasks.

[0055] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0056] Figure 1 This is a flowchart of the method provided in an embodiment of the present invention.

[0057] Figure 2This is a flowchart of the method for training an individual behavior control strategy model provided in the embodiments of the present invention. Detailed Implementation

[0058] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0059] The robot swarm individual behavior control method for spatiotemporal travel tasks provided in this invention can be implemented using any electronic device with computing capabilities, such as a PC, mobile terminal, or server. The robot swarm in this invention can be a swarm of drones, an autonomous vehicle fleet, or collaborative robots in an intelligent manufacturing system.

[0060] The implementation background of this invention is that the target task executed by the robot cluster is the same each time, and the sub-tasks distributed to each individual in the robot cluster are also the same each time the target task is executed. Assume that the robot cluster has n individuals, and each individual needs to execute m tasks when the robot cluster executes the target task.

[0061] In each task, each individual robot in the swarm performs actions according to the task objective. However, due to various factors such as control precision errors, insufficient power, and environmental changes, individuals in the swarm often fail to fully achieve the expected task results. When this situation occurs in the history of robot swarm execution of target tasks, the individual behavior control method adopted is: individuals that fail to achieve the expected task results are adjusted in a timely manner through expert intervention to achieve the predetermined task effect. This individual behavior control method requires human intervention, which has problems such as high cost, slow response speed, and difficulty in adapting to the complexity of large-scale swarm collaborative tasks.

[0062] To address these issues, this invention employs a method that trains an individual behavior control strategy model based on historical data obtained through manual intervention in a target task. This model then predicts the control strategy for the i-th individual performing the j-th task when the robot ensemble is executing the target task. And through predictive control strategies After dynamic adjustment, it is used to control the i-th individual to perform the j-th task, so that the individual behavior control method does not require manual intervention.

[0063] Based on the above, such as Figure 1 As shown, the robot swarm individual behavior control method for spatiotemporal travel tasks provided in this embodiment of the invention includes the following steps S1 to S4:

[0064] S1, obtain the state of the i-th individual in the robot cluster when it performs the j-th task.

[0065] Specific details The method can be obtained from the behavioral trajectory of the i-th individual when the robot swarm historically performed the target task. The method for determining the behavioral trajectory of the i-th individual, and... The specific details will be explained in the following embodiments, and will not be described here.

[0066] S2, will Input a pre-trained individual behavior control strategy model, and the model outputs a predictive control strategy for the i-th individual to perform the j-th task.

[0067] Specifically, the input of the individual behavior control strategy model is the state of an individual before performing a certain task, and the output is the control strategy predicted by the individual behavior control strategy model when an individual performs a certain task.

[0068] It should be noted that, in the process of... Before inputting a pre-trained individual behavior control strategy model, the individual behavior control strategy model needs to be trained first. In this embodiment of the invention, training the individual behavior control strategy model can be achieved through the following steps S01 to S07:

[0069] S01, obtain the actual motion space change matrix of the robot cluster when it performs the target task in history.

[0070] In one specific embodiment, S01 can be implemented by the following steps S011 to S013:

[0071] S011, it is determined that the robot cluster has a total of n individuals, and each individual has performed the target task m times in the past.

[0072] S012, Determine the actual motion space changes of the i-th individual when performing the j-th task in the past when performing the target task. in The changes in angle and distance of individual i in three-dimensional space are represented as follows: Let the angles before and after the task is completed be r1 and r2, respectively, and the coordinates be (x1, y1, z1) and (x2, y2, z2), respectively. This can be expressed by the formula:

[0073]

[0074] S013, Based on the actual motion space changes of each individual robot in the swarm when completing each task, construct the actual motion space change matrix of the robot swarm, represented as:

[0075]

[0076] in, This represents the actual motion space variation of the i-th individual in the robot swarm when it performs the target task j times in history.

[0077] The success of group collaborative tasks depends on the synergy between each individual and the surrounding environment, as well as other individuals. In such tasks, individuals typically influence each other, and an individual's current state is not only related to the states of surrounding individuals but also to its previous historical states. Therefore, considering the complexity of group behavior, this embodiment of the invention models and analyzes individual behavior from both spatial and temporal perspectives, as embodied in steps S02 and S03 below.

[0078] S02, construct the spatial weight vector of the i-th individual when performing the target task in history.

