Robot control method and control device

By obtaining the status information of the robot and target objects and inputting a pre-trained strategy model, the automatic manipulation of the robot arm is realized, solving the limitations of human task pre-analysis and pre-programming, and improving adaptability and control accuracy.

CN118990478APending Publication Date: 2024-11-22ANHUI PEITIAN ROBOT GRP CO LTD

Patent Information

Application Number
CN202411134265.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing robotic arm control methods require artificial task pre-analysis and pre-programming, which is difficult to adapt to complex and changeable environments and task types, resulting in limited use in non-heightly controllable environments.

Method used

By obtaining the status information of the robot and the target object, input a pre-trained strategy model to obtain the target action and parse it into control instructions to achieve the robot automatically manipulating the target object.

Benefits of technology

Improves the adaptability and control accuracy of the robotic arm, allowing it to quickly adapt to environmental changes, reduces the complexity and time of artificial programming, and is suitable for a variety of different task scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118990478A_ABST
    Figure CN118990478A_ABST
Patent Text Reader

Abstract

The invention discloses a robot control method and device, and the method comprises the steps: obtaining the current first state information of a robot, obtaining the current second state information of a target object, and enabling the robot to control the target object at present; inputting the first state information and the second state information into a pre-trained strategy model to obtain a target action; the target action is analyzed into a control instruction, and the control instruction is sent to the robot so that the robot can control the target object. According to the method, the strategy model is trained in advance, and the robot is controlled to move by using the strategy model, so that the robot can process a high-dynamic mechanical arm task, the mechanical arm can quickly adapt to surrounding environment changes, and the adaptability and the control precision of the mechanical arm are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of automation control technology, and in particular to a robot control method and control device. Background Art

[0002] In current technology, the robotic arm has always been installed on a fixed base. The task analysis of the robotic arm is manually analyzed and disassembled into specific parameters such as position, speed, and force by engineering and technical personnel, and these parameters are used as targets and controlled in a pre-programmed manner. When the environment, goals, and task content change, manual re-analysis of the task is required. When the environment that the robotic arm needs to deal with changes drastically, the task types are diverse, and even a large number of unexpected scenarios occur, the task analysis of the robotic arm cannot be completed manually in advance. This traditional strategy model based on task pre-analysis and pre-programming is difficult to adapt to complex and changing requirements, resulting in the composite robot equipped with a robotic arm currently being limited to industrial environments with highly controllable environments. Summary of the invention

[0003] The present application provides a robot control method and a control device to solve the problem in the prior art that tasks need to be pre-analyzed and pre-programmed manually.

[0004] A first aspect of an embodiment of the present application provides a robot control method, the method comprising: obtaining first current state information of the robot and second current state information of a target object, wherein the robot is currently manipulating the target object; inputting the first state information and the second state information into a pre-trained strategy model to obtain a target action; parsing the target action into a control instruction, and sending the control instruction to the robot so that the robot manipulates the target object.

[0005] A second aspect of an embodiment of the present application provides a control device, comprising: an acquisition module, used to acquire a robot's current first state information, and to acquire a target object's current second state information, wherein the robot is currently manipulating the target object; an input module, connected to the acquisition module, used to input the first state information and the second state information into a pre-trained strategy model to obtain a target action; and a parsing module, connected to the input module, used to parse the target action into a control instruction, and send the control instruction to the robot so that the robot manipulates the target object.

[0006] A third aspect of an embodiment of the present application provides a control device, which includes a processor and a memory, wherein the processor is coupled to the memory, wherein program data is stored in the memory, and the processor implements the steps in any of the above methods by executing the program data in the memory.

[0007] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program can be executed by a processor to implement the steps in any of the above methods.

