Robot, double-arm tight coupling cooperation method and device thereof and computer application program
Through a hierarchical framework of two-stage data generation and imitation learning combined with master-slave control, the problem of insufficient data in robotic arms tight coupling cooperation is solved, and collision-free path planning and closed-chain constraint stability are achieved in complex environments, providing an efficient collaboration solution.
Patent Information
- Application Number
- CN202510548809.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-28
- Publication Date
- 2025-08-01
AI Technical Summary
In the prior art, imitation learning strategies are difficult to effectively apply to robots' double-arm tight coupling collaboration tasks due to the lack of sufficient expert data, especially in complex environments, and it is difficult to achieve collision-free path planning and closed-chain constraint stability.
Through a two-stage data generation method, the basic task data is first collected in a simple environment, and then the data set is enhanced using geometric approximation and hallucination mechanisms to generate high-quality training data sets, and real-time collision-free path planning is performed in combination with a hierarchical framework of imitation learning and master-slave control.
Real-time collision-free task execution of robotic arms tightly coupled and cooperative in complex static and dynamic environments is achieved, solving the problem of poor parameter sensitivity and stability in existing methods, and providing a brand new solution.
Smart Images
Figure CN120395824A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure belongs to the technical field of robots, and particularly relates to a robot and its method, device, and computer application program for tightly coupled collaborative operation of two arms. Background Art
[0002] Service robots with a two-arm system have great potential in future service robot applications because of their ability to complete complex operation tasks. Service robots with a two-arm system need to execute tightly coupled collaborative tasks of the two arms. In the tightly coupled collaborative tasks of the two arms, a strict closed-chain constraint is formed by the ends of the two robotic arms and an object, requiring the robotic arms and the object to maintain a highly coordinated synchronous movement. At the same time, a collision-free path planning for the robot must be performed in real time on the basis of ensuring the stability of the closed-chain constraint.
[0003] A relatively mature learning strategy in the field of robots is imitation learning. Imitation learning refers to learning a strategy by observing the demonstration behavior of an expert (such as a state-action trajectory) to make the behavior of the intelligent agent close to that of the expert. For example, a robot completes a grasping task by imitating human operations.
[0004] However, imitation learning requires a large amount of expert data. In the tightly coupled collaborative tasks of the two arms, at least the robotic arm state data, obstacle information data, and target information data at the same moment need to be synchronously and accurately collected. There are many types of data to be acquired, and there are large modal differences. Therefore, it is difficult to obtain sufficient expert data for training the imitation learning model for the tightly coupled collaborative tasks of the two arms. Therefore, imitation learning has made little progress in the tightly coupled collaborative tasks of the two arms. Summary of the Invention
[0005] An embodiment of the present disclosure proposes a tightly coupled collaborative scheme for a robot's two arms based on imitation learning to solve the problem that the imitation learning strategy cannot be successfully applied to the field of tightly coupled collaborative operation of a robot's two arms due to insufficient expert data.
[0006] The first aspect of the embodiment of the present disclosure provides a method for tightly coupled collaborative operation of a robot's two arms, including:
[0007] Obtain task data for the tightly coupled collaborative task of the robot's two arms, and input the task data into an imitation learning model to generate a motion path of the master robotic arm, where the imitation learning model is trained and generated by a supervised learning model using a training data set obtained by randomly adding obstacles in a preset environment according to preset rules, and the task data includes robotic arm state data, obstacle information data, and target information data;
[0008] Control the master robotic arm of the robot to move along the motion path;
[0009] Control the slave robotic arm of the robot to perform synchronous movement following the master robotic arm.
[0010] In some embodiments of the present disclosure, the motion path refers to:
[0011] A path composed of a series of path points in the Cartesian space, satisfying that when the main robotic arm moves along the path, it will not collide with obstacles.
[0012] In some embodiments of the present disclosure, the slave robotic arm of the control robot follows the main robotic arm for synchronous movement, including:
[0013] Performing error compensation on the slave robotic arm in real time according to the movement of the main robotic arm to ensure the stability of the closed-loop constraint between the master and slave robotic arms.
