Robot control method and system and storage medium

Abstract

The invention discloses a robot control method and system and a storage medium. The method comprises the steps that a speed control instruction and action data of a reference object are obtained, the speed control instruction is used for controlling a moving mode to be adopted in a first moving area of the robot with the preset form, and the first moving area comprises the lower limbs and the waist of the robot with the preset form; the action data is used for determining actions to be executed in a second activity area of the preset form robot, and the second activity area comprises upper limbs of the preset form robot; based on the speed control instruction and the action data, determining a whole-body hybrid control mode of the robot in the preset form; and the robot in the preset form is controlled according to the whole-body mixed control mode. The technical problem that control precision and response flexibility are difficult to consider at the same time when robot control is carried out in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a robot control method, system, and storage medium. Background Technology

[0002] In the wave of convergence between artificial intelligence and robotics, teleoperation of humanoid robots has become a bridge connecting humans and machine intelligence, demonstrating unique advantages, especially in performing high-precision, high-dynamic tasks. Teleoperation technology enables operators to remotely control robots, overcoming environmental limitations and expanding the scope of human operation. Related technologies for achieving comprehensive control of humanoid robots rely on either model-based or learning-based controllers. However, model-based controllers, with their approximate modeling of high-degree-of-freedom humanoid robots, suffer from degraded control performance and struggle to cope with dynamically changing environments. While learning-based controllers can dynamically adapt to the environment, they have high training costs, and their performance is compromised during deployment due to differences between simulation and reality, limiting real-time response capabilities. Therefore, the teleoperation control of humanoid robots faces a combination of challenges: limited accuracy and poor adaptability of model-based control, and high training costs, performance degradation due to differences between simulation and reality, and insufficient real-time response of learning-based control. This limits the robot's ability to operate efficiently and accurately in complex and ever-changing environments.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a robot control method, system, and storage medium to at least solve the technical problem in the related art that it is difficult to simultaneously achieve control accuracy and response flexibility when controlling a robot.

[0005] According to one aspect of the embodiments of this application, a robot control method is provided, comprising: acquiring a speed control command and motion data of a reference object, wherein the speed control command is used to control a movement mode to be taken in a first active area of ​​a preset-form robot, the first active area including: the lower limbs and waist of the preset-form robot, and the motion data is used to determine a movement to be performed in a second active area of ​​the preset-form robot, the second active area including: the upper limbs of the preset-form robot; determining a whole-body hybrid control mode of the preset-form robot based on the speed control command and the motion data; and controlling the preset-form robot according to the whole-body hybrid control mode.

[0006] Optionally, determining the whole-body hybrid control method of the robot with the preset shape based on speed control commands and motion data includes: generating joint torque control commands for the first active area based on speed control commands, and generating joint position control commands for the second active area based on motion data; splicing the joint torque control commands and joint position control commands together to determine the whole-body hybrid control method of the robot with the preset shape.

[0007] Optionally, generating joint torque control commands for the first active area based on speed control commands includes: performing whole-body joint torque analysis on the speed control commands using a target reinforcement learning model to obtain whole-body joint torque control commands for the robot with a preset shape; generating joint torque control commands for the first active area based on a first masking mechanism and the whole-body joint torque control commands, wherein the first masking mechanism is used to mask the joint torque control commands for the second active area.

[0008] Optionally, generating joint position control instructions for the second active area based on motion data includes: performing joint mapping on the body configuration of the preset-shape robot based on motion data to obtain mapping results; using a robot kinematic model to perform whole-body joint position control on the mapping results to obtain whole-body joint position control instructions for the preset-shape robot; and generating joint position control instructions for the second active area based on a second masking mechanism and the whole-body joint position control instructions, wherein the second masking mechanism is used to mask the joint position control instructions for the first active area.

[0009] Optionally, the robot control method further includes: acquiring a training dataset, wherein the training dataset includes: training speed commands set according to a specific task in an offline simulation environment; and using the training dataset to perform reinforcement learning training on an initial reinforcement learning model to obtain a target reinforcement learning model.

[0010] Optionally, training the initial reinforcement learning model using the training dataset to obtain the target reinforcement learning model includes: performing whole-body joint torque analysis on the training dataset and current state observation results using the initial reinforcement learning model to obtain whole-body joint torque training instructions for the robot with a preset shape; generating joint torque training instructions for the first active area based on the third masking mechanism and the whole-body joint torque training instructions, wherein the third masking mechanism is used to mask the joint torque training instructions for the second active area; and training the initial reinforcement learning model using reinforcement learning based on the joint torque training instructions for the first active area and the alternative action training instructions for the second active area to obtain the target reinforcement learning model.

[0011] Optionally, the initial reinforcement learning model is trained using reinforcement learning based on the joint torque training instructions of the first active area and the alternative movement training instructions of the second active area to obtain the target reinforcement learning model. This includes: evaluating the current policy of the initial reinforcement learning model based on the joint torque training instructions of the first active area and the alternative movement training instructions of the second active area to obtain the evaluation result; and improving the current policy of the initial reinforcement learning model based on the evaluation result to obtain the target reinforcement learning model.

[0012] Optionally, alternative motion training instructions are generated using a signal generator and / or selected from a preset motion dataset. The alternative motion training instructions are used to control the second active area to perform various types of motions in the simulation environment, including: static motions, regular swinging motions, and pre-recorded motions.

[0013] According to another aspect of the embodiments of this application, a robot control system is also provided, including: a memory storing an executable program; and a controller for running the program, wherein the program executes any of the robot control methods in the embodiments of this application when it runs.

