A method and apparatus for training a robot control model

Abstract

This application provides a method and apparatus for training a robot control model. The method includes: confirming that a first training frequency identifier carried by a target node meets preset conditions; determining a second training frequency identifier corresponding to the target node; updating the second private data priority corresponding to the current training round based on the robot training data reported by each node; and sending the model parameters, the second private data priority, and the second training frequency identifier of the completed federated learning training round to the target node. The robot control model training method and apparatus provided by this invention evaluate the contribution of each participating node from two dimensions: the privacy and scarcity of training data, and control the frequency of each node's participation in federated learning training based on its contribution. This also improves the data consistency of the global model being trained, thereby avoiding situations such as decreased model performance, reduced training efficiency, and increased risk of overfitting.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method and apparatus for training a robot control model. Background Technology

[0002] Currently, in the industrial internet system, enterprises are limited by the availability of data samples when controlling various robots on production lines using artificial intelligence models. This makes it difficult to effectively train the robot control models, a problem particularly severe for small and medium-sized enterprises (SMEs). With the development of federated learning technology, SMEs can collaborate by uploading their respective training data to a federated learning aggregation server for training. This solves the problem of insufficient training samples while ensuring the privacy of each enterprise's data.

[0003] However, as the federated learning training mode of the aforementioned robot control model operates, uneven data volume and inconsistent data quality among various enterprises become apparent. For example, some enterprises neglect to upload their own data and only rely on the training results distributed by the federated learning aggregation server, while others upload low-quality data such as daily robot operation logs and maintenance records, thus providing minimal assistance to the federated learning training. This uneven distribution of training data, a common problem in federated learning, can lead to various adverse consequences, such as decreased model performance, reduced training efficiency, and increased risk of overfitting. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, electronic device, and storage medium for training a robot control model.

[0005] In a first aspect, embodiments of the present invention provide a method for training a robot control model, the method comprising:

[0006] Confirm that the first training frequency identifier carried by the target node meets the preset conditions;

[0007] The second training frequency identifier corresponding to the target node is determined based on the privacy weight and the first privacy data priority of the robot training data reported by the target node.

[0008] Update the priority of the second private data corresponding to the current training round based on the robot training data reported by each node;

[0009] The model parameters, the priority of the second private data, and the second training frequency identifier for the current round of federated learning training are sent to the target node.

[0010] Optionally, the confirmation that the first training frequency identifier carried by the target node meets preset conditions specifically includes:

[0011] The current time is compared with the time in the first training frequency identifier. If the time in the first training frequency identifier is later, then the preset condition is met.

[0012] Optionally, updating the priority of the second private data corresponding to the current training round based on the robot training data reported by each node specifically includes:

[0013] Extract the data types from the robot training data collected in the current training round;

[0014] Calculate the aggregation degree of the robot training data based on the data type and group the training data accordingly;

[0015] Based on the aggregation degree of each training data group, different secondary private data priorities are assigned to different data types in each training data group.

[0016] Optionally, determining the second training frequency identifier corresponding to the target node based on the privacy weight of the robot training data reported by the target node and the first privacy data priority specifically includes:

[0017] The privacy level of each training data sample in the training data is determined, and then the privacy weight of the training data is calculated.

[0018] The robot training data is divided into a private data sample set and a non-private data sample set, and a first private data priority of the robot training data is calculated based on the private data sample set.

[0019] The second training frequency identifier corresponding to the target node is calculated based on the privacy weight and the priority of the first privacy data.

[0020] Optionally, the private data sample set includes the robot's location data, motion trajectory data, working environment data, and task data.

[0021] Optionally, the non-private data sample set includes basic control parameters, equipment status data, public environment data, and non-task-specific data.

[0022] Optionally, the privacy weight of the robot training data is determined based on the quantity and type of the training data.

[0023] Secondly, embodiments of the present invention provide a training device for a robot control model, the device comprising:

[0024] The training admission judgment module is used to confirm that the first training frequency identifier carried by the target node meets the preset conditions.