[0079] Specifically, from a spatial perspective, the relationship between an individual and its surrounding individuals forms the basis for collaboration and task execution. To rationally select which individuals collaborate within a given area, this embodiment of the invention constructs a spatial weight vector. Based on the spatial distance between individuals, the spatial weight vector determines how many neighboring individuals each individual collaborates with to complete the task. For the i-th individual performing the j-th task, assuming its position is (x... i y i , z i In this embodiment of the invention, Euclidean distance is used to calculate the distance between individual i and other individuals. It is assumed that the position of the l-th individual is (x... l y l , z l If ), then the Euclidean distance between individual i and individual l is:

[0080]

[0081] The embodiments of the present invention also pre-set the limit value of Euclidean distance between individuals based on experience. If the Euclidean distance between the l-th individual and the i-th individual Greater than or equal to Therefore, it is assumed that there is no interaction between the l-th individual and the i-th individual. Thus, it can be concluded that the distance to the i-th individual is less than... In this embodiment of the invention, it is assumed that there are x individuals that interact with the i-th individual.

[0082] Based on the above content, said S02 includes, but is not limited to, being implemented through the following steps S021 to S023:

[0083] S021: determining all individuals whose distance to the i-th individual is less than the Euclidean distance limit value , wherein the number of such individuals is x.

[0084] S022: calculating the spatial position weight of each of the x individuals relative to the i-th individual according to the Euclidean distance between each individual in the x individuals and the i-th individual and the sum of the Euclidean distances between them and the i-th individual.

[0085] Specifically, this step is expressed by the formula:

[0086]

[0087] In the formula, represents the Euclidean distance between the l-th individual and the i-th individual among the x individuals screened out according to the Euclidean distance limit value under the j-th task; represents the sum of the Euclidean distances between the x individuals screened out according to the Euclidean distance limit value and the i-th individual under the j-th task, represents the spatial position weight of the l-th individual relative to the i-th individual under the j-th task.

[0088] S023: re-sorting and numbering the x individuals according to the spatial position weight of each individual relative to the i-th individual to obtain a spatial weight vector, which is expressed as: and represents the spatial position weight of the x-th individual relative to the i-th individual after the x individuals screened out according to the Euclidean distance limit value are re-sorted and numbered, represents the spatial weight vector of the i-th individual.

[0089] Wherein, the principle of re-sorting and numbering is as follows: suppose that the x individuals screened out according to the Euclidean distance limit value and the i-th individual are a total of x+1 individuals. For example, the numbers of the x individuals in the robot cluster are [1, 3, …, l, i]. Taking the i-th individual as the center, if the number of the l-th individual is greater than i when compared with that of the i-th individual (l>i), the l-th individual is ranked on one side of the i-th individual, and the greater the spatial position weight is, the closer it is to the i-th individual; if the number of the l-th individual is smaller than i when compared with that of the i-th individual (l<i), the l-th individual is ranked on the other side of the i-th individual, and the greater the spatial position weight is, the closer it is to the i-th individual.

[0090] According to the limit value of Euclidean distance Each of the selected x individuals is sorted in this way, and then the x individuals are reordered from less than i to greater than i, i.e., [1, 2, 3, ..., x]. The change in order can be expressed by the following mathematical relationship:

[0091] f(l) = x; l = f -1 (x);

[0092] Where f represents a function for reordering and numbering, and x represents the number of an individual that was originally numbered l in the robot cluster and is now numbered x after reordering.

[0093] The transformation of function f by reordering and numbering can be completed according to the limit value of Euclidean distance. The selected x individuals are reordered and numbered, and the limit value of Euclidean distance is also known. After the x individuals are selected and reordered, their numbers are ranked in the original population, i.e., a one-to-one correspondence.

[0094] S03, construct the task count weight vector for the j-th task in the history of the i-th individual when performing the target task.

[0095] From the perspective of task frequency, an individual's current task is influenced not only by other individuals in its spatial neighborhood but also by its historical state (and historical tasks). Therefore, this embodiment of the invention constructs a task frequency weight vector to describe the impact of previous tasks on the current task. Using this task frequency weight vector, it is possible to select and analyze how many previous tasks have a significant impact on the current task, thereby better capturing the time dependency in the individual task execution process.