[0014] In some embodiments of the present disclosure, the training data set generated by training in a preset environment with obstacles randomly added according to a preset rule includes:
[0015] Performing the robot dual-arm tightly coupled cooperation task in an environment that meets the preset criteria, obtaining the task data of the task that meets the preset criteria at the target moment, and forming a basic training data set from the task data, where the preset criteria at least include no collision and completing the task within a preset time;
[0016] Randomly adding obstacles in the environment according to a preset rule, and generating an enhanced training data set based on the obstacles, where the preset rule at least includes that the obstacles do not collide with the robot.
[0017] In some embodiments of the present disclosure, the randomly adding obstacles in the environment according to a preset rule includes:
[0018] Approximating the geometric shape of the robotic arm as the union of multiple spheres;
[0019] Given the motion trajectory of the robotic arm, calculating the positions of all spheres forming the robotic arm at any moment to obtain the swept volume of the robotic arm corresponding to the motion trajectory;
[0020] Randomly sampling the center positions of the spheres representing obstacles and specifying the radii in the part of the robotic arm workspace not covered by the swept volume of the robotic arm, and at the same time checking whether the spheres intersect with the swept volume of the robotic arm. If they intersect, discard this sampling and resample until the newly added spheres have no intersecting part with the swept volume of the robotic arm.
[0021] In some embodiments of the present disclosure, the obtaining the enhanced training data set based on the obstacles includes:
[0022] Generating obstacle information data represented by the spheres based on the center positions and radii of the spheres;
[0023] Augment the first task data in the basic training dataset based on the obstacle information data to generate second task data;
[0024] The enhanced training dataset is jointly composed of the first task data and the second task data.
[0025] In some embodiments of the present disclosure, the imitation learning model is generated by training a supervised learning model with a training dataset, including:
[0026] Using the task data in the enhanced training dataset as samples and the motion path of the master manipulator at the next moment corresponding to the task data as labels, train the supervised learning model to generate the imitation learning model.
[0027] A second aspect of the embodiments of the present disclosure provides a robot double-arm tightly coupled cooperation device, including:
[0028] A prediction module, configured to obtain the task data of the robot double-arm tightly coupled cooperation task, input the task data into the imitation learning model, and generate the motion path of the master manipulator, where the imitation learning model is generated by training a supervised learning model with a training dataset obtained by randomly adding obstacles in a preset environment according to preset rules, and the task data includes manipulator state data, obstacle information data, and target information data;
[0029] A first control module, configured to control the master manipulator of the robot to move along the motion path;
[0030] A second control module, configured to control the slave manipulator of the robot to perform synchronous motion following the master manipulator.
[0031] A third aspect of the embodiments of the present disclosure provides a robot, including a memory and a processor,
[0032] The memory is used to store a computer program;
[0033] The processor is configured to, when executing the computer program, implement the robot double-arm tightly coupled cooperation method according to the first aspect of the embodiments of the present disclosure.
[0034] A fourth aspect of the embodiments of the present disclosure provides a computer program product, including computer programs / instructions, and when the computer programs / instructions are executed by a processor, the robot double-arm tightly coupled cooperation method according to the first aspect of the embodiments of the present disclosure is implemented.
