Robot control system and method based on hierarchical collaborative architecture and robot

Abstract

The invention belongs to the technical field of robot control, and particularly relates to a robot control system and method based on a hierarchical collaborative architecture and a robot. The system comprises a brain module deployed at a cloud end and used for performing long-range task planning, decomposing a target task instruction into sub-tasks and sending the sub-tasks to a cerebellum module; the cerebellum module with the body is deployed on the robot body and used for controlling the robot to act and execute the subtask according to the received subtask; the system performs communication through a double-track communication architecture, and comprises a high-frequency data channel, which adopts a shared memory mechanism and is used for data exchange between a cerebellum module with a body and a sensor and an actuator of a robot; and the low-frequency instruction channel adopts network communication based on sockets and is used for task-level information interaction between the brain module and the cerebellum module. According to the invention, the cooperative problem of high-level cognition and bottom-level control is solved, and low delay, high reliability and high autonomy of robot control are realized.

Description

Technical Field

[0001] This application belongs to the field of robot control technology, specifically relating to a robot control system, method, and robot based on a hierarchical collaborative architecture. Background Technology

[0002] The operation and maintenance of nuclear facilities (such as nuclear fuel production lines and nuclear power plants) require extremely high levels of safety and reliability. A large number of maintenance tasks still rely on manual labor in harsh environments with high radiation levels and confined spaces, which is inefficient, increases the risk of human error, and poses health and safety hazards to personnel. To address this issue, the industry has introduced robotics technology.

[0003] However, traditional pre-programmed industrial robots rely on pre-set fixed trajectories for operation. Their structure is usually a fixed robotic arm or a track-based automated guided vehicle (AGV), which is ill-suited for the numerous unstructured and dynamically changing maintenance tasks within nuclear facilities (such as handling accidentally dropped tools or inspecting equipment for corrosion in non-standard locations). Their decision-making capabilities are virtually zero, making them unable to respond to any unforeseen circumstances beyond the pre-programmed procedures.

[0004] While teleoperated or semi-autonomous robots based on traditional control architectures possess some perception capabilities, their control architecture typically follows a serial "perception-planning-control" model. The communication link between high-level task planning modules (e.g., based on visual target recognition) and low-level motion control modules (e.g., servo motor control) is long and has high latency, typically tens to hundreds of milliseconds. This fragmented architecture leads to sluggish robot responses and disjointed movements in dynamic environments. For example, when a robot is performing a grasping task, if the target object suddenly moves slightly, the entire "perception-planning-execution" process needs to be repeated, resulting in task interruption and inefficiency. Furthermore, these systems lack deep semantic understanding of complex and ambiguous human commands, resulting in poor task generalization. Summary of the Invention

[0005] The purpose of this application is to provide a robot control system, method and robot based on a hierarchical collaborative architecture, to solve the problem of collaboration between high-level cognitive decision-making and low-level real-time motion control in robots in the prior art, which leads to high response latency, poor autonomy and insufficient reliability of robots in complex nuclear facility environments.

[0006] The technical solution to achieve the purpose of this application is as follows: The first aspect of this application provides a robot control system based on a hierarchical collaborative architecture, the system comprising: an embodied brain module and an embodied cerebellum module; The embodied brain module is deployed in the cloud and is used for long-term task planning, decomposing the target task instructions into sub-tasks and sending them to the embodied cerebellum module. The embodied cerebellum module is deployed on the robot body and is used to control the robot's actions to execute the sub-task according to the received sub-task; The system communicates via a dual-track communication architecture, which includes a high-frequency data channel and a low-frequency command channel. The high-frequency data channel adopts a shared memory mechanism for data exchange between the embodied cerebellum module and the robot's sensors and actuators; The low-frequency command channel uses socket-based network communication for task-level information interaction between the embodied brain module and the embodied cerebellum module.

[0007] Optionally, the embodied brain module performs long-term task planning based on a visual-language-action model.

[0008] Optionally, the embodied cerebellum module is specifically used to parse the received sub-task into the robot's whole-body motion control commands, read the sensor's status feedback in real time through the high-frequency data channel, and drive the actuator to complete the corresponding specific action.