[0096] For the i-th individual and the j-th task, the difference in the number of tasks between the k-th and j-th tasks is jk times (k < j). Let represent the difference in the number of tasks between the i-th individual and the j-th and k-th tasks. The mathematical formula can be expressed as:

[0097] The embodiments of the present invention set a limit value for the difference in the number of tasks based on experience. If the difference in the number of tasks between the k-th task and the j-th task is greater than or equal to Therefore, it is assumed that the k-th task will not affect the j-th task. Thus, it can be concluded that the difference in the number of tasks executed before the j-th task compared to the j-th task is less than [a certain value]. All tasks, assuming there are a total of y tasks less than

[0098] Based on the above, S03 can be achieved through the following steps S031 and S032:

[0099] S031, determine that the difference in the number of tasks performed by the i-th individual before the j-th task in the history of the target task is less than the limit value of the difference in the number of tasks. The number of tasks, y in total; S032, assign weights to the y tasks, and obtain the task count weight vector for the j-th task in the history of the i-th individual, as follows:

[0100] When assigning weights to y tasks, methods such as reciprocal weight normalization, linear decay normalization, or exponential decay normalization can be used to reasonably allocate weights among different tasks according to requirements. Taking the reciprocal weight normalization method as an example, when assigning weights to y tasks, the weights for each task can be assigned as shown in the following formula:

[0101]

[0102] In the formula, Let represent the weight of the i-th individual in the k-th task relative to the j-th task. It is the sum of weights relative to the j-th task under task y, and

[0103] S04. Construct a spatiotemporal weight matrix centered on the i-th individual based on the spatial weight vector and the task count weight vector.

[0104] Specifically, through steps S02 and S03 above, it is determined which individuals the i-th individual collaborates with spatially to complete the j-th task, and how many previous tasks will affect the current task temporally. Step S04 constructs a spatiotemporal weight matrix centered on the i-th individual based on the spatial weight vector and the task count weight vector for the i-th individual performing the j-th task. The spatiotemporal weight matrix is ​​used to express the relative importance of each position.

[0105] In one specific embodiment, S04 can be implemented by the following steps S041 to S043:

[0106] S041, determine the weight of the row containing the i-th individual in the spatiotemporal weight matrix as follows:

[0107] S042, according to Determine the weight of the row containing the x-th individual that is related to the space of the i-th individual;

[0108] S403, combining the weights of the row containing the i-th individual and the weights of the row containing the x-th individual spatially related to the i-th individual, yields the spatiotemporal weight matrix, expressed as:

[0109]

[0110] S05, determine the spatiotemporal actual action space change matrix centered on the i-th individual.

[0111] Specifically, the spatiotemporal actual action space change matrix is ​​the set of spatiotemporal actual action space changes of individuals related to the time and space of the i-th individual performing the j-th task, historically when performing the target task. In a specific embodiment, step S05 can be implemented by determining the spatiotemporal actual action space change matrix based on the actual action space change matrix of the robot cluster, the spatiotemporal weight matrix, and the correspondence between the x individuals related to the i-th individual before and after reordering and numbering, expressed as:

[0112]

[0113] Where, f(l) = x; l = f -1 (x), where f represents a function that reorders and numbers the elements.

[0114] This step determines the spatiotemporal actual motion space change matrix centered on the i-th individual by combining the spatial weight vector and the task number weight vector, based on the fact that the i-th individual in space collaborates with the surrounding x individuals to complete the task, and that the previous y tasks in time will affect the current task.

[0115] S06, determine the behavioral trajectory of the i-th individual in the robot swarm when performing the target task in history, based on the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix.

[0116] The behavioral trajectory of the i-th individual can characterize the actual motion space changes of the i-th individual in each execution of the target task in the history of the robot swarm. In a specific embodiment, S06 can be implemented by the following steps S061 to S063:

[0117] S061, the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix are used together as the state of the i-th individual executing the j-th task. Alternatively, the product of the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix can be used as the state of the i-th individual performing the j-th task.

[0118] Specifically, regarding how to determine the state of the i-th individual performing the j-th task based on the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix. The embodiments of the present invention provide two methods.

[0119] The first approach is to use the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix together as the state of the i-th individual executing the j-th task.

[0120] The weights are flexible and have the same dimension as the spatiotemporal action space change matrix. To retain more original information, this method uses both the spatiotemporal weight matrix and the spatiotemporal action space change matrix as the state of the i-th individual in the j-th task. This approach allows the model to self-learn and determine the combination of the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix during the subsequent training of the individual behavior control strategy model, and may discover some non-intuitive patterns. Secondly, this approach retains all original data and weight information, preventing the loss of some information. However, this approach has limitations, potentially increasing the complexity of model training because the model needs to process more input data, which may increase training difficulty and computational resource consumption; it may also lead to the need to design more complex network structures to effectively integrate these two types of information.

[0121] The second approach is to use the product of the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix as the state of the i-th individual executing the j-th task.