[0035] In summary, the robot dual-arm tightly coupled cooperation methods, devices, robots, and computer program products provided by the various embodiments of the present disclosure obtain a high-quality training data set through a two-stage data generation method. In particular, in the second stage, the basic training data set is enhanced in the form of randomly adding obstacles based on geometric approximation and hallucination mechanisms, thereby greatly increasing the quantity and diversity of training data, making it possible to train an imitation learning model. Then, through a hierarchical framework that combines imitation learning and master-slave control, real-time collision-free path planning can be performed while ensuring the stability of the closed-chain constraint, enabling real-time collision-free task execution for tightly coupled cooperation in complex static / dynamic environments. At the same time, problems such as parameter sensitivity and poor stability of existing methods are solved, providing a new solution for dual-arm tightly coupled cooperation tasks. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] The features and advantages of the present disclosure will be more clearly understood by referring to the accompanying drawings. The drawings are schematic and should not be construed as imposing any limitation on the present disclosure. In the drawings:
[0037] Figure 1 is a schematic diagram of a dual-arm tightly coupled cooperation data generation method based on imitation learning and its task framework proposed by the present disclosure;
[0038] Figure 2 is a schematic diagram of a computer system applicable to the present disclosure;
[0039] Figure 3 is a flowchart of a robot dual-arm tightly coupled cooperation method according to some embodiments of the present disclosure;
[0040] Figure 4 is the generation process of the training data set of the imitation learning model of the present disclosure;
[0041] Figure 5 is a schematic diagram of a hierarchical framework that combines imitation learning and master-slave control proposed by the present disclosure;
[0042] Figure 6 is based on an embodiment of the present disclosure Figure 3 The time screenshot of performing the mobile coffee task according to the method described in S310 - S330;
[0043] Figure 7 is a schematic diagram of a robot dual-arm tightly coupled cooperation device according to some embodiments of the present disclosure;
[0044] Figure 8 is a schematic diagram of a robot according to some embodiments of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0045] In the following detailed description, numerous specific details are set forth by way of examples in order to provide a thorough understanding of the relevant disclosures. However, it will be apparent to those of ordinary skill in the art that the present disclosure may be practiced without these details. It should be understood that the terms "system", "device", "unit", and / or "module" used in the present disclosure are a means for distinguishing different components, elements, parts, or assemblies at different levels in a sequential arrangement. However, if other expressions can achieve the same purpose, these terms may be replaced by other expressions.
[0046] It should be understood that when a device, unit, or module is referred to as being "on", "connected to", or "coupled to" another device, unit, or module, it can be directly on, connected to, or coupled to or communicate with the other device, unit, or module, or there may be intermediate devices, units, or modules, unless the context clearly dictates otherwise. For example, the term "and / or" used in the present disclosure includes any and all combinations of one or more of the associated listed items.
[0047] The terms used in the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present disclosure. As shown in the specification and claims of the present disclosure, unless the context clearly dictates otherwise, the words "a", "an", "one", and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the features, wholes, steps, operations, elements, and / or components that have been clearly identified, and such expressions do not constitute an exclusive listing, and other features, wholes, steps, operations, elements, and / or components may also be included.
[0048] Referring to the following description and the accompanying drawings, these or other features and characteristics of the present disclosure, the operating methods, the functions of the relevant elements of the structure, the combination of parts, and the economy of manufacture can be better understood, where the description and the drawings form a part of the specification. However, it is clearly understood that the drawings are only for the purpose of illustration and description and are not intended to limit the scope of protection of the present disclosure. It can be understood that the drawings are not drawn to scale.
[0049] A variety of structure diagrams are used in the present disclosure to illustrate various deformations according to the embodiments of the present disclosure. It should be understood that the foregoing or following structures are not used to limit the present disclosure. The scope of protection of the present disclosure is defined by the claims.
[0050] Currently, service robots are attracting more and more attention. Service robots with dual-arm systems have great potential in future service robot applications because they can perform complex operation tasks. As a type of dual-arm cooperation task, the dual-arm tightly coupled cooperation task forms a strict closed-chain constraint between the end-effectors of the two robotic arms and the object, requiring the robotic arms and the object to maintain highly coordinated synchronous motion. In actual applications, it is also necessary to perform real-time collision-free path planning for the robot while ensuring the stability of the closed-chain constraint, which is also a research difficulty in this field. The current mainstream solution is to plan the motion path of the object and let the robotic arm perform path tracking. Among them, a task framework combining reinforcement learning and master-slave control has been proposed, which trains the obstacle avoidance strategy through reinforcement learning and ensures the stability of the closed-chain constraint through master-slave control. However, it is difficult to guarantee the convergence effect and generality of reinforcement learning. There is also a solution that uses a rule-based path planner plus a task priority control strategy to execute this task. However, the rule-based planner is sensitive to parameters and has poor adaptability to the environment, and can often only be used in simple environments, lacking the possibility of application in complex actual scenarios.
