Inherited network generation operating system

Abstract

The invention relates to the field of operating systems, in particular to an inheritance type network generation operating system. The method is applied to a controller with a graphical user interface and a power robot only configured with a posture control model. The graphical user interface at least comprises an input module, a network module, a model generation judgment module and an output control module icon. Whenever a connecting line is drawn between the input mark and the output mark corresponding to the two icons, the controller controls the background corresponding function module to establish communication connection to transmit preset data. According to the method, a control link is established between a background function module required by a power robot attitude control model and a front-end graphical interface icon and graph drawing operation, a user simply and intuitively draws and connects a graphical interface graph according to model training logic, and an attitude control model training control chart is generated. The background starts the corresponding function module, transmits data according to the input and output connecting line, and achieves the training operation of the attitude control model, so that the model training construction is more visual and simpler, and is convenient to use.

Description

Technical Field

[0001] The present invention relates to the field of operating systems, in particular to an inheritance-type network-generated operating system. Background Art

[0002] The control model of a powered robot refers to a set of mathematical models and algorithms used to control the robot's motion and operation. These models and algorithms ensure that the robot can perform actions according to predetermined goals. For example, the control model can be used to control the various joints or wheels that drive the robot and enable it to maintain stability and efficiency in a dynamic environment.

[0003] Usually, in order to enable the power robot to better maintain the target posture in the corresponding usage scenario, a posture control model that conforms to the current usage scenario will be configured for the robot. In the prior art, when training the posture control model of a power robot, multiple steps and technologies are usually involved to ensure that the robot can accurately perform the expected actions and maintain a stable posture. This process usually includes key links such as data collection, model selection, training methods and evaluation. However, when operating these links, it is usually necessary to configure and adjust the corresponding background programs in order to build a complete model training process. It requires high technical capabilities of the operator, and the operation is more complicated and cumbersome, which is not convenient to use. Summary of the Invention

[0004] In order to solve one of the above technical problems, the present invention adopts the following technical solution: According to one aspect of the present invention, an inherited network generation operating system is provided. The system is applied to a controller and a power robot. The controller has a graphical user interface, which includes at least: an input module icon, a network module icon, a model generation and determination module icon, and an output control module icon. The power robot is configured with only a posture control model. Whenever a connection line is drawn between the input mark and the output mark corresponding to the two icons, the controller controls the functional modules corresponding to the two icons in the background to establish a communication connection to transmit preset data; When the connections between the input module icon, the network module icon, the model generation and determination module icon, and the output control module icon are completed, the controller controls the functional modules corresponding to the icons to execute the following steps: The network module corresponding to the network model icon is used to start the posture control model corresponding to the power robot; the posture control model is used to generate a posture adjustment instruction of the power robot according to the input state parameters; the state parameters include the current posture data of the power robot; the posture control model includes an input layer, an output layer and at least one hidden layer, and a full connection is set between two adjacent layers; The output control module corresponding to the output control module icon is used to adjust the posture of the power robot according to the posture adjustment instruction; The input module corresponding to the input module icon is used to obtain the current posture data of the power robot and input it into the posture control model and the model generation determination module corresponding to the model generation determination module icon respectively; The model generation and determination module corresponding to the model generation and determination module icon is used to perform model parameter solidification processing according to the current posture data in each training cycle to update the model weight of the posture control model; The model parameter solidification process includes: Integrating the difference between the current posture data and adjacent historical posture data to generate an adjustment posture coefficient corresponding to the posture adjustment instruction of the current posture data; If the absolute value of the difference between the adjustment posture coefficient corresponding to the current posture data and the target posture coefficient is greater than or equal to the preset validity threshold, the invalid coefficient of model training is accumulated and added by one; If the absolute value of the difference between the adjustment posture coefficient corresponding to the current posture data and the target posture coefficient is less than the preset validity threshold, the model training validity coefficient is accumulated and added by one; When the model solidification node corresponding to the current training cycle is reached, if the ratio between the increment of the effective coefficient of model training and the increment of the invalid coefficient of model training during the training cycle is greater than the preset ratio threshold, the model weight in the posture control model will be updated to the model weight at the current model solidification node.