[0014] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0015] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0016] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0017] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0018] In this embodiment, by acquiring speed control commands and motion data of a reference object, the speed control commands are used to control the movement mode to be taken in the first active area of ​​the preset-form robot. The first active area includes the lower limbs and waist of the preset-form robot. The motion data is used to determine the action to be performed in the second active area of ​​the preset-form robot, which includes the upper limbs of the preset-form robot. Based on the speed control commands and motion data, a full-body hybrid control mode for the preset-form robot is determined. Finally, the preset-form robot is controlled according to the full-body hybrid control mode. Thus, by separating the upper and lower body control strategies of the preset-form robot, high-precision motion replication of the upper limbs and environmental adaptability of the lower limbs are achieved, improving the operational flexibility and environmental adaptability of the preset-form robot. Specifically, by using a full-body hybrid control mode to control the preset-form robot, the upper limbs of the preset-form robot can achieve precise replication of fine movements based on the motion data of the reference object, significantly improving operational accuracy. The control of the lower limbs and waist of the robot in the preset form is achieved through speed control commands, enabling the robot to flexibly adjust its movement mode to adapt to different environmental requirements. This achieves the goal of enabling the upper and lower body of the robot in the preset form to work together through a hybrid control mechanism, thereby improving the robot's control accuracy and flexibility, enhancing the user experience, and solving the technical problem in related technologies where it is difficult to simultaneously achieve control accuracy and response flexibility when controlling a robot. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 This is a flowchart of a robot control method according to an embodiment of this application;

[0021] Figure 2 This is a schematic diagram of a robot control method according to an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of another robot control method according to an embodiment of this application;

[0023] Figure 4 This is a structural block diagram of a robot control device according to an embodiment of this application. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] According to an embodiment of this application, a method embodiment for robot control is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0027] Figure 1 This is a flowchart of a robot control method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0028] Step S11: Obtain speed control instructions and motion data of reference objects. The speed control instructions are used to control the movement mode to be taken in the first active area of ​​the preset shape robot. The first active area includes the lower limbs and waist of the preset shape robot. The motion data is used to determine the action to be performed in the second active area of ​​the preset shape robot. The second active area includes the upper limbs of the preset shape robot.

[0029] Step S12: Based on speed control commands and motion data, determine the full-body hybrid control mode of the preset shape robot;

[0030] Step S13: Control the robot in the preset form according to the whole-body hybrid control method.

[0031] The aforementioned speed control commands can be command signals automatically set by the operator or the system. Their core function is to guide the lower limbs and waist of the robot in a preset form on how to move in three-dimensional space, specifically by specifying the robot's forward speed, turning rate, linear velocity, and angular velocity. Unlike traditional control methods that directly specify joint angles, this embodiment uses speed commands to control the movement mode to be taken in the first active area, allowing the robot in a preset form to flexibly adjust its gait according to environmental changes, thereby demonstrating the flexibility and environmental adaptability of the control strategy.

[0032] The motion data of the aforementioned reference object can be upper limb motion information provided by the operator during actual operation or through a virtual environment, including but not limited to joint angles and rotational speeds, used to map the upper limb movements of a robot with a preset shape. By capturing the details of the operator's upper limb movements, fine-grained control of upper limb movements can be achieved.

[0033] The aforementioned pre-defined robots are robots with specific external shapes and structures, typically designed to resemble humans or similar humanoid forms. This allows them to mimic human behavior patterns when performing tasks, especially in complex environments or situations requiring fine motor skills. Pre-defined robots usually include multiple activity areas, such as upper limbs, lower limbs, and torso, each responsible for different functions and movements.

[0034] The aforementioned first activity area includes the lower limbs and waist of the pre-form robot, primarily responsible for its movement and support. The movement methods to be adopted in this first activity area include, but are not limited to: straight walking, turning, obstacle avoidance, walking uphill and downhill, jumping, or climbing. Specifically, speed control commands can specify the linear velocity of the pre-form robot, allowing it to move smoothly in a straight line. Adjustments using angular velocity commands allow the robot to change its direction of travel, enabling left and right turns. When encountering obstacles in the environment, the pre-form robot needs to quickly adjust its lower limb strides according to the speed control commands to bypass the obstacles and continue moving. On uneven terrain, speed control commands can guide the pre-form robot to adjust its stride size and frequency to adapt to walking on slopes. In special cases, such as crossing deep pits or climbing stairs, speed control commands can include requirements for jump force or climbing speed, enabling the pre-form robot to adopt corresponding movement strategies.

[0035] The aforementioned second activity area includes the upper limbs of the pre-formed robot, focusing on performing fine manipulation tasks. Actions to be performed in this second activity area include, but are not limited to, grasping, placing, and tool manipulation. Specifically, by analyzing motion data from a reference object, the pre-formed robot can mimic an operator's hand movements to precisely grasp and move objects. Conversely, in grasping, the pre-formed robot can place objects in designated locations based on motion data guidance. Furthermore, tasks such as assembling and disassembling using tools like screwdrivers and scissors require the upper limbs to possess the dexterity and strength control to mimic human manipulation. This zoned control strategy maximizes the advantages of each part, improving the overall robot's work efficiency and task execution accuracy.

[0036] Determining a full-body hybrid control method for a robot with a preset shape based on speed control commands and motion data involves fusing these commands and data to develop a control strategy suitable for remote operation and adaptable to environmental changes. Specifically, the speed control commands, including linear and angular velocities, need to be thoroughly analyzed to identify the movement targets of the robot's lower limbs and waist. Real-time calculations ensure that the speed control commands are accurately understood and translated into the robot's movement patterns. Next, the upper limb motion data of the operator or reference object is mapped and analyzed. This data typically includes joint angles, angular velocities, and torques, which must be converted into executable motion commands for the robot's upper limbs to ensure accurate replication of upper limb movements. Finally, the analyzed speed control commands and motion data are fused to determine the full-body hybrid control method.

