Multi-arm Mobile Robot Cooperative Control Framework and Multi-arm Mobile Robot System

Abstract

The present application provides a cooperative control framework for a multi-arm mobile robot and a multi-arm mobile robot system, which is applied to the technical field of mobile composite multi-arm robots. Among them, the end error estimation module of the robotic arm obtains the pose error at the end of each robotic arm; the multi-arm cooperation cost module obtains the state differences between the robotic arms; the reinforcement learning module: takes the difference between the predicted value and the expected value of the pose at the end of each robotic arm, and the state differences between the robotic arms as inputs, and dynamically and adaptively adjusts the cooperation cost weights of the multi-arm cooperation cost module; so that the end error estimation module of the robotic arm and the multi-arm cooperation cost module iteratively pursue minimization, and control multiple robotic arms of the multi-arm mobile robot to perform cooperative operations. Based on this cooperative control framework, the present application can avoid action interference in the cooperative operation control of multi-arm robots, improve the adaptability, motion accuracy and stability of multi-arm robots, and achieve efficient cooperation between multi-arms.

Description

Technical Field

[0001] This application relates to the technical field of mobile composite multi-arm robots, and specifically relates to a cooperative control framework for multi-arm mobile robots and a multi-arm robot system. Background Art

[0002] In many fields such as modern industrial production, logistics handling, and emergency rescue, higher requirements are put forward for the operation ability and flexibility of robots. Mobile composite multi-arm robots, such as four-arm mobile robots with four arms mounted on a mobile chassis, have attracted much attention because they can not only move but also achieve cooperative operation of multiple manipulators.

[0003] However, there are many problems in the cooperative operation control of current four-arm mobile robots. For example, traditional control strategies are difficult to achieve real-time cooperation between multiple arms, resulting in interference between the actions of each manipulator when performing tasks and unable to complete tasks efficiently.

[0004] Based on this, a new technical solution is needed. Summary of the Invention

[0005] In view of this, this application provides a cooperative control framework for multi-arm mobile robots and a multi-arm robot system.

[0006] This application provides the following technical solutions:

[0007] According to a cooperative control framework for multi-arm mobile robots provided by this application, it includes the following modules:

[0008] End-effector error estimation module: Calculate the difference between the pose prediction value and the expected value of the end of each manipulator to obtain the end-effector pose error of each manipulator;

[0009] Multi-arm cooperation cost module: Obtain the state differences between each manipulator, and couple the state differences between each manipulator through the cooperation cost weight;

[0010] Reinforcement learning module: Take the difference between the pose prediction value and the expected value of the end of each manipulator and the state differences between each manipulator as inputs, and dynamically and adaptively adjust the cooperation cost weight of the multi-arm cooperation cost module;

[0011] So that the end-effector error estimation module and the multi-arm cooperation cost module iteratively pursue minimization, and control multiple manipulators of the multi-arm mobile robot to perform cooperative operations.

[0012] According to a multi-arm mobile robot system provided by this application, it includes:

[0013] Task data parsing and allocation module: Parse the motion and manipulator operation tasks of the multi-arm mobile robot to obtain parsed data;

[0014] Cooperative operation behavior planning and chassis motion coupling compensation unit: Apply the cooperative control framework of multi-arm mobile robots, predict the chassis motion behavior, couple the multi-arm kinematics, and constrain the multi-arm workspace according to the parsed data to generate control data;

[0015] Mobile robot control solution unit: Solve the generated control data to generate the chassis motion planning instructions and the control instructions for each manipulator, and send them to the underlying motion controller after dynamic compensation respectively, so as to control the actions of each actuator.

[0016] Preferably, the cooperative operation behavior planning and chassis motion coupling compensation unit includes:

[0017] Chassis motion behavior prediction unit: Based on the dynamic base model of the robot, predict the motion state parameters at the next moment according to the planned motion path of the mobile robot;

[0018] Multi-arm kinematics coupling unit: Based on the manipulator kinematics model, couple the dynamic base coordinate system of the chassis for the manipulator motion planning, eliminate the positioning error of the manipulator end caused by the chassis motion, and realize the real-time conversion of the end target pose of the manipulator based on the global coordinate system to the chassis dynamic coordinate system;

[0019] Workspace constraint unit: Optimize the workspace and trajectory of the multi-arm according to the manipulator perception fusion data and cooperative strategy.