[0005] The present invention has at least one of the following beneficial effects: In the present invention, each background functional module required for the posture control model of the power robot is connected to the icons and dependent graphics drawing operations in the front-end graphical user interface to establish relevant control links. This allows the user to perform relatively simple and intuitive drawing connections on the corresponding graphics in the graphical user interface according to the logic of model training to generate a training control diagram for a certain posture control model. The corresponding background will start the functional modules corresponding to these graphics, and transfer data between different modules according to the input and output connection lines between the graphics to realize the training operation of the posture control model in the background. As a result, the construction work of model training can be made more intuitive and simple, and easy to use.

[0006] At the same time, in the present invention, the model generation and determination module automatically determines whether the current training optimization direction of the model is effective based on whether the posture of the power robot is closer to the target posture after each posture adjustment instruction output by the posture control model during the training cycle. Then, the ratio between the number of invalid optimizations during the entire training cycle is used to determine whether the posture control model is converging to the target state during the current training cycle, and then the model weight parameters are fixed. As a result, each subsequent cycle is based on the model after inheriting the updated parameters obtained in the previous training cycle. Continuing training can make the model gradually converge to the target state quickly, and the training efficiency is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic diagram of a graphical user interface in an inheritance-based network-generated operating system provided by an embodiment of the present invention; Figure 2 A schematic diagram of an interface of a posture control model after updating the model weights after a training cycle is completed during posture control model training using an inheritance-based network generation operating system according to an embodiment of the present invention; Figure 3 A schematic diagram of the interface of the posture control model after the posture control model training is completed using the inheritance network generation operating system provided in an embodiment of the present invention; Figure 4 A flowchart of an inheritance network training method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0010] As a possible embodiment of the present invention, Figure 1 As shown, an inherited network generation operating system is provided. The system is applied to a controller and a power robot. The controller has a graphical user interface, such as Figure 1As shown, the graphical user interface includes at least the following: an input module icon (also corresponding to the angle sensor), a network module icon (also corresponding to the computational network), a model generation and determination module icon (also corresponding to the function), and an output control module icon (also corresponding to the motor control). Power robots are configured with only the posture control model.

[0011] Specifically, in this embodiment, powered robots refer to robots that can move or operate autonomously, possess certain intelligent control capabilities, and are typically driven by electric motors, hydraulic systems, or pneumatic systems. Examples include industrial robots, drones, robot dogs, humanoid robots, and service robots.

[0012] Whenever a connection line is drawn between the input mark and the output mark corresponding to the two icons, the controller controls the functional modules corresponding to the two icons in the background to establish a communication connection to transmit preset data.

[0013] Of course, within the graphical user interface of this operating system, you can also click on an input module icon to start the configuration function of the corresponding function module in the background, and then perform adaptive configuration operations on the function module. For example, you can double-click the input module icon to start the configuration function of the input module, and then configure the posture data to be collected and input as needed, or set the number of output marker graphics to send the acquired data to multiple modules.

[0014] When the connections between the input module icon, the network module icon, the model generation and determination module icon, and the output control module icon are completed, the controller controls the functional modules corresponding to each icon to execute the following steps: The network module corresponding to the network model icon is used to activate the posture control model corresponding to the power robot. The posture control model generates posture adjustment instructions for the power robot based on input state parameters. The state parameters include the current posture data of the power robot. The posture control model includes an input layer, an output layer, and at least one hidden layer, with full connections between adjacent layers.

[0015] Specifically, the attitude control model is a deep learning neural network model. The number of nodes in the input layer and output layer of the attitude control model can be determined according to the types of attitude data to be collected and the number of motion joints of the power robot that need to be controlled. Take the power robot as a four-rotor drone, and the attitude control model is used to maintain the drone in a horizontal centered attitude at all times, and the attitude data includes pitch angle and roll angle as an example. The corresponding input layer has 2 nodes and the output layer has 4 nodes. The number of hidden layers can be one, two or three. Since the attitude control logic of the power robot is relatively simple, it is usually possible to achieve good control of the power robot's attitude by combining multiple or and not strategies. Correspondingly, in a neural learning network, generally three or fewer hidden layers can fully express multiple or and not strategy combinations. Therefore, in this embodiment, the number of hidden layers and the number of nodes in each layer can be set as needed.