[0122] This method synthesizes the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix into a matrix containing weight-adjusted action information. This weight-adjusted action information matrix is ​​then used as the state of the i-th individual in the j-th task. Compared to the first method, this approach reduces the dimensionality of the model input, thus lowering model complexity. In this method, weights directly affect the corresponding action variables, providing a direct reflection of the weights' moderating effect on actions. However, a drawback is that this approach may result in the loss of some information. Directly multiplying two matrices can amplify or suppress subtle action changes after applying weights, potentially leading to the loss of important information from the original data.

[0123] It should be noted that, in specific implementations, calculations other than multiplication can be performed on the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix as needed, and the calculation result can be used as the state of the i-th individual executing the j-th task. The calculation method will not be described in detail in this embodiment of the invention.

[0124] S062, using the spatiotemporal actual action space change of the i-th individual performing the j-th task in the spatiotemporal actual action space change matrix as the control strategy.

[0125] S063, combining the states and control strategies of the i-th individual during multiple task executions, yields the behavioral trajectory of the i-th individual in the history of the robot swarm when performing the target task, denoted as...

[0126] S07, Train the individual behavior control strategy model based on the behavior trajectory of the i-th individual until a well-trained individual behavior control strategy model is obtained.

[0127] Based on the model learning approach, this embodiment of the invention uses the states in the behavioral trajectory of the i-th individual as samples in supervised learning, and the actions corresponding to the states, i.e., the control policies, as labels in supervised learning. The states are treated as inputs to the individual behavior control policy model, and the outputs of the individual behavior control policy model are treated as actions. Finally, the desired actions are used to teach the machine to establish the correspondence between states and actions. Specifically, the... As input to an individual behavior control strategy model, This is taken as the output of the individual behavior control strategy model. Thus, the machine's output is the control strategy that should be output in this state. How much is it?

[0128] Regarding the specific type of prediction model for individual behavior control strategies, this embodiment of the invention recommends selecting a neural network architecture capable of simultaneously capturing temporal dynamics and spatial structural characteristics. Suitable networks include ConvLSTM and attention mechanism networks (such as Transformers), which can model long-short-term dependencies in the temporal dimension while extracting local and global features in the spatial dimension. Furthermore, this embodiment of the invention can also select a hybrid network architecture, such as combining CNNs to process spatial features and LSTMs or Transformers to process temporal features, thereby better handling complex spatiotemporal interaction tasks. These networks can all effectively learn the spatiotemporal features in complex spatiotemporal tasks, thereby improving the adaptability and accuracy of individual behavior control strategy models.

[0129] For details on the specific methods of training individual behavior control strategy models, please refer to the existing neural network model training methods. This embodiment of the invention will not elaborate on these methods.

[0130] like Figure 2 As shown, it is a flowchart of training an individual behavior control strategy model in the method provided by the embodiment of the present invention.

[0131] After training the individual behavior control strategy model through steps S01 to S07, and obtaining the behavior trajectory of the i-th individual based on the historical execution of the target task by the robot swarm, Afterwards, The model takes a pre-trained individual behavior control strategy as input and outputs a predictive control strategy for the i-th individual performing the j-th task. However, since the current task may be affected by the previous task during the execution of a task, in order to minimize the impact of the previous task on the current task, this embodiment of the invention also includes the following step S3. Dynamic adjustments have been made.

[0132] S3, adjusts the strategy dynamically based on preset individual settings and Determine a new control strategy for the i-th individual to perform the j-th task.

[0133] In one specific embodiment, S3 includes: S31, obtaining Control strategy when the i-th individual performs the (j-1)-th task S32, according to and Dynamic adjustment get

[0134] in, For individual behavior control strategy models, in the input The output predictive control policy for task j-1 is then provided. The actual action space changes when the i-th individual performs the (j-1)-th task.

[0135] Specifically, S32 includes:

[0136] if equal Then control equal

[0137] if Greater than Then calculate and The difference is used as an adjustment difference and controlled. equal The sum of the adjustment difference;

[0138] if Less than Then calculate and The difference is used as an adjustment difference and controlled. equal The difference between the adjusted difference and the actual difference.

[0139] Specifically, this can be expressed as:

[0140] If the individual behavior control strategy model predicts the predictive control strategy for the i-th individual in the (j-1)-th task... Equal to the change in its actual action space Therefore, it is assumed that the actual action space of individual i performing the (j-1)th task has changed to the correct position.