[0009] Optionally, the embodied cerebellum module is further configured to determine the execution status of the subtask based on the status feedback of the sensor, and feed back the execution status to the embodied brain module.

[0010] Optionally, the embodied brain module is further configured to perform dynamic task replanning on the target task instructions based on the execution state.

[0011] Optionally, the cloud is an embedded AI platform or a computing server.

[0012] Optionally, the low-frequency command channel is specifically a TCP / IP connection established via an Ethernet network; the high-frequency data channel is implemented using the physical memory of the robot controller.

[0013] The second aspect of this application provides a robot control method based on a hierarchical collaborative architecture, applicable to any one of the robot control systems based on a cerebellar-macrocerebral hierarchical collaborative architecture provided in the first aspect of this application; the method includes: Step S101: The embodied brain module receives the target task instruction and decomposes the target task instruction into sub-tasks; Step S102: The embodied brain module sends a command to the embodied cerebellum module via a low-frequency command channel; the low-frequency command channel uses socket-based network communication for task-level information interaction between the embodied brain module and the embodied cerebellum module. Step S103: The embodied cerebellum module controls the robot to perform the sub-task by using a high-frequency data channel according to the received sub-task; the high-frequency data channel adopts a shared memory mechanism for data exchange between the embodied cerebellum module and the robot's sensors and actuators.

[0014] Optionally, step S101 specifically includes: The embodied brain module performs long-term task planning based on a visual-language-action model.

[0015] Optionally, step S103 specifically includes: The embodied cerebellum module parses the received sub-tasks into full-body motion control commands for the robot; The embodied cerebellum module reads the status feedback of the sensor in real time through the high-frequency data channel; The embodied cerebellum module drives the actuator to complete the corresponding specific action through the high-frequency data channel based on the whole-body motion control command and the status feedback.

[0016] Optionally, step S103 may be followed by: Step S104: The embodied cerebellum module determines the execution status of the subtask based on the status feedback from the sensor; Step S105: The embodied cerebellum module feeds back the execution status to the embodied brain module.

[0017] Optionally, step S105 may be followed by: Step S106: The embodied brain module performs dynamic task replanning on the target task instruction based on the execution state.

[0018] A third aspect of this application provides a robot, including: a controller, sensors, and actuators; the controller is equipped with an embodied cerebellum module, which is one of the robot control systems based on a hierarchical collaborative architecture provided in the first aspect of this application. The embodied cerebellum module interacts with the embodied brain module at the task level through a low-frequency command channel and exchanges data with the sensor and the actuator through a high-frequency data channel. The low-frequency instruction channel uses socket-based network communication, while the high-frequency data channel uses a shared memory mechanism.

[0019] Optionally, the high-frequency data channel is implemented using the physical memory of the controller.

[0020] The fourth aspect of this application provides a computer, wherein the computer is equipped with an embodied brain module, which is one of the robot control systems based on a hierarchical collaborative architecture provided in the first aspect of this application.

[0021] The beneficial technical effects of this application are as follows: This application provides a robot control system, method, and robot based on a hierarchical collaborative architecture. The system includes an embodied brain module and an embodied cerebellum module. The embodied brain module is deployed in the cloud and is used for long-term task planning, decomposing target task instructions into sub-tasks and sending them to the embodied cerebellum module. The embodied cerebellum module is deployed on the robot body and is used to control the robot's actions to execute the sub-tasks according to the received sub-tasks. The system communicates through a dual-track communication architecture. The dual-track communication architecture includes a high-frequency data channel and a low-frequency command channel. The high-frequency data channel adopts a shared memory mechanism for data exchange between the embodied cerebellum module and the robot's sensors and actuators. The low-frequency command channel adopts socket-based network communication for task-level information interaction between the embodied brain module and the embodied cerebellum module. This application embodiment implements a high-frequency control loop through shared memory, achieving a response time in milliseconds, thus ensuring the smoothness and accuracy of the robot's movements in dynamic environments. At the same time, it uses a low-frequency command channel for task-level asynchronous communication, perfectly solving the problem of coordination between high-level cognition and low-level control, and opening up the entire link from understanding complex commands to executing precise actions, thereby achieving low latency, high reliability, and high autonomy in robot control. Attached Figure Description