[0016] The output control module corresponding to the output control module icon is used to adjust the posture of the power robot according to the posture adjustment instruction.

[0017] The input module corresponding to the input module icon is used to obtain the current posture data of the power robot and input it into the posture control model and the model generation determination module corresponding to the model generation determination module icon respectively.

[0018] Obtaining the current posture data of the power robot includes: According to the preset data collection interval, the current posture data of the power robot is obtained based on the information released by the power robot through ROS2 at each collection moment. In each preset data collection interval, the power robot performs at least one posture adjustment according to the posture adjustment instruction.

[0019] Current power robots generally use the topic mechanism in ROS 2 (Robot Operating System 2) to publish the current attitude value of the drone. Therefore, the input module in this embodiment converts the received attitude information into data in the corresponding input format according to the corresponding collection time.

[0020] Typically, after a power robot performs an adjustment based on a posture adjustment command, it needs to reacquire the adjusted posture information and then re-enter the model generation and determination module and the posture control model for the next training adjustment. Therefore, in this embodiment, the preset data collection interval is generally the time interval required for a complete posture adjustment. Furthermore, a training cycle requires multiple posture adjustments, so the duration of the training cycle is longer than the preset data collection interval.

[0021] In addition, the posture input module can also obtain data of other influencing factors in addition to the current posture data. For example, in the graphical user interface, the input module icons can include a posture input module icon and an environmental influencing factor input module icon.

[0022] The input module corresponding to the posture input module icon is used to obtain the current posture data of the power robot and input it into the posture control model and the model generation determination module corresponding to the model generation determination module icon respectively.

[0023] The input module corresponding to the environmental impact factor input module icon is used to obtain environmental data that affects the posture of the power robot in its current environment and input it into the posture control model and the model generation and determination module corresponding to the model generation and determination module icon. For example, the environmental data can be wind speed.

[0024] The model generation and determination module corresponding to the model generation and determination module icon is used to perform model parameter solidification processing based on the current posture data in each training cycle to update the model weights of the posture control model.

[0025] Model parameter solidification processing includes: The difference between the current posture data and the adjacent historical posture data is integrated to generate the adjustment posture coefficient of the posture adjustment instruction corresponding to the current posture data.

[0026] If the absolute value of the difference between the adjustment posture coefficient corresponding to the current posture data and the target posture coefficient is greater than or equal to the preset validity threshold, the invalid coefficient of model training is accumulated and added by one.

[0027] If the absolute value of the difference between the adjustment posture coefficient corresponding to the current posture data and the target posture coefficient is less than the preset validity threshold, the model training validity coefficient is accumulated and added by one.

[0028] When the model solidification node corresponding to the current training cycle is reached, if the ratio between the increment of the effective coefficient of model training and the increment of the invalid coefficient of model training during the training cycle is greater than the preset ratio threshold, the model weight in the posture control model will be updated to the model weight at the current model solidification node.

[0029] When the model solidification node corresponding to the current training cycle is reached, if the ratio between the increment of the effective model training coefficient and the increment of the invalid model training coefficient during the training cycle is less than or equal to the preset ratio threshold, the model weights in the posture control model are randomly changed.

[0030] If the posture control effect of the current posture control model reaches the preset condition, the model parameter solidification process is stopped, and the current model weight is configured into the local posture control model of the power robot.

[0031] The model generation judgment module is the key functional module in this embodiment. It determines whether the posture adjustment instruction output by the model each time is valid by checking whether the posture change trend of the current power robot after each posture adjustment is close to the target posture. At the same time, it also determines whether the model has effectively converged to the target state in the current training cycle by accumulating statistics on the ratio between the number of effective adjustments and the number of invalid adjustments in each training cycle and whether the threshold is reached.

[0032] If so, the model's weight parameters are updated and saved so that they can be further optimized on the current basis in subsequent training cycles, thereby improving the model's training speed by inheriting the training results of the previous cycle.

[0033] If not, the weight parameters of the model are randomly changed to make the model jump into another feasible domain for optimization.