[0020] Compared with the prior art, the beneficial effects that can be achieved by at least one of the above technical solutions adopted in this application at least include:

[0021] In this application, the difference between the predicted value and the expected value of the pose of each manipulator end is calculated through the manipulator end error estimation module to obtain the pose error of each manipulator end; the state differences between the manipulators are obtained through the multi-arm cooperation cost module and coupled through the cooperation cost weight; the reinforcement learning module takes the difference between the predicted value and the expected value of the pose of each manipulator end and the state differences between the manipulators as inputs, and dynamically adjusts the cooperation cost weight adaptively; so that the manipulator end error estimation module and the multi-arm cooperation cost module iteratively pursue minimization, control the multiple manipulators of the multi-arm mobile robot to perform constrained cooperative operations, form a multi-arm multi-modal cooperative control strategy for the multi-arm robot. Under the constraint of this cooperative framework, the multi-arm robot can achieve mutual cooperative operations, avoid action interference in the cooperative operation control of the multi-arm robot, improve the adaptability, motion accuracy and stability of the multi-arm robot, realize efficient cooperation between the multi-arms, and improve the operation ability of the robot in complex environments. Description of the Drawings

[0022] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0023] Figure 1 It is a schematic flowchart of the cooperative control of the multi-arm mobile robot in the present application;

[0024] Figure 2 It is a schematic flowchart of the end-effector error estimation of the robotic arm in the present application;

[0025] Figure 3 It is a schematic flowchart corresponding to the input data of the multi-arm cooperation cost function in the present application;

[0026] Figure 4 It is a schematic flowchart of the reinforcement learning algorithm in the present application;

[0027] Figure 5 It is a schematic diagram of the multi-arm mobile robot system in the present application. Specific embodiments

[0028] The following will describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0029] The following illustrates the implementation manners of the present application through specific specific examples. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. The present application can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0030] It should be noted that the following description relates to various aspects of embodiments within the scope of the appended claims. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is illustrative only. Based on this application, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspects described herein can be used to implement a device and / or practice a method. Additionally, this device and / or method can be implemented using other structures and / or functionality in addition to one or more of the aspects described herein.

[0031] It should also be noted that the diagrams provided in the following embodiments only schematically illustrate the basic concept of this application. The diagrams only show the components related to this application and are not drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in its actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0032] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that these examples can be practiced without these specific details.

[0033] Through in-depth research and improvement exploration of the control of a multi-arm composite mobile robot, the applicant found that: in the face of complex environments and task changes, the adaptability of existing control strategies is poor, and it is unable to quickly adjust the actions of each arm to meet new task requirements. Moreover, existing control methods have deficiencies in ensuring the motion accuracy and stability of multiple arms, which affects the operation quality of the robot.

[0034] Based on this, the following combines the accompanying drawings to illustrate the technical solutions provided by each embodiment of this application.

[0035] An embodiment of this specification proposes a cooperative control framework for a multi-arm mobile robot, as Figure 1 shown, including the following modules:

[0036] End-effector error estimation module for robotic arms: Calculate the difference between the predicted pose value and the expected value of the end-effector of each robotic arm to obtain the end-effector pose error of each robotic arm. Among them, when calculating the end-effector pose error of each robotic arm, the difference between the predicted pose value and the expected value of the end-effector of each robotic arm is used, rather than the difference between the true value and the expected value. This method is conducive to realizing the rapid iteration of this item, thereby accelerating the convergence speed and improving the performance of the system.

[0037] Multi-arm cooperation cost module: Obtain the state differences between each robotic arm and couple the state differences between each robotic arm through the cooperation cost weight.

[0038] Reinforcement learning module: Using the difference between the predicted pose value and the expected value at the end of each robotic arm, and the state differences between the robotic arms as inputs, dynamically and adaptively adjust the collaborative cost weights of the multi-arm collaborative cost module, accelerating or strengthening the iteration of the multi-arm collaborative cost module towards the minimum direction.

[0039] So that the robotic arm end error estimation module and the multi-arm collaborative cost module iteratively pursue minimization, controlling multiple robotic arms of the multi-arm mobile robot to perform collaborative operations, in order to improve the collaborative operation accuracy and multi-arm collaborative performance.

[0040] Among them, the objective function is based on the criterion of minimizing the cumulative sum of the end pose error matrix and the multi-arm collaborative cost function matrix. The robotic arm end error estimation matrix calculates the difference between the predicted pose value and the expected value at the end of each robotic arm to obtain the end pose error of each robotic arm; the state differences between each robotic arm and its adjacent robotic arms are obtained through the multi-arm collaborative cost function matrix, and the state differences between all robotic arms are coupled through the collaborative cost weights.