[0037] For example, during the fusion process, model-based precision control can be applied to the upper limbs to ensure operational accuracy, while learning-based adaptive control is applied to the lower limbs and waist to enhance mobility and stability. In different tasks or environments, the priority of upper and lower body control strategies can be dynamically adjusted according to real-time needs to ensure efficiency and safety. Based on speed control commands and motion data, the overall control strategy of the preset-shape robot can also be optimized in real time to cope with environmental changes, such as uneven ground and obstacles. Using a full-body hybrid control approach to control the preset-shape robot allows it to perform fine-grained tasks in the upper body while enabling the lower body to adapt to various terrains and environmental challenges, effectively improving the robot's operational capabilities and adaptability.

[0038] Based on steps S11 to S13 above, by acquiring speed control commands and motion data of a reference object, the speed control commands are used to control the movement mode to be taken in the first active area of ​​the preset-form robot. The first active area includes the lower limbs and waist of the preset-form robot. The motion data is used to determine the action to be performed in the second active area of ​​the preset-form robot. The second active area includes the upper limbs of the preset-form robot. Then, based on the speed control commands and motion data, the whole-body hybrid control mode of the preset-form robot is determined. Finally, the preset-form robot is controlled according to the whole-body hybrid control mode. Thus, by separating the upper and lower body control strategies of the preset-form robot, high-precision motion replication of the upper limbs and environmental adaptability of the lower limbs are achieved, improving the operational flexibility and environmental adaptability of the preset-form robot. Specifically, by using the whole-body hybrid control mode to control the preset-form robot, the upper limbs of the preset-form robot can achieve accurate replication of fine movements based on the motion data of the reference object, significantly improving the accuracy of operation. The control of the lower limbs and waist of the robot in the preset form is achieved through speed control commands, enabling the robot to flexibly adjust its movement mode to adapt to different environmental requirements. This achieves the goal of enabling the upper and lower body of the robot in the preset form to work together through a whole-body hybrid control method, thereby improving the robot's control accuracy and flexibility, enhancing the user experience, and solving the technical problem in related technologies where it is difficult to simultaneously achieve control accuracy and response flexibility when controlling a robot.

[0039] The robot control method in the embodiments of this application will be further described below.

[0040] In an optional embodiment, in step S12, determining the whole-body hybrid control mode of the preset shape robot based on speed control commands and motion data includes:

[0041] Step S121: Generate joint torque control commands for the first active area based on speed control commands, and generate joint position control commands for the second active area based on motion data;

[0042] Step S122: The joint torque control command and the joint position control command are spliced ​​together to determine the whole-body hybrid control mode of the preset shape robot.

[0043] The aforementioned joint torque control commands are generated based on the speed control commands and are used to directly control the torque of each joint in the first active area, thereby adjusting the movement of the lower limbs and waist of the robot in the preset shape. By adjusting the joint torques, the robot in the preset shape can achieve stable walking, running, or jumping movements to adapt to different ground conditions and execution requirements.

[0044] The aforementioned joint position control commands are generated based on motion data parsing and are used to precisely control the joint positions in the second active area, thereby achieving high-precision control of the upper limb movements of the robot in a preset shape. The joint position control commands can map the operator's upper limb movements, ensuring that the robot's movements in the preset shape are consistent with the operator's intentions. Figure 1 To.

[0045] Furthermore, joint torque control commands and joint position control commands are integrated to form a whole-body hybrid control method. During integration, the two control commands can be spliced ​​together to ensure that the torque control of the lower limbs and the position control of the upper limbs work in coordination, achieving coordinated movement of the robot's entire body. This splicing operation can be implemented using a preset algorithm to ensure the consistency of commands in different activity areas in time and space, thereby avoiding potential conflicts or incoordination in the robot's pre-defined movement patterns.

[0046] In practical applications, such as remote operation scenarios, when an operator sends speed control commands to a robot with a preset shape via a controller, the commands can be quickly parsed to generate joint torque control commands suitable for the first active area, ensuring that the robot's lower limbs and waist can move as expected. Simultaneously, by capturing the operator's upper limb motion data, joint position control commands for the second active area are parsed and generated, enabling the robot's upper limbs to accurately replicate the operator's arm movements. By concatenating the joint torque control commands and the joint position control commands, the robot's upper limb movements are synchronized with the operator's commands during movement, thus achieving high-precision operation and adaptive movement.

[0047] Based on the above optional embodiments, by generating joint torque control commands for the first active area based on speed control commands and joint position control commands for the second active area based on motion data, and then splicing the joint torque control commands and joint position control commands, the whole-body hybrid control mode of the preset shape robot is determined, which significantly improves the flexibility, accuracy and adaptability of the preset shape robot when performing tasks.

[0048] In an optional embodiment, step S121, generating joint torque control commands for the first active region based on speed control commands, includes:

[0049] A target reinforcement learning model is used to perform whole-body joint torque analysis on the speed control command to obtain the whole-body joint torque control command of the robot with the preset shape;

[0050] Based on the first masking mechanism and the whole-body joint torque control commands, joint torque control commands for the first active area are generated. The first masking mechanism is used to mask the joint torque control commands for the second active area.

[0051] The aforementioned target reinforcement learning model is primarily used to enable a pre-defined robot to learn how to maximize cumulative rewards by executing a series of decisions in a given environment. In this embodiment, the target reinforcement learning model can be a learning framework for analyzing speed control commands and generating corresponding whole-body joint torque control commands. The target reinforcement learning model can not only learn based on historical data and experience, but also gradually optimize its decision-making process through simulation and trial and error, ultimately achieving highly efficient and precise control output. Using the target reinforcement learning model to analyze speed control commands enables the pre-defined robot to dynamically adjust its movement strategy in complex and changing environments, achieving intelligent and autonomous walking control.