[0051] Imitation learning is a technique that uses neural networks to learn from expert data, enabling robots to learn to perform specific tasks. Imitation learning has proven its effectiveness in a large number of robot tasks. However, imitation learning requires a large amount of expert data. For the dual-arm tightly coupled cooperation task, at least the robotic arm state data, obstacle information data, and target information data at the same moment need to be collected synchronously and accurately. There are many types of data to be acquired, and there are large modal differences. Therefore, it is difficult to obtain sufficient expert data for training the imitation learning model for the dual-arm tightly coupled cooperation task. Therefore, imitation learning has not been used for this type of task at present.
[0052] In view of this, the present disclosure proposes a method for generating dual-arm tightly coupled cooperation data based on imitation learning and its task framework. First, a high-quality dataset for training the imitation learning model is obtained by using geometric approximation and hallucination mechanisms. Then, a hierarchical framework combining imitation learning and master-slave control is used to perform real-time collision-free path planning while ensuring the stability of the closed-chain constraint. The method for generating dual-arm tightly coupled cooperation data based on imitation learning and its task framework are as Figure 1 shown.
[0053] Figure 2 is a schematic diagram of a computer system applicable to the present disclosure. Figure 2 The computer system shown includes a dual-arm cooperation task server that is data-connected to the robot. The robot has a dual-arm system. The dual-arm cooperation task server can generate a dual-arm tightly coupled cooperation task based on a user instruction and control the robot to implement the dual-arm tightly coupled cooperation task.
[0054] The dual-arm collaborative task server is deployed with an imitation learning model. The imitation learning model can generate the motion path of the master manipulator of the robot based on user instructions or other task requirements. The motion path refers to a path composed of a series of path points in the Cartesian space, ensuring that the master manipulator does not collide with obstacles when moving along the path. The dual-arm collaborative task server can be any one of a single machine, a cluster, or a distributed server. In particular, the dual-arm collaborative task server can be deployed on the robot.
[0055] Figure 3 is a flowchart of a method for tightly coupled collaboration of a robot's dual arms according to some embodiments of the present disclosure. In some embodiments, the method for tightly coupled collaboration of a robot's dual arms is executed by Figure 2 the dual-arm collaborative task server shown, and the method includes the following steps:
[0056] S310, obtain the task data of the tightly coupled collaboration task of the robot's dual arms, input the task data into the imitation learning model, and generate the motion path of the master manipulator, where the imitation learning model is trained and generated by a supervised learning model using a training data set obtained by randomly adding obstacles in a preset environment according to preset rules, and the task data includes manipulator state data, obstacle information data, and target information data.
[0057] The present disclosure first constructs an imitation learning model. The imitation learning model is trained and generated by a mainstream supervised learning model. The training process is executed by a training server. In some embodiments of the present disclosure, the training server is deployed on the dual-arm collaborative task server.
[0058] In the present disclosure, the generation of training data is divided into two stages, as Figure 4 shown.
[0059] In the first stage, data collection is performed on successful tasks in a simple environment. The simple environment refers to an environment with only a few simple obstacles (as Figure 2 shown in the basic data part). A successful task refers to a task that does not collide with obstacles and is completed within a specified time. The purpose of this stage is to obtain high-quality basic task data, and the task data includes at least manipulator state data, obstacle information data, and target information data.
[0060] In the second stage, data augmentation is performed on the basis of the basic task data obtained in the first stage. The specific method is to randomly add obstacles in the environment (such as Figure 2 the obstacles with dashed borders shown in the augmented data part), and the requirement is that the newly added obstacles must ensure that they do not collide with the robot. The purpose of this stage is to enhance the diversity of the data. Finally, a large-scale and high-quality data set can be obtained.