[0041] In one embodiment, as Figure 1 shown, the collaborative control framework further includes a multi-modal data fusion calculation module: performing modal fusion calculation based on the multi-modal data of each robotic arm to obtain the fused modal data. The multi-modal data fusion calculation module sends the fused modal data to the controllers of each robotic arm and the environmental coupling repulsive force field model. Among them, for example, data alignment, time synchronization; anomaly detection / fix; dynamic weight calculation, etc. are performed.

[0042] In the overall system, a multi-modal data hierarchical and dynamic weighted fusion strategy is adopted to improve the operation robustness of each robotic arm. The multi-modal fusion calculation module realizes the modal fusion calculation based on the multi-sensory data of each robotic arm, such as IMU (Inertial Measurement Unit data), vision, force data, etc.

[0043] The result of the multi-modal fusion calculation is sent to the controllers of each robotic arm on the one hand to realize the data fusion of the modal data and the robotic arm solution data; on the other hand, it is sent to the environmental repulsive force field model unit of each robotic arm as its input data.

[0044] In one embodiment, as Figure 1 and Figure 2 shown, the controllers of each robotic arm perform robotic arm kinematic solution on the robotic arm joint encoder data to obtain the solution data, perform fusion calculation on the solution data and the fused modal data to obtain the fusion calculation data, and perform extended Kalman prediction on the fusion calculation data to predict and output the predicted pose value at the end of each robotic arm.

[0045] In the process of obtaining the predicted value of the pose at the end of the robotic arm, a multi-modal data hierarchical and dynamic weighted fusion strategy is adopted to achieve the fusion processing of the modal data of each robotic arm (such as robotic arm joint encoders, visual positioning information, IMU data, and force sensor information), and the kinematic coupling term between robotic arms is introduced as the predicted value of the extended Kalman filter.

[0046] In one embodiment, as Figure 1 and Figure 2 shown, the controller of each robotic arm estimates the predicted value of the pose at the end of each robotic arm by using an extended Kalman filter based on the geometric constraint conditions between the robotic arms;

[0047] Among them, the geometric constraint conditions satisfied by the relative poses of the ends of any two robotic arms in each robotic arm are:

[0048] ;

[0049] Among them, h c represents the geometric constraint condition; i represents one robotic arm, j represents another robotic arm; T i represents the pose transformation matrix corresponding to one robotic arm; T j represents the pose transformation matrix corresponding to another robotic arm; represents the theoretical relative pose. i and j respectively represent different robotic arms, T i and T j are pose transformation matrices.

[0050] Therefore, each robotic arm estimates the end pose value of each robotic arm by using an extended Kalman filter based on the fusion result of its respective multi-modal data and under the geometric constraint conditions between them.

[0051] In one embodiment, as Figure 1 and Figure 3 shown, the environmental coupling repulsive force field model of each robotic arm combines the fused modal data and environmental state perception data to form environmental coupling repulsive force field model data, which is used as the input of the multi-arm collaborative cost module to enable real-time calculation and minimization iteration of the multi-arm collaborative cost module.

[0052] Specifically, each robotic arm performs fusion calculation based on its respective multi-modal perception data, and at the same time uses its respective environmental state perception data as input to separately form its own environmental coupling repulsive force field model data, which is used as the input of the cost function.

[0053] By performing real-time task weight adjustment, non-linear correction calculation, and environmental coupling repulsive force field calculation on each robotic arm, the calculation of the collaborative cost function is made to meet the requirements. Based on the calculation of the established multi-arm collaborative cost function matrix, the real-time calculation and minimization iteration of the multi-arm collaborative cost function are realized, so as to maximize the collaboration between the robotic arms and improve the collaborative performance and accuracy.

[0054] When performing task weight adjustment, according to the dynamic changes in task requirements, the weights of each cost item (such as position error, attitude error, etc.) are adjusted in real time to balance the collaboration between different arms. For example, in actual operation, the position error of some robotic arms is more important than the pose error, while the pose error is more important for some other robotic arms during operation. Therefore, according to the different requirements of the system tasks, the weights of the tasks need to be dynamically adjusted so that the iterative calculation results of the cost function are more in line with the expected requirements.