[0022] Figure 1 A schematic diagram of a robot control system based on a hierarchical collaborative architecture is provided for an embodiment of this application; Figure 2 This is a flowchart illustrating a robot control method based on a hierarchical collaborative architecture, provided as an embodiment of this application. Detailed Implementation

[0023] To enable those skilled in the art to better understand this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only a part of the embodiments of this application, and not all of them. Based on the embodiments described in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] See Figure 1 The figure is a schematic diagram of a robot control system based on a hierarchical collaborative architecture provided in an embodiment of this application.

[0025] This application provides a robot control system based on a hierarchical collaborative architecture, including: an embodied brain module 100 and an embodied cerebellum module 200; The embodied brain module 100 is deployed in the cloud and is used for long-term task planning, breaking down the target task instructions into sub-tasks and sending them to the embodied cerebellum module 200. The cerebellum module 200 is deployed on the robot 300 body and is used to control the robot 300 to perform sub-tasks according to the received sub-tasks. The system communicates via a dual-track communication architecture, which includes a high-frequency data channel 400 and a low-frequency command channel 500. The high-frequency data channel 400 adopts a shared memory mechanism for data exchange between the embodied cerebellum module 200 and the sensors and actuators of the robot 300; The low-frequency command channel 500 uses socket-based network communication for task-level information exchange between the embodied brain module 100 and the embodied cerebellum module 200.

[0026] In this embodiment, the embodied brain module 100 serves as the robot's cognitive center, responsible for handling low-frequency but highly complex cognitive tasks. The embodied brain module 100 can receive and deeply understand complex task instructions, such as "check the corrosion of the low-material container," given by humans in natural language. It also integrates multimodal sensor information (such as vision and LiDAR) to semantically understand the working environment, thereby performing long-term task planning. This intelligently decomposes an abstract maintenance goal into a series of logically clear and executable sub-task instructions. In one example, the embodied brain module 100 can perform long-term task planning based on a Visual-Language-Action (VLA) model. Based on the VLA large model, the embodied brain module 100 endows the robot 300 with the ability to understand complex and ambiguous instructions and autonomously plan and execute them, eliminating the need for tedious programming for each specific task and greatly improving its task adaptability and generalization ability in unstructured scenarios.

[0027] The embodied cerebellum module 200, serving as the motion control center of the robot 300, is responsible for handling high-frequency, low-latency real-time control tasks. It receives sub-task instructions from the embodied brain module 100 and parses them into high-frequency motor instructions for motion planning, trajectory optimization, and joint torque control of the robot 300. It then exchanges data with the robot 300's sensors and actuators to control the robot 300's actions to execute sub-tasks. Furthermore, by reading data from the robot 300's sensors (such as inertial measurement units, encoders, and force sensors) in real time, it can achieve functions such as dynamic balance, precise operation, real-time obstacle avoidance, and force-controlled interaction, ensuring fast, smooth, and accurate movements. It should be noted that the embodied cerebellum module 200 can control the robot 300's movement in any manner; this is not the focus of this application and will not be elaborated upon here.

[0028] It should also be noted that the hierarchical and decoupled architecture of the embodied brain module 100 and the embodied cerebellum module 200 allows the cognitive algorithms of the embodied brain module 100 and the control algorithms of the embodied cerebellum module 200 to be developed, tested, and upgraded independently. This allows for easy replacement of the embodied brain module 100 with a more powerful AI model, or the addition of new motion control algorithms to the embodied cerebellum module 200, without requiring modifications to the entire system, demonstrating good maintainability and technological development potential.

[0029] The systems provided in this application communicate via a dual-track communication architecture. The high-frequency data channel 400 employs a shared memory mechanism for data exchange between the embodied cerebellum module 200 and the sensors and actuators of the robot 300. This ensures that the embodied cerebellum module 200 can perform calculations based on the latest state of the robot 300 during real-time closed-loop control, achieving millisecond-level response. The low-frequency command channel 500 uses socket-based network communication for task-level information interaction between the embodied brain module 100 and the embodied cerebellum module 200. Asynchronous transmission via the low-frequency command channel 500 at a relatively low frequency ensures the reliability and flexibility of complex information transmission.