[0025] The training contribution determination module is used to determine the second training frequency identifier corresponding to the target node based on the private weight of the robot training data reported by the target node and the first private data priority.

[0026] The data priority calculation module is used to update the second private data priority corresponding to the current training round based on the robot training data reported by each node.

[0027] The parameter distribution module is used to distribute the model parameters, the priority of the second private data, and the second training frequency identifier to the target node after completing the current round of federated learning training.

[0028] Thirdly, embodiments of the present invention provide an electronic device, comprising:

[0029] One or more processors;

[0030] Memory, used to store one or more programs;

[0031] Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in the first aspect.

[0032] Fourthly, embodiments of the present invention provide a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method as described in the first aspect.

[0033] In the robot control model training method, apparatus, electronic device, and storage medium provided in this embodiment of the invention, the aggregation server evaluates the contribution of the training data reported by each participating node to the global model training from two dimensions: the privacy and scarcity of the training data, in each federated learning training round. Based on the contribution, it controls the frequency of each node's participation in federated learning training and provides guidance for each node to collect and report data in the next round. This effectively ensures the fairness of the federated learning training process for each participating party and improves the data consistency of the global model being trained, thereby avoiding situations such as model performance degradation, reduced training efficiency, and increased risk of overfitting. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly described below.

[0035] Figure 1 A flowchart illustrating the training method for a robot control model provided in an embodiment of the present invention;

[0036] Figure 2 This is a flowchart illustrating the training frequency calculation method provided in an embodiment of the present invention.

[0037] Figure 3 A flowchart illustrating the method for determining the priority of private data provided in an embodiment of the present invention;

[0038] Figure 4 A schematic diagram of the structure of a training device for a robot control model provided in an embodiment of the present invention;

[0039] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0040] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0041] Similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0042] Currently, in the industrial internet system, enterprises are limited by the availability of data samples when controlling various robots on production lines using artificial intelligence models. This makes it difficult to effectively train the robot control models, a problem particularly severe for small and medium-sized enterprises (SMEs). With the development of federated learning technology, SMEs can collaborate by uploading their respective training data to a federated learning aggregation server for training. This solves the problem of insufficient training samples while ensuring the privacy of each enterprise's data.

[0043] However, as the federated learning training mode of the aforementioned robot control model operates, uneven data volume and inconsistent data quality among various enterprises become apparent. For example, some enterprises neglect to upload their own data and only rely on the training results distributed by the federated learning aggregation server, while others upload low-quality data such as daily robot operation logs and maintenance records, thus providing minimal assistance to the federated learning training. This uneven distribution of training data, a common problem in federated learning, can lead to various adverse consequences, such as decreased model performance, reduced training efficiency, and increased risk of overfitting.

[0044] Based on this, embodiments of the present invention provide a method for training a robot control model. (Appendix) Figure 1 A flowchart illustrating the training method for a robot control model provided in an embodiment of the present invention is shown.

[0045] Step S110: Confirm that the first training frequency identifier carried by the target node meets the preset conditions.

[0046] In this embodiment of the invention, the execution entity for each step is the aggregation server within the federated learning platform. The aggregation server is primarily responsible for collecting, aggregating, and distributing model updates in federated learning. Specifically, it allows each participating node (in this embodiment, various industrial internet companies or other registered entities) to periodically share model updates trained locally on their datasets while retaining their own locally. These updates are then aggregated by the aggregation server to generate a global model. The aggregation server in the federated learning platform possesses characteristics such as protecting data privacy, improving training efficiency, and supporting scalability and fault tolerance.