[0061] Some embodiments of the present disclosure randomly add obstacles based on geometric approximation and hallucination mechanisms, as follows:
[0062] Geometric approximation
[0063] The geometric shape B of the robotic arm can be approximated as the union of N s spheres, as follows:
[0064]
[0065] Where:
[0066] S i (c i , r i ) represents the i-th sphere
[0067] c i is the center coordinate of the sphere
[0068] r i is the radius of the sphere
[0069] All spheres are attached to the links of the robotic arm, and their positions can be dynamically updated through forward kinematics as the robotic arm moves.
[0070] Swept volume construction
[0071] Given the motion trajectory of the robotic arm where q t ∈R n represents the joint configuration of the robotic arm at time t, n is the number of joints of the robotic arm, and T represents the total number of time steps of the trajectory.
[0072] At any moment, the position of each sphere can be obtained through c i (q t ) = f i (q t ).
[0073] Calculating the positions of all spheres at any moment, the swept volume V s (τ) of the robotic arm corresponding to this trajectory can be obtained, and the specific formula is as follows:
[0074]
[0075] It represents the set of the space swept by all links of the robotic arm during the motion
[0076] Adding obstacles based on the hallucination mechanism
[0077] Define the workspace of the robotic arm as V ws
[0078] Then, the free space available for placing obstacles under the given trajectory τ is V free (τ) = V ws - V s (τ), that is, the part of the workspace not covered by the swept volume.
[0079] Randomly sample the center p of the obstacle in the free space V free (τ) and specify its radius r j (s j ∈ [s j , s min , s max ), and ensure that the sphere does not intersect with the swept volume, that is, ensure that the newly added obstacle does not collide with the robotic arm.
[0080] Randomly add N (0 - 20, the number can be customized) obstacles to each basic dataset for data augmentation, and a new dataset can be obtained.
[0081] In the first stage of the present disclosure, existing methods are used for basic data collection, which can ensure the high quality of the robot trajectories in the dataset. In the second stage, geometric approximation and hallucination mechanism are used to augment the basic dataset, which can greatly increase the diversity of the dataset. Through the two-stage data generation method proposed by the present disclosure, a high-quality dataset can be finally obtained.
[0082] If there is only the first stage in data collection, the diversity of the acquired training data is insufficient, and a general imitation learning model cannot be finally trained. If there is only the second stage, the rationality of the robot trajectories in the corresponding environment cannot be guaranteed.
[0083] After obtaining the training dataset, using the task data in the training dataset as samples and the motion path of the main robotic arm at the next moment corresponding to the task data as labels, the supervised learning model is trained to generate an imitation learning model.
[0084] After constructing an imitation learning model based on the augmented training data, the imitation learning model can be deployed on the dual-arm collaboration task server. When the dual-arm collaboration task server receives user requirements and parses out task data based on the user requirements, the task data can be input into the imitation learning model to generate the motion path of the main robotic arm of the robot. Among them, the task data at least includes robotic arm state data, obstacle information data, and target information data.
[0085] S320, control the main robotic arm of the robot to move along the motion path.
[0086] S330, control the slave robotic arm of the robot to perform synchronous motion following the main robotic arm.
[0087] The present disclosure performs a dual-arm tightly coupled collaborative task based on a hierarchical framework that combines imitation learning and master-slave control, such as Figure 5 shown.
[0088] The input of the planner based on imitation learning includes the manipulator state, obstacle information, and task objective information, and the output is collision-free path points in the Cartesian space. Compared with the rule-based planner that does not consider the robot state, the planner of the present disclosure can, through training on a large amount of data, have a certain understanding of the robot state and learn an efficient collision-free planning strategy.
[0089] Master-slave control is a classic scheme for realizing dual-arm coordinated motion in a dual-arm tightly coupled task, which can ensure the stability of the closed-chain constraint to the greatest extent, such as Figure 5 shown in the right half. Under this framework, the master manipulator is responsible for tracking the object's motion, while the slave manipulator performs real-time error compensation according to the motion of the master manipulator to ensure the consistency of the dual-arm motion. By following the path points generated by the planner based on imitation learning through this controller and continuously feeding back the manipulator state to the controller, it is possible to meet the real-time collision-free path planning while ensuring the stability of the constraint.