[0052] Whole-body joint torque analysis is the process of using a target reinforcement learning model to deeply analyze velocity control commands and generate torque control commands for all joints of a robot with a pre-defined shape. Torque is one of the key factors affecting joint movement, directly influencing the power and stability of the robot during operation. Through whole-body joint torque analysis, the robot can autonomously calculate and adjust the torque required for each joint based on different velocity commands, achieving precise control over its movement. Whole-body joint torque analysis combines high-velocity commands with low-level joint torque control, achieving seamless integration from macroscopic commands to microscopic execution, greatly improving the controllability and flexibility of the robot's movement.

[0053] The aforementioned first masking mechanism can filter out instructions related to the first active area from the generated full-body joint torque control instructions, while masking parts unrelated to the second active area. This ensures that the lower body of the robot in the preset form can focus on environmental perception and dynamic adaptation, while the upper body achieves precise operation through other control mechanisms. The first masking mechanism is used to distinguish and isolate the functional requirements of different active areas, avoiding redundant information in the full-body control instructions, reducing computational load, and thus improving the efficiency and accuracy of overall control.

[0054] In practical applications, such as search and rescue missions involving robots with pre-defined shapes, speed control commands, such as forward, backward, and turning, are sent to the robot via remote control. Upon receiving these commands, the target reinforcement learning model optimizes the robot's movement strategy in the current environment. This model can predict and evaluate the impact of different torque control commands on the robot's movement. Through continuous learning and iteration, it ultimately generates a set of full-body joint torque control commands. These commands precisely guide the torque changes of each joint in the robot, enabling effective movement.

[0055] Based on the obtained joint torque control commands for the entire body, a first masking mechanism can be used to filter out commands related to the first active area, while masking commands unrelated to the second active area. This allows the robot in the preset form to focus on processing control signals for the lower body, avoiding distraction or increased computational burden caused by torque commands unrelated to the upper limbs. The joint torque control commands for the first active area generated by the first masking mechanism ensure that the robot's lower body movement is more focused and efficient, contributing to stable walking and rapid response in unstructured environments.

[0056] Based on the above optional embodiments, by using a target reinforcement learning model to perform whole-body joint torque analysis on the speed control command, the whole-body joint torque control command of the preset shape robot is obtained. Then, based on the first mask mechanism and the whole-body joint torque control command, the joint torque control command of the first active area is generated. This not only improves the response speed and adaptability of the lower limbs and waist to the speed control command, but also effectively reduces the interference of upper limb control on the lower body movement strategy. This enables the preset shape robot to maintain stable walking and rapid movement response in complex environments, while retaining the precise operation capability of the upper limbs.

[0057] In an optional embodiment, step S121, generating joint position control commands for the second active area based on motion data includes:

[0058] Based on motion data, joint mapping is performed on the body configuration of the robot with a preset shape to obtain the mapping result;

[0059] The robot kinematics model is used to control the position of all joints in the whole body, and the whole body joint position control commands of the robot with the preset shape are obtained.

[0060] Based on the second masking mechanism and the whole-body joint position control commands, joint position control commands for the second active area are generated. The second masking mechanism is used to mask the joint position control commands for the first active area.

[0061] The aforementioned body configuration can represent the structural attributes of a pre-defined robot, including but not limited to the position, connection method, and degrees of freedom of each joint. In this embodiment, the body configuration provides a hardware foundation for the separation and integration of upper and lower body control of the pre-defined robot, thereby enabling control of the upper limbs through motion data mapping.

[0062] Joint mapping is the process of converting upper limb joint information from motion data into motion commands for the corresponding joints in the second active area of ​​a pre-defined robot. The joint mapping process relies on precise mapping relationships to ensure that the operator's movements are correctly and accurately translated into robot upper limb movements. Through joint mapping, even if the joint layout and number of the humanoid robot and the operator are not exactly the same, natural and accurate motion transmission can be achieved.

[0063] The above mapping results are the output of the joint mapping process, including parameters such as joint position and angular velocity in the second active area of ​​the robot with the preset shape. The mapping results reflect the equivalent representation of the operator's actions on the robot with the preset shape. The mapping results are a prerequisite for further using the robot's kinematic model for whole-body joint position control.

[0064] The aforementioned robot kinematic model is used to describe the relationships between the joint positions of a robot in a preset form and its response to external inputs. In this embodiment, the robot kinematic model receives the mapping results and, based on the body configuration of the robot in the preset form, calculates the joint position control commands for the entire body, ensuring that the robot in the preset form can execute a predetermined sequence of actions. The robot kinematic model can provide precise and efficient control commands, which is key to achieving high-precision operation of the robot arm in the preset form.

[0065] The whole-body joint position control process can be based on the robot's kinematic model, setting specific position targets for each joint of the robot in a preset shape. Through whole-body joint position control, the upper limbs of the robot in a preset shape can be directed to accurately reach the preset position to perform fine operations such as grasping, placing, and assembling, while maintaining stable support for the lower body.

[0066] The aforementioned second masking mechanism is used to filter out commands related to the second active area from the full-body joint position control commands, while simultaneously masking irrelevant control commands for the first active area. This second masking mechanism ensures that the control commands for the upper limbs of the preset-shape robot are not interfered with by the lower body movement strategy, allowing the upper limbs to focus on performing precise tasks while the lower body maintains independent gait control. This further enhances the specificity and efficiency of the control commands, reduces redundancy in the data processing process, and improves the operational accuracy and flexibility of the preset-shape robot.