[0030] The hierarchical decoupled architecture of the embodied brain module 100 and the embodied cerebellum module 200 in this embodiment separates high-level "thinking" from low-level "action," and then tightly and efficiently couples them together through a dual-track communication architecture. A high-frequency control loop is realized through the shared memory of the high-frequency data channel 400, with a response time of milliseconds, ensuring the smoothness and accuracy of the robot's movements in dynamic environments. At the same time, task-level asynchronous communication is carried out through the low-frequency instruction channel 500, which perfectly solves the problem of coordination between high-level cognition and low-level control, and opens up the entire link from understanding complex instructions to executing precise actions, realizing low latency, high reliability and high autonomy of robot control, thereby solving the problems existing in the prior art.

[0031] In some possible implementations of the embodiments of this application, the cerebellum module 200 is specifically used to parse the received sub-task into a whole-body motion control command for the robot 300, read the status feedback of the sensor in real time through the high-frequency data channel 400, and drive the actuator to complete the corresponding specific action.

[0032] Understandably, the embodied cerebellum module 200 performs high-frequency real-time control. Through the high-frequency data channel 400, i.e., shared memory, it reads the robot 300's state feedback in real time, such as joint angles and inertial measurement unit (IMU) data, forming a high-speed local closed loop to drive the actuators to complete the specific actions corresponding to the subtask. The frequency at which the embodied cerebellum module 200 reads data from the shared memory can be 1000Hz; this is not limited here. The embodied cerebellum module 200 achieves a 1000Hz high-frequency control loop through shared memory, with a response time in milliseconds, ensuring the smoothness and accuracy of the robot 300's movements in dynamic environments. Simultaneously, it uses Socket for task-level asynchronous communication, perfectly solving the coordination problem between high-level cognition and low-level control.

[0033] In some possible implementations of the embodiments of this application, the embodied cerebellum module 200 can also be used to determine the execution status of a subtask based on the status feedback of the sensor, and feed back the execution status to the embodied brain module 100.

[0034] Understandably, the task execution status feedback from the embodied cerebellum module 200 can be "target point reached," "grabbing failed," "in progress," "completed," "failure and reason," etc., forming a low-frequency global cognitive closed loop. The embodied cerebellum module 200 transmits asynchronously at a relatively low frequency through the low-frequency command channel 500, ensuring the reliability and flexibility of complex information transmission.

[0035] In one example, the embodied brain module 100 can also be used to dynamically replan target task instructions based on the execution status.

[0036] Understandably, the embodied brain module 100 updates the progress of the entire task based on the received execution status feedback. If a subtask is executed successfully, the next subtask is assigned. If execution fails or a significant change in the environment is detected, the embodied brain module 100 will perform dynamic replanning or decision adjustment, generating a new sequence of subtasks to adapt to the change. This process repeats until the entire complex task is completed. The closed-loop feedback and dynamic replanning mechanism between the embodied brain module 100 and the embodied cerebellum module 200 enables the robot 300 to intelligently respond to abnormal conditions such as sensor noise, communication interference, and sudden obstacles, rather than simply interrupting the task or executing incorrect actions. This significantly improves the safety and reliability of the robot 300 in high-risk environments such as nuclear facilities, achieving a high degree of adaptability to complex and dynamic environments.

[0037] As an example, in practical implementation, the cloud can be an embedded AI platform or a computing server. The embodied brain module 100 can be deployed on the robot-side AI platform or a high-performance edge computing server, communicating with the robot 300 via a wired network or 5G network. The embodied brain module 100 runs a large VLA model with 7 billion parameters, fine-tuned using nuclear fuel field professional manuals, operating procedures (SOPs), and expert demonstration data for long-term task planning. The embodied cerebellum module 200 can be deployed as an embedded real-time controller within the thoracic cavity of the robot 300. This embedded real-time controller can employ an ARM multi-core processor supporting a real-time operating system (RTOS). The embodied cerebellum module 200 integrates a state-aware ontology-based high-dimensional motion planning algorithm, an error-compensated full-process precise control algorithm for the robotic arm, and an integrated high-robustness control and fault-tolerant mechanism algorithm for complex working conditions.