[0047] In this embodiment of the invention, the aggregation server trains the federated learning model according to a preset number of rounds. During each round of training, the aggregation server performs centralized training based on the training data reported by each node, and then distributes the updated global model back to each node. This completes one round of training. Specifically, in this step, since different nodes contribute differently to the federated learning training data, the aggregation server assigns different training frequencies to different nodes. In this embodiment, data contribution refers to the quantity and quality of robot training data provided by a node for federated learning. When a node's data contribution is too low, the aggregation server assigns it a relatively low training frequency, meaning that the node cannot participate in the following federated learning training rounds and cannot obtain the parameter update results of the global model.

[0048] In this embodiment of the invention, the training frequency assigned to a node is calculated based on its data contribution and is represented by a training frequency identifier. The first training frequency identifier in this step is determined and communicated by the aggregation server when the target node participates in a previous federated learning training round. That is, before each target node participates in this round of federated learning training, the aggregation server checks the first training frequency identifier carried by the target node, and only if the condition is met is the target node allowed to participate in this training.

[0049] Specifically, confirming that the first training frequency identifier carried by the target node meets preset conditions can be achieved as follows: compare the current time with the time in the first training frequency identifier. If the time in the first training frequency identifier is later, then the preset admission condition is met. That is, for a target node to participate in federated learning training, its current time must have reached the target time set by the aggregation server. This step prevents low-contribution nodes that do not meet the admission conditions from participating in federated learning training rounds prematurely.

[0050] Step S120: Determine the second training frequency identifier corresponding to the target node based on the privacy weight of the robot training data reported by the target node and the priority of the first privacy data.

[0051] Specifically, as shown in the attached document Figure 2 As shown, the training frequency calculation method in step S120 can be implemented through the following steps S121 to S123.

[0052] Step S121: Determine the privacy level of each training data sample in the training data, and then calculate the privacy weight of the training data.

[0053] Step S122: Divide the robot training data into a private data sample set and a non-private data sample set, and calculate the first private data priority of the robot training data based on the private data sample set;

[0054] Step S123: Calculate the second training frequency identifier corresponding to the target node based on the privacy weight and the first privacy data priority.

[0055] Specifically, in the context of federated learning in this embodiment of the invention, robot control parameter data can be categorized into private and non-private data based on factors such as the likelihood of privacy breaches, data importance, and scarcity. For robot control parameter data, the following types of data are typically considered sensitive data: location data, i.e., the specific geographical location or spatial coordinates of the robot's work location; motion trajectory data, i.e., motion parameters such as the robot's path, speed, and acceleration; working environment data, i.e., environmental information such as temperature, humidity, and lighting; and task-related data, i.e., the specific task type, task parameters, and task results performed by the robot. The following types of data are typically considered non-private data: basic control parameters, such as the robot's movement speed, acceleration, and turning angle; device status data, such as the robot's power status, network connection status, and sensor status; public environment data, such as environmental parameters like temperature and humidity in public areas; and non-task-specific data, such as the robot's daily operation logs and maintenance records.

[0056] It is worth noting that the above-mentioned private and non-private data are only examples for illustration. The specific data included in actual applications is determined according to the corresponding scenario of the robot application.

[0057] Furthermore, in this embodiment of the invention, the training data also needs to be differentiated into different privacy levels according to the requirements of model training in federated learning. For example, location data and motion trajectory data have the highest privacy level, which can be set to 1.0; non-task-specific data have the lowest privacy level, which can be set to 0.1. The above is only an example; in actual applications, how the aggregation server sets the privacy level according to the type of training data is determined based on the corresponding scenario of the robot application. For a specific model, once the pre-setting is completed before federated learning training, the privacy level of each relevant training data type will not change.

[0058] In this embodiment of the invention, the training data uploaded by the target node in a federated learning training round contains multiple training data samples, each with its corresponding privacy level. By combining the privacy levels of these training data samples, the privacy weight of the entire training data uploaded by the target node in that training round can be determined. The privacy weight of the robot training data is determined based on the quantity and data type of the training data, and can be calculated using the following formula:

[0059]

[0060] Where W represents the private weights of the entire training data, n represents the number of training data samples, and w i Let s represent the private weight of the i-th data sample. i This represents a certain importance indicator of the i-th data sample, such as sample size or information content.