[0008] The beneficial effect is: the present application can realize the robot's manipulation of the target object by pre-training the strategy model and inputting the robot's current first state information and the target object's current second state information into the strategy model, so that the robot can handle highly dynamic robotic arm tasks, and the robotic arm can quickly adapt to surrounding environmental changes, thereby improving the adaptability and control accuracy of the robotic arm. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0010] Figure 1 It is a flowchart of an implementation method of the robot control method of the present application;

[0011] Figure 2 It is a flowchart of an implementation method of the robot control method of the present application;

[0012] Figure 3 It is a flowchart of an implementation method of the robot control method of the present application;

[0013] Figure 4 It is a flowchart of an implementation method of the robot control method of the present application;

[0014] Figure 5 It is a structural schematic diagram of an implementation method of a control device of the present application;

[0015] Figure 6 It is a structural schematic diagram of another implementation mode of the control device of the present application;

[0016] Figure 7 It is a structural schematic diagram of an implementation method of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0018] It should be noted that the terms "first" and "second" in this application are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0019] See also Figure 1 , Figure 1 1 is a flow chart of an embodiment of a control method of a robot of the present application. In one embodiment, the control method of the robot includes:

[0020] S110: Acquire current first state information of the robot, and acquire current second state information of the target object, wherein the robot is currently manipulating the target object.

[0021] The control method of the present application is executed by a control device, the control device is used to obtain the current first state information of the robot and the current second state information of the target object, and the robot is currently manipulating the target object. In one embodiment, a table tennis racket is fixed at the end of the robot, and when the robot is required to swing the table tennis racket to hit the table tennis ball or flip the table tennis ball up and down, the target object is the table tennis ball.

[0022] In one application scenario, in a robot-bouncing ping-pong ball task, the robot's current first state information includes but is not limited to: the position and speed of each axis of the robot, the position and speed of the robot's end. The target object's current second state information includes but is not limited to: the position and speed of the ping-pong ball.

[0023] S120: Input the first state information and the second state information into a pre-trained strategy model to obtain a target action.

[0024] Specifically, the strategy model is pre-trained, and after receiving the first state information and the second state information, it can predict the next action to be performed by the robot, that is, the target action.

[0025] For example, in a table tennis juggling task, the target action output by the strategy model can be: the position of the robot's end (that is, the position of the TCP point at the next moment), that is, the strategy model outputs the position of the robot's end at the next moment.

[0026] S130: Parse the target action into a control instruction, and send the control instruction to the robot so that the robot can manipulate the target object.

[0027] Specifically, after obtaining the target action, the target action can be parsed into a control instruction that the robot can recognize, and finally the control instruction is sent to the robot so that the robot executes the target action output by the strategy model. In one embodiment, the target action output by the strategy model in step S120 is the position of the robot end, and step S130 reversely resolves the position of the robot end into the position of each joint axis of the robot, and sends the position of each joint axis of the robot to the robot.

[0028] In one application scenario, when parsing the target action, step S130 may obtain one control instruction or multiple control instructions. If multiple control instructions are obtained, the multiple control instructions may be sent to the robot in sequence. For example, when the multiple control instructions obtained by parsing are the positions of the robot's six joint axes, the positions of the six joint axes are sent to the robot in sequence.

[0029] After step S130, return to execute step S110, that is, continuously obtain the current first state information of the robot and the current second state information of the target object, input the first state information and the second state information into the pre-trained strategy model to obtain the target action, parse the target action into control instructions and send the control instructions to the robot, so that the robot can manipulate the target object.

[0030] The implementation method of the present application can realize the robot controlling the target object by pre-training the strategy model and inputting the current first state information of the robot and the current second state information of the target object into the strategy model. The traditional way of controlling robots requires complex and cumbersome control programming, which is difficult to quickly adapt to new task requirements. However, the solution of the present application does not require complex control algorithm programming. The robot can control the target object only through the trained strategy model. The method of operating the robot is simple and efficient.

[0031] At the same time, in the process of manipulating the target object, the current first state information of the robot and the current second state information of the target object are continuously obtained, so that the strategy model can output different target actions according to different state information to achieve the purpose of adjusting the target action.

[0032] Moreover, the control instructions parsed out according to different target actions are also different. The robot manipulates the target object according to the constantly adjusted control instructions, so that the robot can handle highly dynamic robotic arm tasks, and the robotic arm can quickly adapt to changes in the surrounding environment, thereby improving the adaptability and control accuracy of the robotic arm.