[0055] Due to factors such as robotic arm dynamics modeling and environmental constraints, non-linear errors will be generated, and certain corrections need to be made to the non-linear errors in order to improve the accuracy of the model and the collaboration of control. Therefore, for different robotic arms, certain non-linear corrections need to be made according to factors such as their dynamics models and motion constraints, so that the pose estimation or calculation in the cost function is more accurate.

[0056] The essence of the environmental coupling repulsive force field model is to avoid collisions between robotic arms during collaborative actions. The input of the environmental coupling repulsive force field model calculation is the data of multi-modal fusion and the data of environmental perception as the input constraint data of the model. The output of the environmental coupling repulsive force field model will, together with the task weight and non-linear correction calculation, affect the iterative calculation direction and results of the cost function.

[0057] In one embodiment, as Figure 1 and Figure 4 shown, the reinforcement learning module calculates synchronously with the robotic arm end error estimation module and the multi-arm collaborative cost module. Using the state difference data set as the input state space, reinforcement training and learning are carried out based on the state difference data set, and the weight values in the multi-arm collaborative cost module are dynamically corrected, and the dynamic weight coefficient matrix in the multi-arm collaborative cost function matrix is dynamically and adaptively adjusted.

[0058] Among them, for the reinforcement learning algorithm, the input state space is the state difference between the robotic arms in the collaborative cost matrix; reinforcement training and learning are carried out based on this state difference data set, and the rapid realization of its minimum value is used as its target task. The output of the reinforcement learning algorithm is the dynamic weight coefficient matrix in the collaborative cost function, which dynamically adjusts the weights in the collaborative cost function. Multiple combinations can be used for the reward function in the reinforcement learning algorithm; for example Figure 4In an embodiment, the end - effector execution parameter function of the robotic arm is used as the reward function, or it can be a combination of multiple parameters, such as comprehensively considering the end - effector error, force - sense uniformity, energy consumption, etc. as the reward function.

[0059] In one embodiment, as Figure 1 , Figure 2 and Figure 3 shown, the robotic - arm end - effector error estimation module includes a robotic - arm end - effector pose error estimation matrix:

[0060] ;

[0061] wherein, represents the desired pose of the robotic - arm end - effector; represents the predicted result value of the actual pose of the robotic - arm end - effector; represents the pose - error weight matrix; T represents the transpose of the matrix.

[0062] The multi - arm cooperation cost module includes a multi - arm cooperation cost function matrix, that is, the difference between the states of each robotic arm and the states of adjacent robotic arms:

[0063] ;

[0064] wherein, u i represents the end - effector state matrix of one robotic arm; u j represents the end - effector state matrix of another robotic arm; R u represents the cooperation - cost function dynamic weight matrix. After the state differences between all arms are coupled by the cooperation - cost dynamic weight, a multi - arm cooperation cost function is formed; the multi - arm cooperation cost function matrix has the same minimum - strategy pursuit as the aforementioned robotic - arm end - effector error matrix, and multi - arm cooperation work is realized on the premise of pursuing the minimum value.

[0065] In one embodiment, as Figure 1 shown, the objective function includes three parts, namely: the robotic - arm end - effector error matrix, the multi - arm cooperation cost function matrix, and the reinforcement - learning algorithm. The expression of the objective function of this cooperative - control framework:

[0066]

[0067] wherein, represents the reinforcement - learning algorithm term, which is the objective function of the cumulative reward; represents the N th power of the discount factor, which is used to adjust the weight of future rewards, that is, to adjust the importance of future rewards. A larger value means a greater weight of future rewards; N represents the total number of steps or total time;t Represents time or the number of steps; Is the reward function, representing the immediate reward based on the instruction value and the feedback prediction value; Represents the immediate feedback prediction value, Represents the immediate instruction value; E Is the expectation operator, representing taking the average. In the end - effector pose error term, the predicted value of the actual end - effector pose is adopted instead of the traditional actual end - effector value to improve the convergence speed of the pose error term and reduce the actual pose error, thereby improving the cooperation accuracy.

[0068] The reinforcement learning algorithm term also belongs to one of the objective functions, but it is embedded in the objective function during the calculation to perform dynamic adaptive adjustment on the cost coefficient R u in the multi - arm cooperation cost function matrix, thereby enhancing the multi - arm cooperation performance and accelerating the convergence speed of the objective function.