[0061] Furthermore, for the entire training data, in addition to distinguishing the privacy level of each sample, it is also necessary to determine the priority of private data. In this embodiment of the invention, the priority of private data is set to assess the scarcity of training data. The priorities of different training data are dynamically determined by the aggregation server based on the training data collected from all nodes. Generally, non-private data is easier for the aggregation server to obtain and is usually uniformly set to the lowest priority; only private data has different priorities.

[0062] Therefore, in this embodiment of the invention, after the robot receives the robot training data uploaded by the target node, it first distinguishes the data into a private data sample set and a non-private data sample set, and then extracts the private data sample set. Therefore, it is necessary to calculate the first private data priority of the robot training data based on the private data sample set. In this embodiment of the invention, the first private data priority is used to evaluate the priority of the robot training data uploaded by the target node in this round of federated learning training. A specific calculation method can be considered as follows:

[0063]

[0064] Where P represents the first priority result for private data, n represents the number of samples in the private data sample set, and p i t represents the priority data of the i-th private data sample. i Let represent the weight of the i-th private data sample. This weight can be determined based on factors such as the importance of the sample, its information content, and the sample size. Here, can be understood as the specific priority p of each sample. i The values ​​were calculated and determined by the aggregation server during the previous round of federated learning training. Accordingly, during the current round of federated learning training, the aggregation server will calculate and determine the priority values ​​of training samples of different data types in the next round.

[0065] After determining the privacy weight and first privacy data priority of the robot training data reported by the target node, this embodiment of the invention will calculate the second training frequency identifier corresponding to the target node based on the privacy weight and the first privacy data priority. That is, the time when the target node can participate in the next federated learning training is determined based on the above two aspects of the target node's contribution to the current round of federated learning training.

[0066] Specifically, to determine the second training frequency identifier of the target node based on the contributions of both private weights and private data priorities, and to ensure that the larger the private weights and private data priorities are, the smaller the second training frequency identifier is, thus shortening the time available for the next federated learning training, the following formula can be used:

[0067] f = f max -(α·w+β·p)

[0068] Where w represents the privacy weight, p represents the first privacy data priority, f represents the second training frequency identifier, and α and β are weighting coefficients that adjust the influence of the privacy weight and privacy data priority, which can be adjusted according to specific needs. max This is the maximum possible value of the second training frequency identifier. α and β can be used to balance the influence of private weights and private data priority. For example, if we consider private weights more important, we can increase the value of α; if we consider private data priority more important, we can increase the value of β. To ensure that the value of f is always within a reasonable range (e.g., non-negative), some boundary conditions can be added. For example:

[0069] f = max(0, f max -(α·w+β·p))

[0070] Furthermore, if the numerical ranges of privacy weights and privacy data priorities differ significantly, they can be normalized to make their contributions to the final result more balanced. For example: suppose wnorm and p norm These are the normalized privacy weight and the priority of the first private data, respectively. Therefore, the formula can be further modified as follows:

[0071] f = max(0, f max -(α·w norm +β·p norm ))

[0072] Normalization can be achieved in the following ways:

[0073]

[0074] Among them, w min and w max These are the minimum and maximum values ​​of the privacy weight, p. min and p max These are the minimum and maximum values ​​of the first private data priority. In this way, the formula can dynamically adjust the second training frequency identifier based on changes in the privacy weight and private data priority, while ensuring the reasonableness and effectiveness of the results. In summary, the second training frequency identifier calculated here objectively evaluates the contribution of the robot training data uploaded by the target node in this session through two dimensions: privacy and scarcity, providing a basis for rationally allocating federated learning resources to the target node in the next training session.

[0075] Step S130: Update the priority of the second private data corresponding to the current training round based on the robot training data reported by each node.