[0067] When an operator performs a specific task, such as operating tools or moving materials, motion data is collected in real time through a capture device and then mapped to joints to ensure that the upper limbs of the robot in the preset form can accurately map the operator's movements, thereby converting human gestures, strength, and operational skills into instructions that the robot in the preset form can execute.

[0068] Once the mapping results are ready, the optimal positions of all joints in the body are calculated using a robot kinematic model to achieve precise upper body manipulation. This involves complex mathematical calculations and physical simulations, ensuring that each joint moves along the optimal path, thus achieving high efficiency and accuracy in the operation. A second masking mechanism filters control commands for the entire body's joint positions, allowing commands related to the second active area to pass through while preventing commands unrelated to the first active area from entering the upper limb control chain. This keeps the upper limb control commands pure, unaffected by unnecessary lower body movements, significantly improving the upper limb's focus and accuracy during task execution.

[0069] Based on the above optional embodiments, the body configuration of the preset shape robot is mapped by motion data to obtain the mapping result. Then, the robot kinematics model is used to control the position of the whole body joints of the mapping result to obtain the whole body joint position control command of the preset shape robot. Finally, based on the second mask mechanism and the whole body joint position control command, the joint position control command of the second active area is generated, thereby further improving the flexibility and reliability of the preset shape robot.

[0070] In an optional embodiment, the robot control method of this application further includes:

[0071] Obtain the training dataset, which includes: training speed instructions set according to a specific task in an offline simulation environment;

[0072] The initial reinforcement learning model is trained using the training dataset to obtain the target reinforcement learning model.

[0073] The aforementioned training dataset is a series of input-output data pairs generated in an offline simulation environment based on a preset specific task objective, used for training the reinforcement learning model. In this embodiment, the training dataset includes training speed commands, which can simulate a set of speed commands sent by an operator or system to a robot of a preset shape in different task scenarios. The training speed commands may include the linear velocity and angular velocity of the robot of the preset shape, used to guide its movement and turning.

[0074] The aforementioned initial reinforcement learning model, before training begins, uses a pre-defined model architecture and parameters as the starting point for reinforcement learning training. This initial reinforcement learning model can typically be a model that has not been trained on any data, or a preliminary model based on existing knowledge and algorithms. In this embodiment, the initial reinforcement learning model needs to be learned and adjusted using a training dataset to adapt to the control requirements of a robot with a preset shape.

[0075] The aforementioned target reinforcement learning model is the final reinforcement learning model obtained after learning and optimization on a sufficient training dataset. Compared with the initial reinforcement learning model, the target reinforcement learning model has a more refined control strategy, can better respond to speed control commands, and achieves efficient movement of the first active area in complex environments.

[0076] During reinforcement learning training, the initial reinforcement learning model interacts with the simulation environment, trying different actions to respond to training speed commands, while receiving reward or penalty signals from the environment. The initial reinforcement learning model uses these signals to adjust its strategy, learning how to maximize rewards when performing specific tasks—that is, finding the most suitable joint torque output under various speed commands. After multiple iterations and optimizations, the final target reinforcement learning model can accurately predict and execute movement strategies matching speed commands in actual deployment, significantly improving the performance of robots with pre-defined shapes in complex tasks.

[0077] Based on the above optional embodiments, by acquiring a training dataset and then using the training dataset to train the initial reinforcement learning model, a target reinforcement learning model is obtained. This significantly improves the flexibility, adaptability, and robustness of the robot's movement in the preset form, reduces the trial-and-error costs in actual operation, and ensures that the robot in the preset form can quickly respond to the operator's instructions and accurately reach the designated position during task execution.

[0078] In one optional embodiment, the initial reinforcement learning model is trained using a training dataset to obtain the target reinforcement learning model, which includes:

[0079] An initial reinforcement learning model is used to perform whole-body joint torque analysis on the training dataset and current state observation results to obtain whole-body joint torque training instructions for the robot with the preset shape.

[0080] Based on the third masking mechanism and the whole-body joint torque training instructions, joint torque training instructions for the first active area are generated. The third masking mechanism is used to shield the joint torque training instructions for the second active area.

[0081] Based on the joint torque training instructions of the first active area and the alternative movement training instructions of the second active area, the initial reinforcement learning model is trained to obtain the target reinforcement learning model.

[0082] The aforementioned current state observation results refer to the information about itself and its surrounding environment obtained by the robot with the pre-defined shape through sensors at each time step, including but not limited to joint angles, velocity, acceleration, and the position of obstacles in the environment. The current state observation results are an important basis for the initial reinforcement learning model to make decisions, enabling the initial reinforcement learning model not only to learn from the training dataset, but also to adjust its strategy according to real-time environmental changes, thereby improving the robot with the pre-defined shape's adaptability to unexpected situations.

[0083] The aforementioned third masking mechanism can filter out instructions related to the first range of motion from the full-body joint torque training instructions, while simultaneously masking joint torque instructions for the second range of motion. This third masking mechanism ensures that the reinforcement learning process can focus on optimizing the mobility of the lower limbs and waist, without being distracted by the control needs of the upper body, effectively improving the efficiency and focus of training.

[0084] The aforementioned alternative motion training instructions refer to a series of simulated motion instructions designed for the second activity area during training. These instructions are used to place the robot's upper limbs in simulated real-world working conditions without directly training the upper limb control strategy. Alternative motion training instructions include, but are not limited to, remaining stationary, regular swinging, and pre-recorded different types of movements. The aim is to enrich the training scenarios, enhance the effectiveness of training instructions in the first activity area, and avoid unnecessary impact on upper limb control.