[0034] Specifically, taking the power robot as a quad-rotor drone, the attitude control model is used to maintain the drone in a horizontal attitude at all times, the attitude data includes the pitch angle, the preset data collection interval is 20ms, the target pitch angle is 0°, the preset ratio threshold is 0.8, and the preset validity threshold is 30° as an example, the model parameter solidification process within a certain training cycle (1S) is explained: The pitch angle currently collected is A t , the previous adjacent historical pitch angle is A t-1 , A t The attitude adjustment coefficient Pt corresponding to the attitude adjustment instruction satisfies the following conditions: ; Where b is the current number of data collection times.

[0035] Therefore, the above integral operation can represent the degree of deviation of the current UAV attitude from 0° after each attitude adjustment. Therefore, if the absolute value of Pt is greater than 30°, it is considered that the current adjustment has deviated significantly from the target and the adjustment strategy is invalid. The invalid coefficient of model training is accumulated by 1; otherwise, the valid coefficient of model training is accumulated by 1.

[0036] Finally, during the entire training cycle, there are 50 validity checks. If the proportion of valid times to the total exceeds 0.8, or the ratio of valid times to invalid times is greater than 4, the model weights in the posture control model are updated to the model weights when the current model is solidified. Otherwise, the model weights in the posture control model are randomly changed.

[0037] In this embodiment, each backend functional module required for the power robot's posture control model is linked to icons and associated graphics drawing operations within the frontend graphical user interface (GUI). This allows users to generate a training control diagram for a posture control model by simply and intuitively drawing connections based on the model training logic. The corresponding backend will activate the functional modules corresponding to these graphics and, based on the input and output connection lines between the graphics, transfer data between the modules to implement posture control model training operations in the backend. This makes model training more intuitive and simple, making it easier to use.

[0038] At the same time, in this embodiment, the model generation and determination module automatically determines whether the model's current training optimization direction is effective based on whether the power robot's posture is closer to the target posture after each posture adjustment instruction output by the posture control model during the training cycle. The ratio of invalid optimization attempts within the entire training cycle is then used to determine whether the posture control model is converging to the target state within the current training cycle, thereby fixing the model weight parameters. Thus, each subsequent cycle is based on the model that inherits the updated parameters obtained in the previous training cycle. Continued training can enable the model to gradually and quickly converge to the target state, resulting in higher training efficiency.

[0039] This embodiment also provides the use of the inherited network generation operating system in this embodiment to train and generate a posture control network model of a self-adjusting gimbal robot (also known as a camera stabilizer). The gimbal robot uses four drive motors below to control the swing of the bracket to change the posture of the camera above. The corresponding posture control model is used to ensure that the camera above is in a horizontal state. The input layer of the model is 6 input nodes, corresponding to the X-direction acceleration, Y-direction acceleration, Z-direction acceleration, roll angle, pitch angle and yaw angle respectively. The output layer is 4 corresponding output control nodes, which control the four motors respectively, and a hidden layer.

[0040] In the process of training and generating a model using the system in this embodiment, Figure 2 As shown in the figure, it is the posture control model after a certain training cycle, and the content displayed in the lower left corner is the model weight recorded at each time when the model weight is fixed. Figure 3 As shown in Figure 2, the posture control model after training is completed.

[0041] As another embodiment of the present invention, Figure 4As shown, an inheritance network training method is also provided. This method can be used to further train a power robot equipped with a mature original posture control model. Without fusing and superimposing the two models, the new model can inherit the capabilities of the original posture control model and further optimize and improve the posture control capabilities of the power robot in new usage scenarios.

[0042] Specifically, the method includes the following steps: S100: A new posture control optimization model for the power robot is created. This posture control optimization model (equivalent to the posture control model in the previous embodiment) is used to generate posture adjustment instructions for the power robot based on input state parameters. The state parameters include the power robot's current posture data. The posture control optimization model includes an input layer, an output layer, and at least one hidden layer, with full connections between adjacent layers. The power robot is already configured with an original posture control model. This original posture control model is the posture control model generated by the power robot to meet the requirements of the first usage task.

[0043] In this embodiment, there are differences between the original posture control model and the newly created posture control optimization model. The difference is not only a difference in model capabilities, but also a difference in model types. For example, the original posture control model is a PID control model, while the posture control optimization model is a deep learning neural network model.