[0069] This application provides a coordinated operation control strategy and control architecture for a mobile composite multi - arm robot, which is applicable to a composite multi - arm cooperative operation robot, forming a multi - arm multi - modal cooperative control strategy for the multi - arm robot. Under the constraint of this cooperation framework, the multi - arm robot can achieve mutual cooperative operation. It solves the problems of action interference, poor adaptability, and insufficient motion accuracy and stability existing in the existing multi - arm robot during cooperative operation control, realizes efficient cooperation between multiple arms, and improves the operation ability of the robot in a complex environment.

[0070] This application constrains the multi - arm mobile robot through the objective function of the cooperation control framework, and controls multiple manipulators of the multi - arm mobile robot to perform cooperative operations. Among them, the reinforcement learning algorithm is also included in the objective function and is calculated synchronously with the other two items. The reinforcement learning calculation uses the state difference data set generated by the cost function matrix as the training sample data to dynamically correct the weight value in the multi - arm cooperation cost function, so that the multi - arm cooperation cost function converges faster and obtains a lower value, thereby having higher cooperation accuracy.

[0071] This embodiment of the specification also discloses a multi - arm mobile robot system, as Figure 5 shown, including:

[0072] Task data parsing and distribution module: Parses the motion and manipulator operation tasks of the multi - arm mobile robot to obtain parsed data. For example, parses the motion and manipulator operation tasks of a four - arm mobile robot according to the content such as a robot order, and distributes the data to the cooperative operation behavior planning and chassis motion coupling compensation unit.

[0073] The cooperative operation behavior planning and chassis motion coupling compensation unit performs chassis motion behavior prediction, multi - arm kinematic coupling, and multi - arm workspace constraint according to the parsed data to generate control data.

[0074] Mobile robot control and calculation unit: It calculates the generated control data to generate robot chassis motion planning instructions and each robotic arm control instruction, and sends them to the underlying motion controller after dynamic compensation respectively, so as to control the actions of each actuator.

[0075] Specifically, it analyzes and calculates the data generated by the above-mentioned collaborative operation and chassis coupling compensation unit to become the robot chassis motion planning instruction and each robotic arm control instruction respectively, and sends them to the underlying motion controller after dynamic compensation respectively to execute the motion instructions of each motor. For example, it calculates the motion state parameters at the next moment predicted, the end target pose of the robotic arm in the chassis dynamic coordinate system in real time, and the optimized multi-arm workspace and trajectory and other data.

[0076] In one embodiment, the collaborative operation behavior planning and chassis motion coupling compensation unit includes:

[0077] Chassis motion behavior prediction unit: According to the motion path planned by the mobile robot, based on the dynamic base model of the robot, it predicts the motion state parameters at the next moment.

[0078] Multi-arm kinematic coupling unit: Based on the robotic arm kinematic model, it couples the chassis dynamic base coordinate system for the robotic arm motion planning, eliminates the positioning error of the robotic arm end caused by the chassis motion, and realizes the real-time conversion of the end target pose of the robotic arm based on the global coordinate system into the chassis dynamic coordinate system.

[0079] Workspace constraint unit: According to the robotic arm perception fusion data and collaborative strategy, it optimizes the multi-arm workspace and trajectory to avoid interference between robotic arms and between the robotic arm and the chassis motion; the constraint unit may include such as robotic arm singular pose, collaborative strategy constraint conditions, etc.

[0080] For a four-arm mobile robot, when the four arms perform collaborative operations, it is also affected by the movement behavior of the mobile chassis. Therefore, in system control, the influencing factors of the chassis motion on the four-arm collaborative operation should be considered. Based on this idea, this application proposes a four-arm mobile robot system framework to compensate for the chassis motion in the four-arm collaborative control.

[0081] In this specification, the same or similar parts between each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the embodiments described later, the description is relatively simple, and the relevant parts can refer to the partial description of the foregoing embodiments.

[0082] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A collaborative control framework for a multi-arm mobile robot, characterized in that: Includes the following modules: Manipulator end error estimation module: calculates the difference between the predicted value and the expected value of the pose of each manipulator end to obtain the pose error of each manipulator end; Multi-arm collaborative cost module: obtains the state difference between each robotic arm and couples the state difference between each robotic arm through collaborative cost weight; Reinforcement learning module: using the difference between the predicted value and the expected value of the pose of each robotic arm and the state difference between each robotic arm as input, the collaborative cost weight of the multi-arm collaborative cost module is dynamically and adaptively adjusted; The robot end error estimation module and the multi-arm collaborative cost module are iteratively pursued to be minimized, and multiple robot arms of the multi-arm mobile robot are controlled to work collaboratively.