[0076] As mentioned earlier, the private data priority in this embodiment of the invention is an indicator designed by the aggregation server to evaluate the scarcity of training data. The private data priority is not only used by the aggregation server to evaluate the scarcity of training data uploaded by each node, but is also dynamically updated by the aggregation server in each round based on the training data collected from all nodes, and is sent to each node at the end of each round of training to guide each node to improve the quality of the robot training data uploaded next time.

[0077] The first priority of private data is determined by the aggregation server during the previous round of federated learning training. Correspondingly, during the current round of federated learning training, the aggregation server will calculate and determine the second priority of private data for training samples of different data types in the next round.

[0078] Specifically, such as Figure 3 As shown, the method for determining the priority of private data in step S130 can be implemented according to steps S131 to S133.

[0079] Step S131: Extract the data types from the robot training data collected in the current training round.

[0080] Step S132: Calculate the aggregation degree of the robot training data based on the data type and group the training data.

[0081] Step S133: Based on the aggregation degree of each training data group, assign different second private data priorities to different data types in each training data group.

[0082] Specifically, for the robot training data collected from various nodes in the current round, the aggregation server first needs to clean the collected data, removing duplicate, invalid, or abnormal data. Then, the cleaned data is standardized to ensure comparability between data from different sources. Next, the training data can be grouped according to the types of robot training data mentioned throughout the text. After grouping, the aggregation server calculates the aggregation degree of each group, which can be achieved by calculating indicators such as data similarity, consistency, or correlation. Each group is then sorted according to the calculated aggregation degree, with groups of data with higher aggregation degrees appearing first. Data with high aggregation degrees indicates high similarity or consistency among them, suggesting a relatively rich data source, which in this embodiment manifests as low scarcity. This type of data is marked with a lower second-private data priority value. Conversely, training data types in groups with lower aggregation degrees have higher data scarcity and are marked with a higher second-private data priority value. The specific value assigned can be set according to the user's actual needs.

[0083] As described in this step, the aggregation server dynamically updates the priority values ​​of different private data in each federated learning training round, namely the second private data priority in this step, as a reflection of the scarcity of the current training data. In practice, the second private data priority can be calculated directly using the robot training data of the current round; alternatively, after calculating the scarcity of the robot training data for the current round, the calculation result can be updated to the priority values ​​of the historical total training data in the form of weight adjustments.

[0084] Step S140: The model parameters, the priority of the second private data, and the second training frequency identifier of the completed current round of federated learning training are sent to the target node.

[0085] After step S130 is completed, the aggregation server can perform federated learning training for the current round. This embodiment of the invention does not limit the specific model type used for the robot control model. For example, a federated deep learning model can be used to train the robot control model. Such models typically use deep neural networks as function approximators and are trained collaboratively among multiple robots or data holders through federated learning. For example, convolutional neural networks are suitable for processing image data, such as target detection and recognition in robot vision tasks, and can be trained collaboratively among multiple robots through federated learning, improving the model's accuracy and generalization ability. Another example is recurrent neural networks (RNNs) or long short-term memory networks (LSTMs), which are suitable for processing sequence data, such as robot trajectory prediction and control. Federated learning allows multiple robots to share information in the sequence data, thereby improving the model's prediction and control performance. The specific model type used should be determined based on the robot's application scenario in the Industrial Internet and negotiated by the participating parties in the federated learning process.

[0086] The trained model parameters will be distributed normally to the nodes participating in this round of federated learning. Furthermore, unlike the typical federated learning process, this embodiment of the invention will also distribute a second private data priority and a second training frequency identifier to the target node.