[0038] In one example, the low-frequency instruction channel 500 can be a TCP / IP connection established via an Ethernet network; the high-frequency data channel 400 can be implemented using the physical memory of the robot 300 controller.

[0039] In practical implementation, the embodied brain module 100 and the embodied cerebellum module 200 can establish a reliable TCP / IP connection via Ethernet and communicate using Socket. Commands and status data are encapsulated in JSON format. The high-frequency data channel 400 can allocate a shared memory area in the physical memory of the real-time controller. The robot 300's dual-arm joint encoders, end-effector six-dimensional force sensors, chassis IMU, and wheel speedometers directly access their data to this shared memory at a frequency of 1000Hz via DMA. The control algorithm of the embodied cerebellum module 200 directly reads the status from this memory area and writes the calculated target torque or speed commands for each joint back to this area, which are then read and executed by the underlying servo driver. The entire read-write-calculation-execution closed-loop cycle is stable at 1 millisecond. The robot 300 may include a mobile chassis with 3 degrees of freedom, two robotic arms with 6 degrees of freedom each, and multimodal sensors such as depth cameras, LiDAR, and infrared thermal imagers, which will not be listed here.

[0040] This application provides a robot control system, method, and robot based on a hierarchical collaborative architecture. The system includes an embodied brain module and an embodied cerebellum module. The embodied brain module is deployed in the cloud and is used for long-term task planning, decomposing target task instructions into sub-tasks and sending them to the embodied cerebellum module. The embodied cerebellum module is deployed on the robot body and is used to control the robot's actions to execute the sub-tasks according to the received sub-tasks. The system communicates through a dual-track communication architecture. The dual-track communication architecture includes a high-frequency data channel and a low-frequency command channel. The high-frequency data channel adopts a shared memory mechanism for data exchange between the embodied cerebellum module and the robot's sensors and actuators. The low-frequency command channel adopts socket-based network communication for task-level information interaction between the embodied brain module and the embodied cerebellum module. This application embodiment implements a high-frequency control loop through shared memory, achieving a response time in milliseconds, thus ensuring the smoothness and accuracy of the robot's movements in dynamic environments. At the same time, it uses a low-frequency command channel for task-level asynchronous communication, perfectly solving the problem of coordination between high-level cognition and low-level control, and opening up the entire link from understanding complex commands to executing precise actions, thereby achieving low latency, high reliability, and high autonomy in robot control.

[0041] Based on the robot control system based on the hierarchical collaborative architecture provided in the above embodiments, this application also provides a robot control method based on the hierarchical collaborative architecture, which can be applied to any one of the robot control systems based on the hierarchical collaborative architecture provided in the above embodiments.

[0042] See Figure 2The figure is a flowchart illustrating a robot control method based on a hierarchical collaborative architecture provided in an embodiment of this application.

[0043] This application provides a robot control method based on a hierarchical collaborative architecture, comprising: Step S101: The embodied brain module 100 receives the target task instruction and decomposes the target task instruction into sub-tasks; Step S102: The embodied brain module 100 sends a command to the embodied cerebellum module 200 via the low-frequency command channel 500; the low-frequency command channel 500 uses socket-based network communication for task-level information interaction between the embodied brain module 100 and the embodied cerebellum module 200. Step S103: The embodied cerebellum module 200 controls the robot 300 to perform the sub-task by using the high-frequency data channel 400 according to the received sub-task; the high-frequency data channel 400 adopts a shared memory mechanism for data exchange between the embodied cerebellum module 200 and the sensors and actuators of the robot 300.