[0044] The original posture control model is usually a general control model trained by the manufacturer when the power robot leaves the factory. The first usage task requirement usually targeted by the model during training is different from the second usage task requirement targeted by the posture control optimization model in this embodiment.

[0045] Taking a humanoid power robot as an example, the first usage task requirement is that the power robot can stand up again after falling down in different usage scenarios, while the second usage task requirement is that the power robot can stand up again after falling down in different usage scenarios after carrying a certain load.

[0046] S200: The acquired current posture data of the power robot is input into the posture control optimization model and the training and judgment module. The posture control optimization model is a neural learning network model newly trained to meet the second task requirement. The first and second task requirements have different factors affecting the posture of the power robot.

[0047] S300: The training determination module is used to perform model parameter solidification processing according to the current posture data in each training cycle to update the model weights of the posture control optimization model.

[0048] Model parameter solidification processing includes: S301: Integrate the difference between the current posture data and the adjacent historical posture data to generate an adjustment posture coefficient corresponding to the posture adjustment instruction of the current posture data.

[0049] S302: If the absolute value of the difference between the adjustment posture coefficient corresponding to the current posture data and the target posture coefficient is greater than or equal to the preset validity threshold, the invalid coefficient of model training is accumulated and added by one.

[0050] S303: If the absolute value of the difference between the adjustment posture coefficient corresponding to the current posture data and the target posture coefficient is less than the preset validity threshold, the model training validity coefficient is accumulated and added by one.

[0051] S304: When the model solidification node corresponding to the current training cycle is reached, if the ratio between the increment of the effective coefficient of model training and the increment of the invalid coefficient of model training during the training cycle is greater than the preset ratio threshold, the model weight in the posture control optimization model is updated to the model weight at the current model solidification node.

[0052] S305: When the model solidification node corresponding to the current training cycle is reached, if the ratio between the increment of the effective coefficient of model training and the increment of the invalid coefficient of model training during the training cycle is less than or equal to the preset ratio threshold, the model weights in the posture control optimization model are randomly changed.

[0053] In this embodiment, when a power robot trains a newly constructed posture control optimization model to adapt to a new usage scenario, the posture control optimization model is trained based on the control instructions output by the model and the posture information after the power robot's adjustments, which serve as the basis for determining whether the model's current adjustments are effective, thereby driving the model to converge toward the target state. During this process, the original posture control model deployed in the power robot also runs in parallel. That is, the posture information after the power robot's adjustments includes the adjustment information issued by the original posture control model to the power robot. Therefore, when the present invention trains the newly constructed posture control optimization model, its training data (i.e., posture information) includes the adjustment information issued by the original posture control model to the power robot, as well as the compensation adjustment information output by the newly constructed posture control optimization model. Accordingly, the posture adjustment instructions output by the trained posture control optimization model take into account the compensation information of the posture adjustment instructions output by the original posture control model. This eliminates the need to fuse and superimpose the current posture control optimization model with the original posture control model, thereby achieving posture optimization adjustment of the power robot when adapting to a new usage scenario based on the original posture control model. This method is less time-consuming, more efficient, and does not cause incompatibility issues between models.

[0054] Furthermore, in this embodiment, the training determination module determines whether the model's current training optimization direction is effective based on whether the power robot's posture approaches the target posture after each posture adjustment command output by the posture control optimization model during the training cycle. The module then determines whether the posture control optimization model is converging toward the target state within the current training cycle by comparing the number of invalid optimization attempts within the entire training cycle, thereby fixing the model's weight parameters. Consequently, each subsequent cycle is based on the model's updated parameters from the previous training cycle. Continuing training allows the model to gradually and rapidly converge toward the target state, resulting in higher training efficiency.

[0055] Furthermore, although the steps of the method of the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that the steps must be performed in this particular order, or that all steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0056] Through the description of the above embodiments, it will be readily understood by those skilled in the art that the example embodiments described herein can be implemented via software or via a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or mobile hard drive) or on a network and includes several instructions for enabling a computing device (such as a personal computer, server, mobile terminal, or network device) to execute the methods according to the embodiments of the present disclosure.