[0033] In one embodiment, see Figure 2 In step S120, the first state information and the second state information are input into a pre-trained strategy model, and the target action is obtained, specifically including:

[0034] S121: Convert the first state information and the second state information into target data according to a preset rule, wherein the target data carries the first state information and the second state information.

[0035] S122: Input the target data into the strategy model to obtain the target action.

[0036] Specifically, the control device converts the acquired first state information and second state information into target data according to preset rules and then inputs the target data into the strategy model. The target data can be directly recognized by the strategy model, which can improve the processing speed of the strategy model and ensure the efficiency of the algorithm.

[0037] In other embodiments, the control device can also convert the state information of the robot at the previous moment, the state information of the target object at the previous moment, the current first state information of the robot, and the current second state information of the target object into target data. At this time, the target data carries the state information of the robot at the previous moment, the current first state information of the robot, the state information of the target object at the previous moment, and the current second state information of the target object. This target data is input into the strategy model, and the target action obtained is more accurate. In other words, the target data can carry not only the information of the current moment, but also the information of the historical moment.

[0038] In this embodiment, the robot control method further includes: upon receiving a switching instruction, switching the strategy model to another strategy model.

[0039] In the control method of this embodiment, instructions can be switched according to user needs. For example, if the robot is currently performing a table tennis juggling task and the user requires the robot to stop the table tennis juggling task and perform a badminton swinging task, the control device will issue a switching instruction according to the user's needs to switch the table tennis juggling strategy model to a badminton swinging strategy model. The robot of this embodiment can handle tasks of multiple different modes and is suitable for different scenarios, thereby improving the universality of the robot.

[0040] Continue reading Figure 2 In one embodiment, before step S130, the method further includes:

[0041] S123: Determine whether the target action meets the preset requirements.

[0042] If the preset requirement is not met, the process returns to step S120, and the first state information and the second state information are re-input into the pre-trained strategy model to obtain the target action again.

[0043] If the target action meets the preset requirements, step S130 is executed to parse the target action into a control instruction, and the control instruction is sent to the robot so that the robot controls the target object.

[0044] Specifically, under normal circumstances, the target action needs to meet reasonable action requirements, such as whether the juggling height or the robot's arm angle is suitable for juggling, or whether the position of the robot's end is within the robot's range of motion. If the height and angle are not suitable for juggling, or the position of the robot's end has exceeded the robot's maximum range of motion, then juggling will fail or even damage the robot. Therefore, in order to achieve the target action and extend the robot's service life, this embodiment needs to determine whether the target action meets the preset requirements.

[0045] Of course, in other implementations, after the strategy model outputs the target action, step S123 may not be executed, but step S130 may be executed directly.

[0046] In one embodiment, in order to avoid falling into an infinite loop, before returning to execute step S120, the number of times of returning to execute step S120 can be counted. If the number reaches a threshold, the model is retrained. If the number does not reach the threshold, return to execute step S120.

[0047] The following describes the training process of the above strategy model, see Figure 3 , the training process includes:

[0048] S101: Inputting the scene parameters of the target task into the strategy model, wherein the target task is the robot manipulating the target object, and the scene parameters include at least one of the parameters of the robot, the parameters of the robot's end tool, the parameters of the target object, and the parameters of the surrounding environment when the robot manipulates the target object.

[0049] In order to enable the robot to manipulate the target object, it is necessary to build corresponding scene parameters, which include at least one of the parameters of the robot, the parameters of the robot's end tool, the parameters of the target object, and the parameters of the surrounding environment when the robot manipulates the target object.

[0050] For example, in the task of juggling a ping-pong ball, the parameters of the robot include the model of the robot, the installation position of the robot and other parameters; the parameters of the robot's end tool include the size, installation position and rebound coefficient of the ping-pong racket and other parameters; the parameters of the target object include the size, weight, rebound coefficient and other parameters of the ping-pong ball; and the parameters of the surrounding environment when the robot manipulates the target object include the table top, fence, etc.

[0051] In other implementations, the surrounding environment parameters also include parameters such as the model, installation location, angle, etc. of a camera used to monitor the movement of the target object.

[0052] This implementation method builds scenario parameters based on real scenarios and technical solutions, so that the ability of the strategy model to complete the target task is the same or equivalent to the real scenario.