[0085] In the simulation environment, the pre-defined robot, through an initial reinforcement learning model, performs full-body joint torque analysis based on the training dataset and current state observations to obtain torque control commands for each joint under different task scenarios. This process simulates the robot's perception and response to environmental changes when performing a specific task, providing detailed dynamic guidance for subsequent training.

[0086] Subsequently, through the third masking mechanism, torque commands specific to the lower limbs and waist are extracted from the whole-body joint torque training commands, while upper limb-related commands are masked. This allows the initial reinforcement learning model to focus on optimizing the movement strategy of the first active area without being disturbed by upper body movements, thereby ensuring the effective allocation of learning resources and the precise improvement of the target area control strategy.

[0087] Finally, based on the joint torque training commands for the first active area and the alternative movement training commands for the second active area, the initial reinforcement learning model was trained using reinforcement learning. Through multiple trials and learning, the strategy was gradually adjusted to more accurately control the lower limbs and waist when faced with different speed commands and environmental changes. Meanwhile, the diversity of upper limb movements was maintained through alternative commands, without interfering with the training process of the lower limbs. After sufficient training, the initial reinforcement learning model evolved into a target reinforcement learning model, significantly improving the performance of the robot in motion control within the preset form.

[0088] Based on the above optional embodiments, by using an initial reinforcement learning model to perform whole-body joint torque analysis on the training dataset and the current state observation results, the whole-body joint torque training instructions for the preset shape robot are obtained. Then, based on the third mask mechanism and the whole-body joint torque training instructions, the joint torque training instructions for the first active area are generated. Finally, the initial reinforcement learning model is trained by reinforcement learning based on the joint torque training instructions for the first active area and the alternative action training instructions for the second active area. The resulting target reinforcement learning model can respond to speed control instructions more accurately and flexibly, significantly enhancing the walking stability and adaptability of the preset shape robot in complex terrain and dynamic environments.

[0089] In one optional embodiment, the initial reinforcement learning model is trained using reinforcement learning based on the joint torque training instructions of the first active area and the alternative movement training instructions of the second active area to obtain the target reinforcement learning model, including:

[0090] Based on the joint torque training instructions of the first active area and the alternative movement training instructions of the second active area, the current policy of the initial reinforcement learning model is evaluated to obtain the evaluation results;

[0091] Based on the evaluation results, the current policy of the initial reinforcement learning model is improved to obtain the target reinforcement learning model.

[0092] In reinforcement learning, policy evaluation is a crucial step in the model training process. It involves analyzing the performance of the initial reinforcement learning model's current policy based on training instructions for joint torques in the first active region and training instructions for alternative actions in the second active region. The evaluation process typically simulates various task scenarios and environmental conditions, collecting behavioral data of the initial reinforcement learning model in each scenario, including factors such as the probability of successfully completing the task, the accuracy of the actions, and energy consumption, to quantify the merits of the current policy.

[0093] Policy improvement is the process of adjusting and optimizing the current policy of the initial reinforcement learning model based on the policy evaluation results. Policy improvement aims to identify deficiencies in the model's behavior based on evaluation feedback and update the policy through algorithms to improve the control accuracy, efficiency, and robustness of the initial reinforcement learning model in the first active area. Improved strategies include, but are not limited to, updating model parameters, adjusting the learning rate, and introducing new reward functions, with the goal of enabling the initial reinforcement learning model to perform better when facing unknown future challenges. Through the policy improvement process, the initial reinforcement learning model iterates step by step until the target reinforcement learning model exhibits higher accuracy, faster response speed, and stronger environmental adaptability in the control of the first active area. While improving the policy, it also ensures that the movement diversity in the second active area is not affected, maintaining the flexibility and precision of the upper limbs when performing tasks.

[0094] Based on the above optional embodiments, the current strategy of the initial reinforcement learning model is evaluated according to the joint torque training instructions of the first active area and the alternative action training instructions of the second active area. The evaluation results are obtained, and then the current strategy of the initial reinforcement learning model is improved based on the evaluation results to obtain the target reinforcement learning model. This ensures the scientificity and effectiveness of the model training process and further improves the model performance.

[0095] In one alternative embodiment, the alternative motion training instructions are generated using a signal generator and / or selected from a preset motion dataset. The alternative motion training instructions are used to control the second active area to perform various types of motions in the simulation environment, including: stationary motions, regular swinging motions, and pre-recorded motions.

[0096] The aforementioned signal generator can generate random signals for use as alternative action training commands. These random signals represent various dynamic conditions that the upper limbs may encounter, such as sudden stops, slight vibrations, or swings at specific frequencies. The signal generator can simulate the uncertainties that the upper limbs may experience in the real world, making the pre-defined robot more closely resemble the real operating environment during training, thereby improving the robustness and adaptability of upper limb movements.

[0097] The aforementioned pre-recorded action dataset is a database containing a series of pre-recorded, stereotyped movements. During reinforcement learning training, alternative action training instructions can be selected from this dataset. The use of pre-recorded movements ensures that the upper limbs perform a series of validated, task-related actions during training, such as mimicking human arm movements when grasping, carrying, or assembling objects. The pre-recorded action dataset provides a rich behavioral library for upper limb training, contributing to improved proficiency and accuracy in performing specific tasks.

[0098] In alternative motion training instructions, static motions allow the upper limbs to remain stationary in a simulated environment, helping to simulate the static state of the robot's upper limbs when the operator is observing or thinking. Regular swinging involves the upper limbs swinging according to a pre-defined pattern, such as simulating the natural swing of the arm during human walking, or the periodic movements of the upper limbs when performing repetitive tasks. Pre-recorded motions are specific motion sequences selected from a pre-set motion dataset, used to train the upper limbs' ability to perform specific tasks. The introduction of these three motion types provides a comprehensive dynamic range for upper limb training, ensuring that the upper limbs exhibit appropriate dynamic behavior in any task scenario, enhancing the flexibility and realism of the robot in performing tasks.