[0141] If the individual behavior control strategy model predicts the predictive control strategy for the i-th individual in the (j-1)-th task... Greater than its actual change in motion space It is then assumed that the actual action space change of individual i during the (j-1)th task is greater than that of the predictive control strategy. To minimize the change in the actual motion space, this embodiment of the invention adds the portion of the motion space that changes less during the execution of the j-th task to the predictive control strategy of the j-th task. superior.

[0142] If the individual behavior control strategy model predicts the predictive control strategy for the i-th individual in the (j-1)-th task... Smaller than its actual change in motion space It is then assumed that the actual action space change of individual i during the (j-1)th task is greater than that of the predictive control strategy. Therefore, when performing the j-th task, this embodiment of the invention reduces the portion of the actual action space with significant variations to the predictive control strategy of the j-th task. superior.

[0143] Through such individual dynamic adjustment strategies and Compare and decide How to dynamically adjust based on previous status information, more closely aligning with the adjustment logic of industrial sites.

[0144] It should be noted that when dynamically adjusting through the above-mentioned individual dynamic adjustment strategy... When j=1, i.e., for the first task, directly take... equal

[0145] S4, control the i-th individual according to Perform the j-th task and collect the actual action space changes of the i-th individual performing the j-th task.

[0146] As new changes in the actual action space are continuously collected, this embodiment of the invention utilizes the previously mentioned individual behavior control learning method to continuously update and expand the state-action pair dataset. Specifically, the newly collected dataset is merged with the existing dataset, and the merged data is used to retrain the individual behavior control policy model. During this process, the individual behavior control policy model is tested on a validation set to ensure error convergence and to continuously optimize its performance. In this way, the database gradually improves, the individual behavior control policy prediction model is trained to handle more diverse data, and its prediction and control effects are enhanced. This automatic learning mechanism ensures that the model evolves gradually over time. Continuously improve its accuracy and reliability in practical applications.

[0147] In summary, the method provided in this embodiment of the invention, after training an individual behavior control strategy prediction model based on historical data, in practical applications, for predicting the state of the i-th individual and the j-th task, uses the method described above to create an individual behavior trajectory to obtain the state of the i-th individual and the j-th task. When this is input into the individual behavior control strategy model, the model will output the predictive control strategy for the i-th individual and the j-th task. Then, based on the preset individual dynamic adjustment strategy, a new control strategy for the i-th individual and the j-th task is obtained, based on the previous actual state. Will By issuing instructions to the i-th individual and waiting for the individual to complete its action, the actual action space changes of the i-th individual in the j-th task are collected to facilitate the subsequent creation of individual behavior trajectories.

[0148] This invention focuses on the individual behavior control of robot swarms for spatiotemporal movement tasks, proposing a method for such control. By recording the spatiotemporal behavior trajectory and adjustment methods of individuals in each task, the machine can mimic the process of expert-led individual behavior adjustments, gradually learning and mastering reasonable individual control strategies. Based on this, the predictive control strategy learned from the individual behavior control strategy model is compared with the actual changes in the action space to optimize the predictive control strategy, achieving dynamic tuning of individuals. This method not only significantly improves the individual response efficiency and task completion accuracy in spatiotemporal movement tasks but also effectively addresses spatiotemporal task changes in complex dynamic environments, providing theoretical support and practical reference for behavior control in collaborative robot tasks.

[0149] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks, characterized in that, include: S1, obtain the state of the i-th individual in the robot cluster when it performs the j-th task. ;in, The spatiotemporal weight matrix is ​​determined based on the spatiotemporal weight matrix centered on the i-th individual and the spatiotemporal actual action space change matrix centered on the i-th individual. The spatiotemporal weight matrix is ​​constructed based on the spatial weight vector of the i-th individual when performing the target task in history and the task number weight vector of the i-th individual when performing the target task for the j-th time in history. The spatiotemporal actual action space change matrix is ​​constructed based on the correspondence between the actual action space change matrix of the robot cluster, the spatiotemporal weight matrix, and the x individuals related to the i-th individual before and after reordering and numbering. S2, will Input a pre-trained individual behavior control strategy model, and the model outputs a predictive control strategy for the i-th individual to perform the j-th task. ; S3, adjusts the strategy dynamically based on preset individual settings and Determine a new control strategy for the i-th individual to perform the j-th task. ; S3 includes: S31, obtain Control strategy when the i-th individual performs the (j-1)-th task ; S32, according to and Dynamic adjustment ,get S32 includes: if equal Then control equal ; if Greater than Then calculate and The difference is used as an adjustment difference and controlled. equal The sum of the adjustment difference; if Less than Then calculate and The difference is used as an adjustment difference and controlled. equal The difference between the adjusted difference and the actual difference; S4, control the i-th individual according to Perform the j-th task and collect the actual action space changes of the i-th individual performing the j-th task.