[0053] S102: Setting target parameters of the target task, wherein the target parameters include at least one of parameters of the robot and parameters of the target object when the robot manipulates the target object to reach the target.

[0054] Specifically, the target parameters are various parameters of the robot when performing the target task under ideal conditions.

[0055] For example, in a ping-pong ball juggling task, the target task is to juggle the ping-pong ball to a height of, for example, 30 cm. At this time, the target parameters may include the parameters of the robot when juggling the ping-pong ball to a height of 30 cm (the position and speed of each axis of the robot, the position and speed of the end), and the parameters of the ping-pong ball (such as the position and speed of the ping-pong ball, etc.).

[0056] Setting the target parameters can quantify the state information of the target task to evaluate whether the target task is completed. This application quantifies the state information and can handle complex dynamic nonlinear tasks, which are generally difficult to model using traditional control methods. For example, a neural network is used to represent the strategy model. By setting the network structure and target parameters, the complex model can be replaced to achieve control of dynamic nonlinear tasks.

[0057] S103: Building a simulation environment, and in the simulation environment, controlling the robot to manipulate the target object according to the actions output by the strategy model.

[0058] The strategy model of this implementation is based on simulation technology, which means using 3D physics engine technology to establish physical constraints that conform to real rules in a virtual environment to simulate the behavior of robots and other objects in the environment. This technology can be used to collect state information, motion, images and other data of the robot and the surrounding environment from the simulation environment, which greatly reduces time cost, resource cost, safety risk, etc. compared with real robot collection.

[0059] This embodiment builds a simulation environment in the simulation software. Specifically, the initial state of the robot, target object and surrounding environment, simulation step length, gravity friction and other dynamic parameters are set in the simulation software. At the same time, the action rules of the robot action and the rendering rules of the simulated camera are also set. In addition, in order to be consistent with the real environment, parameters such as control delay, observation delay, control noise, observation noise, environmental noise, noise of object interaction, and the safe motion range of the robot can be further set.

[0060] In the simulation environment, the robot is controlled to manipulate the target object according to the actions output by the policy model, and the policy model is further trained according to the parameters collected in the simulation environment. Only when the policy model enables the robot to complete the target task in the simulation environment can the policy model be finally deployed in the real environment.

[0061] S104: Training the strategy model according to the robot's state information, the target object's state information and the target parameters.

[0062] Specifically, the policy model can be trained based on the gap between the robot's state information, the target object's state information and the target parameters, so that the robot can perfectly perform the target task.

[0063] Each time the robot performs a target task, the policy model can be trained based on the robot's state information, the target object's state information, and the target parameters. For example, in the table tennis juggling task, each time the robot juggles the ball, the robot's state information and the target object's state information are collected, and the policy model is further trained based on the robot's state information, the target object's state information, and the target parameters, thereby optimizing the policy model.

[0064] This application enables users to only configure various parameters such as the robot's state information, the target object's state information and the target parameters in the upper-level interface, and obtain a strategy model that can complete the target task through training.

[0065] The strategy model of this embodiment is based on reinforcement learning technology, which is the behavior of a robot that constantly interacts and learns with the surrounding environment. Specifically, the robot performs certain actions by observing the state of the surrounding environment and receives reward or punishment signals from the surrounding environment. Based on these feedback signals, the robot can gradually learn and improve its strategy model to obtain better results. There are multiple algorithms in reinforcement learning technology that can iteratively optimize the strategy model, such as PPO, SAC algorithm, etc. These algorithms continuously collect data on the interaction between the robot and the surrounding environment and the artificially set reward function, and calculate the optimization direction until the strategy model enables the robot to complete the target task. For complex target tasks, the strategy model can generally be represented by a neural network, and the complexity of the neural network depends on the task configuration. The duration and complexity of the entire training process depend on multiple factors, such as task complexity, task parameter settings, algorithm selection, algorithm hyperparameters, reward function settings, etc. Therefore, during the training process, it is necessary to adjust these settings according to the training situation so that the training can achieve the desired effect.