[0099] Figure 2 This is a schematic diagram of a robot control method according to an embodiment of this application, such as... Figure 2 As shown, a speed control command is acquired, which is used to control the movement mode to be taken in the first active area of ​​the robot with a preset shape. A target reinforcement learning model is used to perform whole-body joint torque analysis on the speed control command, and infer and output the whole-body joint torque control command of the robot with the preset shape. Based on the first mask mechanism and the whole-body joint torque control command, the joint torque control command of the first active area is generated. The first mask mechanism is used to mask the joint torque control command of the second active area. During the training of the target reinforcement learning model in the simulation environment, the initial reinforcement learning model is used to perform whole-body joint torque analysis on the training dataset and the current state observation results to obtain the whole-body joint torque training command of the robot with the preset shape. Then, based on the third mask mechanism and the whole-body joint torque training command, the joint torque training command of the first active area is generated. Then, based on the joint torque training command of the first active area and the alternative action training command of the second active area, the initial reinforcement learning model is trained by reinforcement learning to obtain the target reinforcement learning model. Alternative motion training instructions are generated by a signal generator and / or randomly selected from a preset motion dataset. These alternative motion training instructions are used to control the second active area to perform various types of motions in the simulation environment, including: stationary motions, regular swinging motions, and pre-recorded motions.

[0100] Figure 3 This is a schematic diagram of another robot control method according to an embodiment of this application, such as... Figure 3As shown, the system acquires speed control commands and motion data of a reference object. A target reinforcement learning model is used to perform full-body joint torque analysis on the speed control commands, resulting in full-body joint torque control commands for a pre-defined robot. Based on a first masking mechanism and the full-body joint torque control commands, joint torque control commands for the first active region are generated. The first masking mechanism is used to mask the joint torque control commands for the second active region. The pre-defined robot is controlled using a full-body hybrid control method. Joint mapping is performed on the body configuration of the pre-defined robot based on the motion data, yielding the mapping results. A robot kinematics model is used to perform full-body joint position control on the mapping results, resulting in full-body joint position control commands for the pre-defined robot. Based on a second masking mechanism and the full-body joint position control commands, joint position control commands for the second active region are generated. The second masking mechanism is used to mask the joint position control commands for the first active region. The joint torque control commands and joint position control commands are concatenated to determine the full-body hybrid control method for the pre-defined robot. The underlying controller can control the pre-defined robot using this full-body hybrid control method.

[0101] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0102] According to an embodiment of this application, an apparatus embodiment for a robot control method is provided. It should be noted that the apparatus can be used to execute the above-described robot control method.

[0103] Figure 4 This is a structural block diagram of a robot control device according to an embodiment of this application, such as... Figure 4 As shown, the device includes:

[0104] The acquisition module 401 is used to acquire speed control instructions and motion data of a reference object. The speed control instructions are used to control the movement mode to be taken in the first active area of ​​the preset shape robot. The first active area includes the lower limbs and waist of the preset shape robot. The motion data is used to determine the action to be performed in the second active area of ​​the preset shape robot. The second active area includes the upper limbs of the preset shape robot.

[0105] The determination module 402 is used to determine the whole-body hybrid control mode of the robot with a preset shape based on speed control commands and motion data;

[0106] The control module 403 is used to control the robot in a preset form according to the whole-body hybrid control method.

[0107] Optionally, the determining module 402 is further configured to: generate joint torque control commands for the first active area based on speed control commands, and generate joint position control commands for the second active area based on motion data; and splice the joint torque control commands and joint position control commands to determine the whole-body hybrid control mode of the preset shape robot.

[0108] Optionally, the determining module 402 is further configured to: perform whole-body joint torque analysis on the speed control command using a target reinforcement learning model to obtain the whole-body joint torque control command of the robot with the preset shape; and generate the joint torque control command of the first active area based on the first masking mechanism and the whole-body joint torque control command, wherein the first masking mechanism is used to shield the joint torque control command of the second active area.

[0109] Optionally, the determining module 402 is further configured to: perform joint mapping on the body configuration of the preset shape robot based on motion data to obtain mapping results; use a robot kinematic model to perform whole-body joint position control on the mapping results to obtain whole-body joint position control commands for the preset shape robot; and generate joint position control commands for the second active area based on the second masking mechanism and the whole-body joint position control commands, wherein the second masking mechanism is used to shield the joint position control commands for the first active area.

[0110] Optionally, the acquisition module 401 is also used to acquire a training dataset, wherein the training dataset includes: training speed instructions set according to a specific task in an offline simulation environment; the robot control device further includes: a training module 404, used to perform reinforcement learning training on the initial reinforcement learning model using the training dataset to obtain a target reinforcement learning model.

[0111] Optionally, the training module 404 is further configured to: perform whole-body joint torque analysis on the training dataset and current state observation results using the initial reinforcement learning model to obtain whole-body joint torque training instructions for the robot with a preset shape; generate joint torque training instructions for the first active area based on the third masking mechanism and the whole-body joint torque training instructions, wherein the third masking mechanism is used to mask the joint torque training instructions for the second active area; and perform reinforcement learning training on the initial reinforcement learning model based on the joint torque training instructions for the first active area and the alternative action training instructions for the second active area to obtain the target reinforcement learning model.

[0112] Optionally, the training module 404 is further configured to: evaluate the current policy of the initial reinforcement learning model based on the joint torque training instructions of the first active area and the alternative action training instructions of the second active area, and obtain the evaluation result; and improve the current policy of the initial reinforcement learning model based on the evaluation result to obtain the target reinforcement learning model.