2. The multi-arm mobile robot collaborative control framework according to claim 1, characterized in that: The collaborative control framework also includes: a multimodal data fusion calculation module: performing modal fusion calculation based on the multimodal data of each robotic arm to obtain fused modal data; The multimodal data fusion calculation module sends the fused modal data to the controller of each robotic arm and the environmental coupling repulsive field model.

3. The multi-arm mobile robot collaborative control framework according to claim 2, characterized in that: The controller of each robot arm performs robot arm kinematics solution on the robot arm joint encoder data to obtain solution data, fuses the solution data with the fused modal data to obtain fused calculation data, performs extended Kalman prediction on the fused calculation data, and predicts and outputs the posture prediction value of each robot arm end.

4. The multi-arm mobile robot collaborative control framework according to claim 3, characterized in that: The controller of each robotic arm estimates the predicted value of the position and posture of each robotic arm end using an extended Kalman filter based on the geometric constraints between the robotic arms. Among them, the geometric constraints satisfied by the relative posture of the ends of each pair of manipulators are: ； in, h c Represents geometric constraints; i represents a robotic arm, j represents another robotic arm; T i Represents the posture transformation matrix corresponding to a robotic arm; T j Represents the posture transformation matrix corresponding to the other robotic arm; represents the theoretical relative pose.

5. The multi-arm mobile robot collaborative control framework according to claim 2, characterized in that: The environmental coupling repulsive field model of each robotic arm will fuse the modal data and environmental state perception data to form environmental coupling repulsive field model data as the input of the multi-arm collaborative cost module to enable real-time calculation and minimize iteration of the multi-arm collaborative cost module.

6. The multi-arm mobile robot collaborative control framework according to claim 1, characterized in that: The reinforcement learning module calculates synchronously with the robot end error estimation module and the multi-arm collaborative cost module, uses the state difference data set generated by the state difference of each robot arm as the input state space, performs reinforcement training and learning based on the state difference data set, and dynamically corrects the weight value in the multi-arm collaborative cost module.

7. The multi-arm mobile robot collaborative control framework according to claim 1, characterized in that: The manipulator end error estimation module includes a manipulator end posture error estimation matrix: ； in, Indicates the desired position of the end effector of the robot arm; Represents the predicted result value of the actual position and posture of the end effector of the robot arm; represents the pose error weight matrix; T Represents the transpose of a matrix; The multi-arm coordination cost module includes a multi-arm coordination cost function matrix: ； in, u i Represents the end state matrix of a robotic arm; u j represents the end state matrix of the other robot; R u Represents the dynamic weight matrix of the collaborative cost function.

8. The multi-arm mobile robot collaborative control framework according to claim 7, characterized in that: The objective function expression of the collaborative control framework is: ； in, Represents the reinforcement learning algorithm term; The discount factor N The power is used to adjust the weight of future rewards; N Indicates the total number of steps; is the reward function, which represents the immediate reward based on the instruction value and the feedback prediction value; represents the immediate feedback prediction value, Indicates the immediate instruction value; E is the expectation operator, which means taking the average value.

9. A multi-arm mobile robot system, characterized in that: include: Task data analysis and allocation module: analyzes the motion of the multi-arm mobile robot and the robot arm operation tasks to obtain analysis data; Collaborative operation behavior planning and chassis motion coupling compensation unit: applying the collaborative control framework of the multi-arm mobile robot described in any one of claims 1 to 8, performing chassis motion behavior prediction, multi-arm kinematic coupling and multi-arm workspace constraints according to analytical data to generate control data; Mobile robot control solving unit: solves the generated control data, generates robot chassis motion planning instructions and each robotic arm control instructions, and sends them to the underlying motion controller after dynamic compensation, thereby controlling the actions of each actuator.

10. The multi-arm mobile robot system according to claim 9, characterized in that: The collaborative operation behavior planning and chassis motion coupling compensation unit includes: Chassis motion behavior prediction unit: According to the motion path planned by the mobile robot and based on the dynamic base model of the robot, the motion state parameters at the next moment are predicted; Multi-arm kinematic coupling unit: Based on the kinematic model of the robot arm, the robot arm motion planning is coupled to the chassis dynamic base coordinate system, eliminating the robot arm end positioning error caused by chassis movement, and realizing the real-time conversion of the end target posture of the robot arm based on the global coordinate system into the chassis dynamic coordinate system; Workspace constraint unit: optimizes the workspace and trajectory of multiple arms based on the robot arm perception fusion data and collaborative strategy.

Images