[0090] Although the training data are all static data, because (1) the data set is diverse enough and large enough, and (2) the entire framework is for real-time planning and real-time control, the above method can also effectively handle dynamic obstacles by treating the dynamically changing scene as a series of continuous static scenes.
[0091] An embodiment of the present disclosure performs a mobile coffee task based on Figure 3 the method described in S310 - S330 (referred to as our method) in Figure 6 shown. Figure 6 In
[0092] Figure 7 is a schematic diagram of a robot dual-arm tightly coupled collaborative device according to some embodiments of the present disclosure. As Figure 7 shown, the robot dual-arm tightly coupled collaborative device 700 includes a prediction module 710, a first control module 720, and a second control module 730. In some embodiments of the present disclosure, the robot dual-arm tightly coupled collaborative function is executed by Figure 2 the dual-arm collaborative task server shown in
[0093] A prediction module 710, configured to obtain task data of a robot's two-arm tightly coupled collaborative task, input the task data into an imitation learning model, and generate a motion path of the main manipulator, where the imitation learning model is trained and generated by a supervised learning model using a training data set obtained by randomly adding obstacles according to preset rules in a preset environment, and the task data includes manipulator state data, obstacle information data, and target information data;
[0094] A first control module 720, configured to control the main manipulator of the robot to move along the motion path;
[0095] A second control module 730, configured to control the slave manipulator of the robot to perform synchronous motion following the main manipulator.
[0096] Figure 8 is a schematic diagram of a robot according to some embodiments of the present disclosure. As Figure 8 shown, the robot 800 includes a memory 820 and a processor 810. The memory 820 is configured to store a computer program; the processor 810 is configured to, when executing the computer program, implement Figure 3 the robot two-arm tightly coupled collaborative method described in S310 - S330 in
[0097] An embodiment of the present disclosure provides a computer program product, including computer programs / instructions, which, when executed by a processor, implement Figure 3 the robot two-arm tightly coupled collaborative method described in S310 - S330 in
[0098] In summary, for the robot two-arm tightly coupled collaborative method, device, robot, and computer program product provided by the embodiments of the present disclosure, a high-quality training data set is obtained through a two-stage data generation method. In particular, in the second stage, the basic training data set is enhanced in the form of randomly adding obstacles based on geometric approximation and hallucination mechanisms, thereby greatly increasing the quantity and diversity of training data, making it possible to train an imitation learning model; then, through a hierarchical framework combining imitation learning and master-slave control, real-time collision-free path planning can be performed while ensuring the stability of the closed-chain constraint, enabling real-time collision-free task execution for tightly coupled collaboration in complex static / dynamic environments, and at the same time solving problems such as parameter sensitivity and poor stability of existing methods, providing a new solution for two-arm tightly coupled collaborative tasks.
[0099] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding descriptions in the foregoing device embodiments and will not be elaborated herein.
[0100] Although the subject matter described herein is provided in the general context of execution in conjunction with the execution of an operating system and applications on a computer system, those skilled in the art will recognize that other implementations may also be performed in conjunction with other types of program modules. In general, program modules include routines, programs, components, data structures, and other types of structures that perform particular tasks or implement particular abstract data types. Those skilled in the art will understand that the subject matter described herein may be practiced using other computer system configurations, including handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, minicomputers, mainframe computers, etc., and may also be used in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.
[0101] Those of ordinary skill in the art will appreciate that the elements and method steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether such functions are executed in hardware or software depends upon the particular application and design constraints of the technical solution. Skilled artisans may use different methods for each particular application to implement the described functions, but such implementation should not be considered to exceed the scope of the present disclosure.
[0102] It should be understood that the above specific embodiments of the present disclosure are merely for illustrative or explanatory purposes of the principles of the present disclosure, and do not constitute a limitation to the present disclosure. Therefore, any modifications, equivalent replacements, improvements, etc. made without departing from the spirit and scope of the present disclosure shall be included within the protection scope of the present disclosure. In addition, the appended claims of the present disclosure are intended to cover all changes and modifications that fall within the scope and boundaries of the appended claims, or within the equivalent forms of such scope and boundaries.