[0113] Optionally, alternative motion training instructions are generated using a signal generator and / or selected from a preset motion dataset. The alternative motion training instructions are used to control the second active area to perform various types of motions in the simulation environment, including: static motions, regular swinging motions, and pre-recorded motions.

[0114] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0115] Embodiments of this application also provide a robot control system, including: a memory storing an executable program; and a controller for running the program, wherein the program executes any of the robot control methods described in the embodiments of this application during runtime.

[0116] Optionally, in this embodiment, the controller can be configured to perform the following steps via a computer program:

[0117] S1, acquire speed control instructions and motion data of reference objects, wherein the speed control instructions are used to control the movement mode to be taken in the first active area of ​​the preset shape robot, the first active area includes: the lower limbs and waist of the preset shape robot, and the motion data is used to determine the action to be performed in the second active area of ​​the preset shape robot, the second active area includes: the upper limbs of the preset shape robot.

[0118] S2, based on speed control commands and motion data, determines the whole-body hybrid control mode of the robot in the preset shape;

[0119] S3 controls the robot in a preset form using a full-body hybrid control method.

[0120] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0121] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0122] S1, acquire speed control instructions and motion data of reference objects, wherein the speed control instructions are used to control the movement mode to be taken in the first active area of ​​the preset shape robot, the first active area includes: the lower limbs and waist of the preset shape robot, and the motion data is used to determine the action to be performed in the second active area of ​​the preset shape robot, the second active area includes: the upper limbs of the preset shape robot.

[0123] S2, based on speed control commands and motion data, determines the whole-body hybrid control mode of the robot in the preset shape;

[0124] S3 controls the robot in a preset form using a full-body hybrid control method.

[0125] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0126] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0127] Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of this application.

[0128] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0129] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0130] 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 units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0131] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0132] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or 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 methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0133] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A robot control method, characterized in that, include: Acquire speed control commands and motion data of a reference object, wherein the speed control commands are used to control the movement mode to be taken in the first active area of ​​the preset shape robot, the first active area including: the lower limbs and waist of the preset shape robot, and the motion data is used to determine the action to be performed in the second active area of ​​the preset shape robot, the second active area including: the upper limbs of the preset shape robot. Based on the speed control command and the motion data, the whole-body hybrid control mode of the preset shape robot is determined; The robot in the preset form is controlled according to the whole-body hybrid control method.

2. The robot control method according to claim 1, characterized in that, Based on the speed control command and the motion data, the whole-body hybrid control mode of the preset shape robot is determined to include: Based on the speed control command, a joint torque control command for the first active area is generated, and based on the motion data, a joint position control command for the second active area is generated; The joint torque control command and the joint position control command are spliced ​​together to determine the whole-body hybrid control mode of the preset shape robot.

3. The robot control method according to claim 2, characterized in that, Generating the joint torque control command for the first active area based on the speed control command includes: A target reinforcement learning model is used to perform whole-body joint torque analysis on the speed control command to obtain the whole-body joint torque control command of the robot with the preset shape. Based on the first masking mechanism and the whole-body joint torque control command, the joint torque control command for the first active area is generated, wherein the first masking mechanism is used to mask the joint torque control command for the second active area.

4. The robot control method according to claim 2, characterized in that, Generating the joint position control command for the second active area based on the motion data includes: Based on the motion data, joint mapping is performed on the body configuration of the preset-shape robot to obtain the mapping result; The robot kinematics model is used to perform whole-body joint position control on the mapping results to obtain the whole-body joint position control command of the robot with the preset shape; Based on the second masking mechanism and the whole-body joint position control commands, the joint position control commands for the second active area are generated, wherein the second masking mechanism is used to mask the joint position control commands for the first active area.

5. The robot control method according to claim 3, characterized in that, The robot control method further includes: Obtain a training dataset, wherein the training dataset includes: training speed instructions set according to a specific task in an offline simulation environment; The initial reinforcement learning model is trained using the training dataset to obtain the target reinforcement learning model.

6. The robot control method according to claim 5, characterized in that, The initial reinforcement learning model is trained using the training dataset to obtain the target reinforcement learning model, which includes: The initial reinforcement learning model is used to perform whole-body joint torque analysis on the training dataset and the current state observation results to obtain the whole-body joint torque training instructions for the robot with the preset shape. Based on the third masking mechanism and the whole-body joint torque training instructions, joint torque training instructions for the first active area are generated, wherein the third masking mechanism is used to shield the joint torque training instructions for the second active area. Based on the joint torque training instructions of the first active area and the alternative movement training instructions of the second active area, the initial reinforcement learning model is trained by reinforcement learning to obtain the target reinforcement learning model.

7. The robot control method according to claim 6, characterized in that, Based on the joint torque training instructions of the first active area and the alternative movement training instructions of the second active area, the initial reinforcement learning model is trained using reinforcement learning to obtain the target reinforcement learning model, which includes: Based on the joint torque training instructions of the first active area and the alternative action training instructions of the second active area, the current policy of the initial reinforcement learning model is evaluated to obtain the evaluation result; Based on the evaluation results, the current policy of the initial reinforcement learning model is improved to obtain the target reinforcement learning model.

8. The robot control method according to claim 6, characterized in that, The alternative action training instructions are generated using a signal generator and / or selected from a preset action dataset. The alternative action training instructions are used to control the second active area to perform various types of actions in the simulation environment. The various types of actions include: static actions, regular swinging, and pre-recorded actions.

9. A robot control system, characterized in that, include: Memory, which stores executable programs; A controller for running the program, wherein the program executes the robot control method according to any one of claims 1 to 8 when it runs.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the robot control method according to any one of claims 1 to 8.

Images