[0087] The second private data priority is a scarcity index calculated by the aggregation server for different types of robot training data. This index is the same for each node and guides each node on how to increase its contribution when reporting training data in the next round, thus avoiding situations where it cannot participate in federated learning training at a high frequency. The second training frequency identifier is an index that directly evaluates the target node's current contribution to federated learning training. This index is different for each node and indicates the round in which the target node can participate in federated learning training. The target node can refer to the second training frequency identifier to implement targeted data collection of different types within a specific time. It can be understood that each participating node in the federated learning training collects and reports training data according to the two indicators of the second private data priority and the second training frequency identifier it receives from the target node. This effectively ensures the fairness of the federated learning training process for all participants and improves the data consistency of the global model being trained, thereby avoiding situations such as model performance degradation, reduced training efficiency, and increased risk of overfitting.

[0088] In the robot control model training method provided in this embodiment of the invention, the aggregation server evaluates the contribution of the training data reported by each participating node to the global model training from two dimensions: privacy and scarcity of the training data in each federated learning training round. Based on the contribution, the frequency of each node's participation in federated learning training is controlled, and guidance is given to each node to collect and report data in the next round. This effectively ensures the fairness of the federated learning training process for each participating party and improves the data consistency of the global model being trained, thereby avoiding situations such as model performance degradation, reduced training efficiency, and increased risk of overfitting.

[0089] Based on any of the above embodiments, the appendix Figure 4 A schematic diagram of the structure of the training device for the robot control model provided in an embodiment of the present invention is shown, the details of which are as follows:

[0090] The training admission judgment module 410 is used to confirm that the first training frequency identifier carried by the target node meets the preset conditions.

[0091] The training contribution determination module 420 is used to determine the second training frequency identifier corresponding to the target node based on the private weight of the robot training data reported by the target node and the first private data priority.

[0092] The data priority calculation module 430 is used to update the second private data priority corresponding to the current training round based on the robot training data reported by each node.

[0093] The parameter distribution module 440 is used to distribute the model parameters, the second private data priority, and the second training frequency identifier to the target node after completing the current round of federated learning training.

[0094] In the robot control model training device provided in this embodiment of the invention, the aggregation server evaluates the contribution of the training data reported by each participating node to the global model training from two dimensions: privacy and scarcity of the training data, in each federated learning training round. Based on the contribution, it controls the frequency of each node's participation in federated learning training and provides guidance for each node to collect and report data in the next round. This effectively ensures the fairness of the federated learning training process for each participating party and improves the data consistency of the global model being trained, thereby avoiding situations such as model performance degradation, reduced training efficiency, and increased risk of overfitting.

[0095] Based on any of the above embodiments, the appendix Figure 4The diagram illustrates the physical structure of an electronic device according to an embodiment of the present invention. This electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540. The processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions stored in the memory 530 to execute the following method:

[0096] Confirm that the first training frequency identifier carried by the target node meets the preset conditions;

[0097] The second training frequency identifier corresponding to the target node is determined based on the privacy weight and the first privacy data priority of the robot training data reported by the target node.

[0098] Update the priority of the second private data corresponding to the current training round based on the robot training data reported by each node;

[0099] The model parameters, the priority of the second private data, and the second training frequency identifier for the current round of federated learning training are sent to the target node.

[0100] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment of the invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this embodiment of the invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] On the other hand, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments, including, for example:

[0102] Confirm that the first training frequency identifier carried by the target node meets the preset conditions;

[0103] The second training frequency identifier corresponding to the target node is determined based on the privacy weight and the first privacy data priority of the robot training data reported by the target node.

[0104] Update the priority of the second private data corresponding to the current training round based on the robot training data reported by each node;

[0105] The model parameters, the priority of the second private data, and the second training frequency identifier for the current round of federated learning training are sent to the target node.