[0044] Understandably, the embodied brain module 100, as the cognitive center of robot 300, is responsible for handling low-frequency but highly complex cognitive tasks. This includes receiving and understanding complex task instructions from humans, integrating multimodal sensor information to perform semantic understanding of the working environment, and conducting long-term task planning. It intelligently decomposes an abstract operational goal into a series of logically clear and executable sub-task instruction sequences. The embodied cerebellum module 200, as the robot's motion control center, is deployed on the robot 300 body and is responsible for handling high-frequency, low-latency real-time control tasks. This includes receiving sub-task instructions from the embodied brain module 100 and parsing them into high-frequency motor instructions such as whole-body motion planning, trajectory optimization, and joint torque control, thereby achieving dynamic balance, precise operation, and real-time obstacle avoidance for robot 300.

[0045] The high-frequency data channel 400 and the low-frequency command channel 500 are crucial bridges connecting the embodied brain module 100, the embodied cerebellum module 200, and the robot 300. They innovatively employ two different communication methods: the high-frequency data channel 400 uses a shared memory mechanism for data exchange between the sensors and actuators of the embodied cerebellum module 200 and the robot 300, reading and writing to shared memory at a frequency of 1000Hz or higher to ensure data freshness during real-time control loop operations in the embodied cerebellum module 200. The low-frequency command channel 500 uses socket-based network communication for task-level information interaction between the embodied brain module 100 and the embodied cerebellum module 200, transmitting asynchronously at a relatively low frequency (e.g., 10-100Hz). Through a layered architecture and dual-track communication, low-latency, high-synchronization collaboration between advanced cognition and real-time control is achieved. The architecture is clear, decoupled, and facilitates independent upgrades and maintenance of each module, exhibiting excellent scalability.

[0046] In one example, step S101 may specifically include: The embodied brain module is based on a visual-language-motor model for long-term task planning.

[0047] In practical applications, the embodied brain module 100 can be based on a VLA (Virtual Language Analysis) model finely tuned with expertise in specific application scenarios (such as the nuclear domain) to enhance its understanding of specific scenarios and its task planning capabilities. Based on the VLA model, the embodied brain module 100 endows the robot 300 with powerful autonomous decision-making and task generalization capabilities, enabling it to adapt to unstructured and complex environments.

[0048] The embodied cerebellum module 200 integrates a state-aware ontology-based high-dimensional motion planning algorithm, an error-compensated full-process precise control strategy for the robotic arm, and an integrated, highly robust control and fault-tolerant mechanism for complex working conditions, thereby improving the operational accuracy and safety of the robot 300 in complex environments. Other relevant details can be found in the above embodiments and will not be repeated here.

[0049] In some possible implementations of the embodiments of this application, step S103 may specifically include: The cerebellum module 200 parses the received subtasks into full-body motion control commands for the robot 300; The cerebellum module 200 reads the sensor status feedback in real time through the high-frequency data channel 400; The embodied cerebellum module 200 drives the actuator to complete the corresponding specific action based on the whole-body motion control command and status feedback through the high-frequency data channel 400.

[0050] In one example, step S103 may be followed by: Step S104: The embodied cerebellum module 200 determines the execution status of the subtask based on the status feedback from the sensors; Step S105: The embodied cerebellum module 200 feeds back the execution status to the embodied brain module 100.

[0051] In another example, step S105 may be followed by: Step S106: The embodied brain module 100 performs dynamic task replanning of the target task instructions based on the execution status.

[0052] It is understood that the relevant descriptions of the above steps can be found in the relevant content of the above system partial embodiments, and will not be repeated here.

[0053] The following example illustrates in detail a robot control method based on a hierarchical collaborative architecture provided in this application.

[0054] Taking the task of "performing infrared temperature measurement and dust removal on the control cabinet of a CL-15 CNC lathe" as an example: This application provides a robot control method based on a hierarchical collaborative architecture, which may include: Step S1: The operator gives the voice command "Perform infrared temperature measurement and dust removal on the control cabinet of the CL-15 CNC lathe".

[0055] Step S2: The VLA model of the embodied brain module 100 receives the instruction and calls the vision module to identify the CL-15 control cabinet in the scene. It decomposes the task into a sequence of sub-tasks, including 6 sub-tasks: 1. Navigate to a point 1 meter in front of the CL-15 control cabinet; 2. Identify and use the left arm to open the cabinet door; 3. Use the right arm, equipped with an infrared thermal imager, to scan all wiring terminals; 4. Identify any abnormal points with temperatures exceeding 40℃ and record their coordinates; 5. Replace the right arm with a compressed air nozzle to perform a zigzag cleaning of the cabinet interior; 6. Use the left arm to close the cabinet door; 7. Report task completion and abnormal point information via the voice synthesis module.