[0057] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. An inheritance-type network generation operating system, characterized in that: The system is applied to a controller and a power robot, wherein the controller has a graphical user interface, which includes at least: an input module icon, a network module icon, a model generation and determination module icon, and an output control module icon; the power robot is only equipped with a posture control model; Whenever a connection line is drawn between the input mark and the output mark corresponding to the two icons, the controller controls the functional modules corresponding to the two icons in the background to establish a communication connection to transmit preset data; When the connections between the input module icon, the network module icon, the model generation and determination module icon, and the output control module icon are completed, the controller controls the functional modules corresponding to the icons to execute the following steps: The network module corresponding to the network model icon is used to start the posture control model corresponding to the power robot; the posture control model is used to generate a posture adjustment instruction of the power robot according to the input state parameters; the state parameters include the current posture data of the power robot; the posture control model includes an input layer, an output layer and at least one hidden layer, and a full connection is set between two adjacent layers; The output control module corresponding to the output control module icon is used to adjust the posture of the power robot according to the posture adjustment instruction; The input module corresponding to the input module icon is used to obtain the current posture data of the power robot and input it into the posture control model and the model generation determination module corresponding to the model generation determination module icon respectively; The model generation and determination module corresponding to the model generation and determination module icon is used to perform model parameter solidification processing according to the current posture data in each training cycle to update the model weight of the posture control model; The model parameter solidification process includes: Integrating the difference between the current posture data and adjacent historical posture data to generate an adjustment posture coefficient corresponding to the posture adjustment instruction of the current posture data; If the absolute value of the difference between the adjustment posture coefficient corresponding to the current posture data and the target posture coefficient is greater than or equal to the preset validity threshold, the invalid coefficient of model training is accumulated and added by one; If the absolute value of the difference between the adjustment posture coefficient corresponding to the current posture data and the target posture coefficient is less than the preset validity threshold, the model training validity coefficient is accumulated and added by one; When the model solidification node corresponding to the current training cycle is reached, if the ratio between the increment of the effective coefficient of model training and the increment of the invalid coefficient of model training during the training cycle is greater than the preset ratio threshold, the model weight in the posture control model will be updated to the model weight at the current model solidification node.

2. The inherited network generation operating system according to claim 1, characterized in that: The obtaining of the current posture data of the power robot includes: According to the preset data collection interval, the posture data corresponding to the power robot at each collection moment is obtained; in each of the preset data collection intervals, the power robot performs at least one posture adjustment according to the posture adjustment instruction.

3. The inherited network generation operating system according to claim 2, characterized in that: The acquisition of the posture data corresponding to the power robot at each acquisition moment includes: According to the information released by the power robot through ROS2 at each collection moment, the current posture data of the power robot is obtained.

4. The inherited network generation operating system according to claim 1, characterized in that: The posture control model is a deep learning neural network model.

5. The inherited network generation operating system according to claim 4, characterized in that: The number of hidden layers in the posture control model is one, two or three.

6. The inherited network generation operating system according to claim 1, characterized in that: The power robot is a drone, and the current posture data includes a pitch angle and / or a roll angle.

7. The inheritance-based network generation operating system according to claim 1, characterized in that: When reaching the model solidification node corresponding to the current training cycle, the model parameter solidification process further includes: If, during the training cycle, the ratio of the model training effective coefficient increment to the model training invalid coefficient increment is less than or equal to a preset ratio threshold, the model weights in the posture control model are randomly changed.

8. The inherited network generation operating system according to claim 1, characterized in that: The model generation and determination module is further configured to perform the following steps: If the posture control effect of the current posture control model reaches the preset condition, the model parameter solidification process is stopped, and the current model weight is configured into the local posture control model of the power robot.

9. The inherited network generation operating system according to claim 1, characterized in that: The input module icons include a posture input module icon and an environmental influence factor input module icon; The input module corresponding to the posture input module icon is used to obtain the current posture data of the power robot and input it into the posture control model and the model generation determination module corresponding to the model generation determination module icon respectively; The input module corresponding to the environmental influencing factor input module icon is used to obtain environmental data that affects the posture of the power robot in the current environment, and input it into the posture control model and the model generation judgment module corresponding to the model generation judgment module icon respectively.

10. The inheritance-type network generation operating system according to claim 9, characterized in that: The environmental data includes wind speed.

Images