[0066] In one embodiment, in the early stage of training, the policy model is unstable and the actions it outputs may not make the robot move. Therefore, in the early stage of training, some action instructions can be randomly generated to make the robot manipulate the target object, and then the policy model is trained according to the robot's state information, the target object's state information and the target parameters. As the training process continues, the policy model becomes more and more stable, and the probability of randomly generating action instructions is correspondingly reduced. Finally, the robot is controlled to manipulate the target object completely according to the output actions of the policy model.

[0067] See also Figure 4 In one implementation, step S104 specifically includes:

[0068] S1041: Determine a reward function based on the robot's state information, the target object's state information, and the target parameters.

[0069] Specifically, the target parameters are the various parameters of the robot when performing the target task under an ideal state. For example, in the table tennis juggling task, the target task is to juggle the table tennis ball to a height of, for example, 30 cm. At this time, the target parameters may include the robot parameters when juggling the table tennis ball to a height of 30 cm (the position and speed of each axis of the robot, the position and speed of the end), and the parameters of the table tennis ball (such as the position and speed of the table tennis ball, etc.). The reward function represents the gap between the robot and the target when currently performing the task.

[0070] For example, in the table tennis bouncing task, the reward function is the relative position, speed, number of bouncing, etc. of the table tennis ball and the table tennis racket.

[0071] S1042: Train the policy model according to the reward function.

[0072] The policy model continuously collects data on the robot's interaction with the surrounding environment and the reward function, and calculates the optimization direction until the policy model is trained to enable the robot to complete the target task.

[0073] Continue reading Figure 4 In one embodiment, after step S104, the following steps are further included:

[0074] S105: After periodically controlling the robot to execute multiple target tasks according to the actions output by the strategy model, the strategy model is evaluated to obtain an evaluation result.

[0075] During the policy model training process, the policy model will be regularly evaluated in the simulation environment. This is because there is noise in the simulation environment, and the obtained policy model may be a probabilistic model. By evaluating the policy model multiple times, multiple evaluation results are obtained to ensure the accuracy of the final policy model output action.

[0076] S106: Retrain the strategy model based on the evaluation results.

[0077] According to the evaluation results, the strategy model is retrained, and the optimal strategy model is obtained after multiple trainings.

[0078] For example, in the task of the robot bouncing a ping-pong ball, each time the robot bouncing the ball, it collects the robot's state information and the target object's state information, and further trains the policy model based on the robot's state information, the target object's state information and the target parameters, and repeats the above process continuously. At the same time, in the above process, the policy model is used regularly or irregularly to make the robot bouncing the ping-pong ball multiple times, and the policy model is evaluated based on the results of the robot bouncing the ping-pong ball multiple times, and then the policy model is retrained based on the evaluation results.

[0079] See also Figure 5 , Figure 5 It is a structural diagram of an implementation of a control device of the present application. The control device 100 includes an acquisition module 110, an input module 120 and a parsing module 130 which are connected in sequence.

[0080] The acquisition module 110 is used to acquire the current first state information of the robot and the current second state information of the target object, wherein the robot is currently manipulating the target object.

[0081] The input module 120 is connected to the acquisition module 110 and is used to input the first state information and the second state information into a pre-trained strategy model to obtain a target action.

[0082] The parsing module 130 is connected to the input module 120 and is used to parse the target action into a control instruction and send the control instruction to the robot so that the robot can manipulate the target object.

[0083] Among them, the control device 100 can be any device that can control the robot. For example, the control device 100 can be a robot control cabinet. Regarding its specific structure, this application does not impose any limitation.

[0084] When the control device 100 is working, it executes the method steps in any of the above-mentioned implementation modes. The detailed method steps can be found in the above-mentioned related contents, which will not be described again here.

[0085] See also Figure 6 , Figure 6 2 is a schematic diagram of another embodiment of the control device 200 of the present application. The control device 200 includes a processor 210 and a memory 220. The processor 210 is coupled to the memory 220. The memory 220 stores program data. The processor 210 implements the steps in any of the above-mentioned implementation methods by executing the program data in the memory 220. The detailed steps can be found in the above-mentioned implementations and will not be repeated here.