[0056] Step S3: The embodied brain module 100 sends the first subtask instruction to the embodied cerebellum module 200 through the low-frequency instruction channel 500: {"task_id": 1, "type": "navigate", "target_pose": [x, y, theta]}.

[0057] Step S4: After receiving the instruction, the cerebellum module 200 generates a collision-free path through its motion planning unit and transmits it to the shared memory (i.e., the high-frequency data channel 400). Subsequently, the robot 300's whole-body controller reads the IMU and wheel speed meter data from the shared memory at a frequency of 1000Hz, calculates and outputs real-time speed commands to drive the chassis hub motors, and drives the robot 300 to move smoothly.

[0058] Step S5: When the positioning system of robot 300 shows that it has reached the target position and the posture is stable, the embodied cerebellum module 200 sends feedback to the embodied brain module 100 through the low frequency command channel 500: {"task_id": 1, "status": "success"}.

[0059] Step S6: After receiving the successful feedback, the embodied brain module 100 continues to issue the second subtask {"task_id": 2, "type": "open_door", "arm": "left"}.

[0060] Step S7: The embodied cerebellum module 200 drives the left arm to perform the door-opening action. During this process, if the value fed back by the force sensor at the end of the left arm in the shared memory is continuously higher than the preset threshold, it indicates that the door is stuck. The fault tolerance mechanism of the embodied cerebellum module 200 will immediately stop the action and feed back the abnormal status {"task_id": 2, "status": "error", "code": "E102", "message": "Door stuck"} to the embodied brain module 100 through the low-frequency command channel 500.

[0061] Step S8: After receiving the anomaly, the embodied brain module 100 can decide to execute a new recovery strategy, such as issuing a sub-task of "slightly shaking the cabinet door from side to side," or directly requesting manual intervention from the operator. The task process proceeds dynamically based on feedback until all sub-tasks are completed.

[0062] This application embodiment implements a high-frequency control loop through shared memory, achieving a response time in milliseconds, thus ensuring the smoothness and accuracy of the robot's movements in dynamic environments. At the same time, it uses a low-frequency command channel for task-level asynchronous communication, perfectly solving the problem of coordination between high-level cognition and low-level control, and opening up the entire link from understanding complex commands to executing precise actions, thereby achieving low latency, high reliability, and high autonomy in robot control.

[0063] Based on the robot control system and method based on the hierarchical collaborative architecture provided in the above embodiments, this application also provides a robot, including: a controller, a sensor and an actuator; the controller is equipped with an embodied cerebellum module from any of the robot control systems based on the hierarchical collaborative architecture provided in the above embodiments. The embodied cerebellum module interacts with the embodied brain module at the task level through a low-frequency command channel and exchanges data with sensors and actuators through a high-frequency data channel. The low-frequency instruction channel uses socket-based network communication, while the high-frequency data channel uses a shared memory mechanism.

[0064] In one example, a high-frequency data channel can be implemented using the controller's physical memory.

[0065] Based on the robot control system and method based on the hierarchical collaborative architecture provided in the above embodiments, this application also provides a computer having a embodied brain module of any of the robot control systems based on the hierarchical collaborative architecture provided in the above embodiments.

[0066] The present application has been described in detail above with reference to the accompanying drawings and embodiments. However, the present application is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present application. All content not described in detail in this application can be derived from existing technology.

Claims

1. A robot control system based on a hierarchical collaborative architecture, characterized in that, The system includes: an embodied brain module and an embodied cerebellum module; The embodied brain module is deployed in the cloud and is used for long-term task planning, decomposing the target task instructions into sub-tasks and sending them to the embodied cerebellum module. The embodied cerebellum module is deployed on the robot body and is used to control the robot's actions to execute the sub-task according to the received sub-task; The system communicates via a dual-track communication architecture, which includes a high-frequency data channel and a low-frequency command channel. The high-frequency data channel adopts a shared memory mechanism for data exchange between the embodied cerebellum module and the robot's sensors and actuators; The low-frequency command channel uses socket-based network communication for task-level information interaction between the embodied brain module and the embodied cerebellum module.