[0106] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0107] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for training a robot control model, characterized in that, The method includes: Confirm that the first training frequency identifier carried by the target node meets the preset conditions; The second training frequency identifier corresponding to the target node is determined based on the privacy weight and the first privacy data priority of the robot training data reported by the target node. Update the priority of the second private data corresponding to the current training round based on the robot training data reported by each node; The model parameters, the priority of the second private data, and the second training frequency identifier that have been completed for the current round of federated learning training are sent to the target node. The step of determining the second training frequency identifier corresponding to the target node based on the privacy weight of the robot training data reported by the target node and the priority of the first privacy data specifically includes: The privacy level of each training data sample in the training data is determined, and then the privacy weight of the training data is calculated. The robot training data is divided into a private data sample set and a non-private data sample set, and a first private data priority of the robot training data is calculated based on the private data sample set. The second training frequency identifier corresponding to the target node is calculated based on the privacy weight and the first privacy data priority; The privacy weight of the robot training data is determined based on the quantity and data type of the training data; The specific calculation method for the priority of the first private data is as follows: in, P This indicates the result with the highest priority for private data. n This indicates the number of samples in the private data sample set. Indicates the first i Priority data for a sample of private data. Indicates the first i Weights of private data samples; The second training frequency identifier uses the following formula: f = f max -( α • w + β • p ) in, w Indicates privacy weight. p This indicates the priority of the first private data. f Indicates the second training frequency identifier. α and β This involves adjusting the weighting coefficients that influence privacy weight and privacy data priority; these should be adjusted according to specific needs. f max It is the maximum possible value of the second training frequency identifier.

2. The training method for the robot control model according to claim 1, characterized in that, The confirmation that the first training frequency identifier carried by the target node meets the preset conditions specifically includes: The current time is compared with the time in the first training frequency identifier. If the time in the first training frequency identifier is later, then the preset condition is met.

3. The training method for the robot control model according to claim 1, characterized in that, The step of updating the priority of the second private data corresponding to the current training round based on the robot training data reported by each node specifically includes: Extract the data types from the robot training data collected in the current training round; Calculate the aggregation degree of the robot training data based on the data type and group the training data accordingly; Based on the aggregation degree of each training data group, different secondary private data priorities are assigned to different data types in each training data group.

4. The training method for the robot control model according to claim 1, characterized in that, The private data sample set includes robot location data, motion trajectory data, working environment data, and task data.

5. The training method for the robot control model according to claim 1, characterized in that, The non-private data sample set includes basic control parameters, equipment status data, public environment data, and non-task-specific data.

6. A training device for a robot control model, characterized in that, The device includes: The training admission judgment module is used to confirm that the first training frequency identifier carried by the target node meets the preset conditions. The training contribution determination module is used to determine the second training frequency identifier corresponding to the target node based on the private weight of the robot training data reported by the target node and the first private data priority. The data priority calculation module is used to update the second private data priority corresponding to the current training round based on the robot training data reported by each node. The parameter distribution module is used to distribute the model parameters, the priority of the second private data, and the second training frequency identifier to the target node after completing the current round of federated learning training. The step of determining the second training frequency identifier corresponding to the target node based on the privacy weight of the robot training data reported by the target node and the priority of the first privacy data specifically includes: The privacy level of each training data sample in the training data is determined, and then the privacy weight of the training data is calculated. The robot training data is divided into a private data sample set and a non-private data sample set, and a first private data priority of the robot training data is calculated based on the private data sample set. The second training frequency identifier corresponding to the target node is calculated based on the privacy weight and the first privacy data priority. The privacy weight of the robot training data is determined based on the quantity and data type of the training data; The specific calculation method for the priority of the first private data is as follows: in, P This indicates the result with the highest priority for private data. n This indicates the number of samples in the private data sample set. Indicates the first i Priority data for a sample of private data. Indicates the first i Weights of private data samples; The second training frequency identifier uses the following formula: f = f max -( α • w + β • p ) in, w Indicates privacy weight. p This indicates the priority of the first private data. f Indicates the second training frequency identifier. α and β This involves adjusting the weighting coefficients that influence privacy weight and privacy data priority; these should be adjusted according to specific needs. f max It is the maximum possible value of the second training frequency identifier.

7. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium having executable instructions stored thereon, characterized in that, When the executable instructions are executed by the processor, the processor performs the method as described in any one of claims 1 to 5.

Images