[0086] The control device 200 may be a robot control cabinet, and its specific structure is not limited here.

[0087] See also Figure 7 , Figure 7 The computer-readable storage medium 300 stores a computer program 310, which can be executed by a processor to implement the steps in any of the above methods.

[0088] Among them, the computer-readable storage medium 300 can specifically be a device that can store the computer program 310, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or it can also be a server that stores the computer program 310. The server can send the stored computer program 310 to other devices for execution, or it can also execute the stored computer program 310 by itself.

[0089] The above description is only an embodiment of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A robot control method, characterized in that: The method comprises: Acquiring current first state information of the robot and current second state information of the target object, wherein the robot is currently manipulating the target object; Inputting the first state information and the second state information into a pre-trained strategy model to obtain a target action; The target action is parsed into a control instruction, and the control instruction is sent to the robot so that the robot manipulates the target object.

2. The method according to claim 1, characterized in that: The step of inputting the first state information and the second state information into a pre-trained strategy model to obtain a target action includes: Converting the first state information and the second state information into target data according to a preset rule, wherein the target data carries the first state information and the second state information; The target data is input into the policy model to obtain the target action.

3. The method according to claim 1, characterized in that The method further comprises: When a switching instruction is received, the policy model is switched to another policy model.

4. The method according to claim 1, characterized in that: Before parsing the target action into a control instruction, the method further includes: Determining whether the target action meets preset requirements; If the preset requirement is not met, the process returns to the step of inputting the first state information and the second state information into a pre-trained strategy model to obtain a target action.

5. The method according to claim 1, characterized in that Before obtaining the current first state information of the robot and obtaining the current second state information of the target object, the method further includes: Inputting scene parameters of a target task into the strategy model, wherein the target task is for the robot to manipulate the target object, and the scene parameters include at least one of the parameters of the robot, the parameters of the robot end tool, the parameters of the target object, and the parameters of the surrounding environment when the robot manipulates the target object; Setting target parameters of the target task, wherein the target parameters include at least one of a parameter of the robot and a parameter of the target object when the robot manipulates the target object to reach the target; Building a simulation environment, and in the simulation environment, controlling the robot to manipulate the target object according to the output action of the strategy model; The strategy model is trained according to the state information of the robot, the state information of the target object and the target parameters.

6. The method according to claim 5, characterized in that The step of training the strategy model according to the state information of the robot, the state information of the target object and the target parameter comprises: Determining a reward function according to the state information of the robot, the state information of the target object, and the target parameter; The policy model is trained according to the reward function.

7. The method according to claim 5, characterized in that After the strategy model is trained according to the state information of the robot, the state information of the target object and the target parameter, the method further includes: After periodically controlling the robot to perform the target task multiple times according to the actions output by the strategy model, the strategy model is evaluated to obtain an evaluation result; The strategy model is retrained according to the evaluation results.

8. A control device, characterized in that: The control device comprises: An acquisition module, used to acquire current first state information of the robot and current second state information of a target object, wherein the robot is currently manipulating the target object; an input module, connected to the acquisition module, and configured to input the first state information and the second state information into a pre-trained strategy model to obtain a target action; The parsing module is connected to the input module and is used to parse the target action into a control instruction and send the control instruction to the robot so that the robot can manipulate the target object.

9. A control device, characterized in that: The control device includes a processor and a memory, the processor is coupled to the memory, program data is stored in the memory, and the processor implements the steps in the method according to any one of claims 1 to 7 by executing the program data in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program can be executed by a processor to implement the steps in the method according to any one of claims 1 to 7.

Citation Information

Patent Citations

  • Humanoid robot throwing and stability control method based on vision and movement

    CN105128011A

  • Anti-disturbance model training method and device, anti-disturbance model control method and device, equipment, robot and medium

    CN112276950A

  • Legged robot motion control method and device based on meta-reinforcement learning and medium

    CN112684794A

  • Robot control method and device and strategy model training method and device

    CN116945165A

  • Off-line learning for robot control using a reward prediction model

    US20230256593A1

Cited By

  • Servo control system for multi-axis robotic arm for coal mine drilling rig

    WO2026184026A1