Claims
1. A method for tightly coupled cooperation of a robot's two arms, characterized in that, Including: Obtain the task data of the robot's two-arm tightly coupled collaborative task, input the task data into the imitation learning model to generate the motion path of the main manipulator, where the imitation learning model is trained and generated by a supervised learning model using a training data set obtained by randomly adding obstacles in a preset environment according to preset rules, and the task data includes manipulator state data, obstacle information data, and target information data; Control the main manipulator of the robot to move along the motion path; Control the slave manipulator of the robot to perform synchronous motion following the main manipulator.
2. The method according to claim 1, wherein The motion path refers to: A path composed of a series of path points in the Cartesian space, satisfying that the main manipulator will not collide with obstacles when moving along the path.
3. The method according to claim 1, characterized in that The control of the slave manipulator of the robot to perform synchronous motion following the main manipulator includes: Perform error compensation on the slave manipulator in real time according to the motion of the main manipulator to ensure the stability of the closed-loop constraint between the master and slave manipulators.
4. The method according to claim 1, wherein The training and generation by a supervised learning model using a training data set obtained by randomly adding obstacles in a preset environment according to preset rules includes: Execute the robot's two-arm tightly coupled collaborative task in an environment meeting the preset standards, obtain the task data of the task meeting the preset standards at the target moment, and form a basic training data set from the task data, where the preset standards at least include no collision and completion of the task within a preset time; Randomly add obstacles in the environment according to preset rules, and generate an enhanced training data set based on the obstacles, where the preset rules at least include that the obstacles do not collide with the robot.
5. The method according to claim 4, wherein The randomly adding obstacles in the environment according to preset rules includes: Approximate the geometric shape of the manipulator as the union of multiple spheres; Given the motion trajectory of the manipulator, calculate the positions of all spheres forming the manipulator at any moment to obtain the swept volume of the manipulator corresponding to the motion trajectory; Randomly sample the center positions of the spheres representing obstacles and specify the radii in the part of the manipulator's workspace not covered by the swept volume of the manipulator, and at the same time check whether the spheres intersect with the swept volume of the manipulator. If they intersect, abandon this sampling and resample until the newly added spheres have no intersecting parts with the swept volume of the manipulator.
6. The method according to claim 5, wherein The obtaining of the enhanced training data set based on the obstacles includes: Generate the obstacle information data represented by the spheres based on the center positions and radii of the spheres; Expand the first task data in the basic training data set based on the obstacle information data to generate second task data; The enhanced training data set is jointly composed of the first task data and the second task data.
7. The method according to claim 6, characterized in that, The training and generation of the imitation learning model by a supervised learning model using a training data set includes: Using the task data in the enhanced training data set as samples and the motion path of the main manipulator at the next moment corresponding to the task data as labels, train the supervised learning model to generate the imitation learning model.
8. A robot double-arm tightly coupled cooperation device, characterized in that, Including: A prediction module, configured to obtain task data of the tightly coupled collaborative task of the robot's two arms, input the task data into an imitation learning model, and generate a motion path of the main manipulator, wherein the imitation learning model is trained and generated by a supervised learning model using a training data set obtained by randomly adding obstacles according to preset rules in a preset environment, and the task data includes manipulator state data, obstacle information data, and target information data; A first control module, configured to control the main manipulator of the robot to move along the motion path; A second control module, configured to control the slave manipulator of the robot to perform synchronous movement following the main manipulator.
9. A robot, characterized in that: Comprising a memory and a processor, The memory is configured to store a computer program; The processor is configured to, when executing the computer program, implement the robot two-arm tightly coupled collaboration method according to any one of claims 1-7.
10. A computer program product, comprising computer program / instructions, which when executed by a processor implement the robot two-arm tightly coupled collaboration method according to any one of claims 1-7.