2. The robot control system based on a hierarchical collaborative architecture according to claim 1, characterized in that, The embodied brain module performs long-term task planning based on a visual-language-action model.

3. The robot control system based on a hierarchical collaborative architecture according to claim 1, characterized in that, The embodied cerebellum module is specifically used to parse the received sub-tasks into the robot's whole-body motion control commands, read the sensor status feedback in real time through the high-frequency data channel, and drive the actuator to complete the corresponding specific actions.

4. The robot control system based on a hierarchical collaborative architecture according to claim 3, characterized in that, The embodied cerebellum module is also used to determine the execution status of the sub-task based on the status feedback of the sensor, and to feed back the execution status to the embodied brain module.

5. The robot control system based on a hierarchical collaborative architecture according to claim 4, characterized in that, The embodied brain module is also used to dynamically replan the target task instructions based on the execution state.

6. The robot control system based on a hierarchical collaborative architecture according to any one of claims 1-5, characterized in that, The cloud refers to an embedded AI platform or a computing server.

7. The robot control system based on a hierarchical collaborative architecture according to any one of claims 1-5, characterized in that, The low-frequency command channel is specifically a TCP / IP connection established via an Ethernet network; the high-frequency data channel is implemented using the physical memory of the robot's controller.

8. A robot control method based on a hierarchical collaborative architecture, characterized in that, The method is applied to the robot control system based on a hierarchical collaborative architecture as described in any one of claims 1-7; the method includes: Step S101: The embodied brain module receives the target task instruction and decomposes the target task instruction into sub-tasks; Step S102: The embodied brain module sends a command to the embodied cerebellum module via a low-frequency command channel; the low-frequency command channel uses socket-based network communication for task-level information interaction between the embodied brain module and the embodied cerebellum module. Step S103: The embodied cerebellum module controls the robot to perform the sub-task by using a high-frequency data channel according to the received sub-task; the high-frequency data channel adopts a shared memory mechanism for data exchange between the embodied cerebellum module and the robot's sensors and actuators.

9. The robot control method based on a hierarchical collaborative architecture according to claim 8, characterized in that, Step S101 specifically includes: The embodied brain module performs long-term task planning based on a visual-language-action model.

10. The robot control method based on a hierarchical collaborative architecture according to claim 8, characterized in that, Step S103 specifically includes: The embodied cerebellum module parses the received sub-tasks into full-body motion control commands for the robot; The embodied cerebellum module reads the status feedback of the sensor in real time through the high-frequency data channel; The embodied cerebellum module drives the actuator to complete the corresponding specific action through the high-frequency data channel based on the whole-body motion control command and the status feedback.

11. The robot control method based on a hierarchical collaborative architecture according to claim 8, characterized in that, Step S103 is followed by: Step S104: The embodied cerebellum module determines the execution status of the subtask based on the status feedback from the sensor; Step S105: The embodied cerebellum module feeds back the execution status to the embodied brain module.

12. The robot control method based on a hierarchical collaborative architecture according to claim 11, characterized in that, Step S105 is followed by: Step S106: The embodied brain module performs dynamic task replanning on the target task instruction based on the execution state.

13. A robot, characterized in that, include: A controller, sensors, and actuators; the controller is equipped with an embodied cerebellum module, as described in any one of claims 1-7, for a robot control system based on a hierarchical collaborative architecture. The embodied cerebellum module interacts with the embodied brain module at the task level through a low-frequency command channel and exchanges data with the sensor and the actuator through a high-frequency data channel. The low-frequency instruction channel uses socket-based network communication, while the high-frequency data channel uses a shared memory mechanism.

14. The robot according to claim 13, characterized in that, The high-frequency data channel is implemented using the physical memory of the controller.

15. A computer, characterized in that, The computer is equipped with the embodied brain module of the robot control system based on the hierarchical collaborative architecture as described in any one of claims 1-7.

Images