2. The method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks according to claim 1, characterized in that, The S2 is in Before inputting the pre-trained individual behavior control strategy model, the following steps are also included: S01, Obtain the actual motion space change matrix of the robot cluster when it performs the target task in history; S02, Construct the spatial weight vector of the i-th individual when performing the target task in history; S03, Construct the task count weight vector for the j-th task in the history of the i-th individual when performing the target task; S04, construct a spatiotemporal weight matrix centered on the i-th individual based on the spatial weight vector and the task count weight vector; S05, determine the spatiotemporal actual action space change matrix centered on the i-th individual; S06, Determine the behavioral trajectory of the i-th individual in the robot swarm when it performs the target task in history, based on the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix; S07, Train the individual behavior control strategy model based on the behavior trajectory of the i-th individual until a well-trained individual behavior control strategy model is obtained.

3. The method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks according to claim 2, characterized in that, S01 includes: S011, It is determined that the robot cluster has a total of n individuals, and each individual has performed the target task m times in the past. S012, Determine the actual motion space changes of the i-th individual when performing the j-th task in the past when performing the target task. ,in This represents the changes in angle and distance of individual i in three-dimensional space; S013, Based on the actual motion space changes of each individual robot in the swarm when completing each task, construct the actual motion space change matrix of the robot swarm, represented as: ； in, This represents the actual motion space variation of the i-th individual in the robot swarm when it performs the target task j times in history.

4. The method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks according to claim 2 or 3, characterized in that, S02 includes: S021, determine that the distance between the i-th individual and the i-th individual is less than the Euclidean distance limit. All individuals, and the number of individuals is x; S022, calculate the spatial position weight of each of the x individuals relative to the i-th individual based on the sum of the Euclidean distance between each individual and the i-th individual and the sum of their Euclidean distances to the i-th individual; S023, re-sorting and numbering x individuals according to the spatial position weight of each individual relative to the i-th individual, to obtain a spatial weight vector, expressed as: and ; wherein, the principle of re-sorting and numbering is: taking the i-th individual as the center, if the l-th individual is compared with the i-th individual and l>i, the l-th individual is arranged on one side of the i-th individual, and the spatial position weight is larger, the closer it is to the i-th individual; if l<i, the l-th individual is arranged on the other side of the i-th individual, and the spatial position weight is larger, the closer it is to the i-th individual.

5. The method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks according to claim 4, characterized in that, S03 includes: S031, determine that the difference in the number of tasks performed by the i-th individual before the j-th task in the history of the target task is less than the limit value of the difference in the number of tasks. The number of times the task is performed, and the total number of times the task is performed is y. S032 assigns weights to tasks y, obtaining the task count weight vector for the j-th task in the history of the i-th individual, as follows: .

6. The method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks according to claim 5, characterized in that, S04 includes: S041, determine the weight of the row containing the i-th individual in the spatiotemporal weight matrix as follows: ; S042, according to Determine the weight of the row containing the x-th individual that is related to the space of the i-th individual; S403, combining the weights of the row containing the i-th individual and the weights of the row containing the x-th individual spatially related to the i-th individual, yields the spatiotemporal weight matrix, expressed as: 。 7. The method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks according to claim 6, characterized in that, S05 includes: Based on the actual motion space change matrix of the robot swarm, the spatiotemporal weight matrix, and the correspondence between the x individuals related to the i-th individual and the previous and subsequent reordering and numbering, the spatiotemporal actual motion space change matrix is ​​determined, and is expressed as: ； in, , f A function that represents reordering and numbering.

8. The method for controlling the behavior of individual robots in a swarm for spatiotemporal travel tasks according to claim 7, characterized in that, S06 includes: S061, the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix are used together as the state of the i-th individual executing the j-th task. Alternatively, the product of the spatiotemporal weight matrix and the spatiotemporal actual action space change matrix can be used as the state of the i-th individual executing the j-th task. S062, using the spatiotemporal actual action space change of the i-th individual performing the j-th task in the spatiotemporal actual action space change matrix as the control strategy. ; S063, combining the states and control strategies of the i-th individual during multiple task executions, yields the behavioral trajectory of the i-th individual in the history of the robot swarm when performing the target task, denoted as... .

Images