Robot flexible sleeve assembly system and method based on self-adaptive variable impedance

Abstract

The invention discloses a robot flexible sleeve assembly system and method based on self-adaptive variable impedance, solves the problem that a robot in the prior art is difficult to mount a flexible sleeve, and enables the robot to achieve the beneficial effect of sleeving of the flexible sleeve. According to the specific scheme, the robot flexible casing pipe assembling system based on the self-adaption variable impedance comprises a visual module, a first camera used for collecting images of a flexible casing pipe and a second camera used for collecting images of a bent hard pipe; the force sensor, the robot, the first camera, the second camera and the force sensor are separately connected with the controller, and the controller recognizes the pose of the flexible sleeve according to the flexible sleeve image and captures the pose track of the bent hard tube according to the bent hard tube image. And the controller constructs a variable impedance parameter learning network according to the hard tube pose trajectory data and the contact force dynamic data sent by the force sensor, performs real-time analysis on the contact force state, and outputs a rigidity parameter and a damping parameter matched with the current state.

Description

Technical Field

[0001] The present invention relates to the technical field of robot adaptive assembly, in particular to a robot flexible sleeve assembly system and method based on adaptive variable impedance. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Robots can install pipelines industrially, effectively replacing manual operations. For example, a straight pipe can be sleeved on a straight pipe. However, in the technical field of robot adaptive assembly, achieving high-precision assembly under complex physical interaction conditions is still a technical challenge that needs to be overcome urgently. This problem is particularly prominent in the assembly scenarios of complex working conditions of flexible sleeves such as the air-conditioning pipelines of new energy vehicles. That is, when a robot holds a flexible sleeve and wants to sleeve it on a bent hard pipe, during the assembly process of the robot flexible sleeve and the bent hard pipe (such as the nesting of a metal bent pipe and a flexible foam pipe), the non-linear elastic deformation characteristics of the flexible sleeve lead to a strong dynamic coupling effect between the contact force and the deformation, resulting in frequent occurrence of assembly deviation and pipe body damage problems. At the same time, due to the differences in pipe body length (50 - 200 mm) and the number of spatial inflection points for different specifications of sleeve tasks, there are significant differences in the impedance parameters during the deformation process. If a robot is used for installation, traditional impedance control relies on manually preset fixed stiffness and damping parameters. Although it can achieve basic force interaction, its fixed parameters cannot adapt to the dynamic deformation process of multi-specification sleeves and cannot match the deformation degree and the change rate of the contact force. Some solutions use geometric modeling methods, but their accuracy is limited because they cannot represent the dynamic contact force-deformation relationship. The physical model method is restricted by the complexity of material parameter calibration. The force control strategy based on supervised learning is limited in generalization ability due to the strong randomness of the deformation of the flexible sleeve and is prone to problems such as sudden changes in contact force and pipe body damage under different working conditions of sleeve tasks.

[0004] The above reasons make it impossible for the robot to sleeve the flexible sleeve on the bent hard pipe by itself, resulting in the current installation of the flexible sleeve still being manually operated, facing problems such as low efficiency and poor assembly consistency. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a robot flexible sleeve assembly system based on adaptive variable impedance, which takes into account the characteristics of the flexible sleeve and the bent hard pipe, so that the robot can smoothly sleeve the flexible sleeve on the bent hard pipe.

[0006] To achieve the above purpose, the present invention is realized through the following technical solutions:

[0007] A robot flexible sleeve assembly system based on adaptive variable impedance, comprising:

[0008] A robot, the robot is provided with a mechanical gripper for grasping a flexible sleeve;

[0009] A vision module, including a first camera for collecting images of the flexible sleeve and a second camera for collecting images of the bent rigid tube;

[0010] A force sensor, the force sensor is arranged at the mechanical gripper;

[0011] A controller, the robot, the first camera, the second camera and the force sensor are respectively and independently connected to the controller. The controller identifies the pose of the flexible sleeve according to the image of the flexible sleeve so that the mechanical gripper of the robot can grasp the flexible sleeve, and captures the pose trajectory of the bent rigid tube according to the image of the bent rigid tube. The controller constructs a variable impedance parameter learning network according to the pose trajectory data of the rigid tube and the dynamic contact force data sent by the force sensor. The controller analyzes the contact force state in real time according to the variable impedance parameter learning network, and outputs the stiffness parameter and damping parameter matching the current state. The controller generates an end trajectory correction instruction according to the output stiffness parameter and damping parameter, drives the work of the robot to perform compliant adjustment, and realizes the installation of the flexible sleeve at the bent rigid tube.

[0012] A robot flexible sleeve assembly system based on adaptive variable impedance as described above, the first camera is installed at the end of the robotic arm of the robot, and the second camera is fixed obliquely above the bent rigid tube.

[0013] A robot flexible sleeve assembly system based on adaptive variable impedance as described above, the controller captures the pose trajectory of the bent rigid tube through the following:

[0014] The controller filters the noise of the obtained image of the bent rigid tube, reconstructs the structure of the bent rigid tube based on the curve skeleton extraction algorithm, generates candidate skeleton nodes discretely distributed inside the bent rigid tube. After spatial filtering and shrinkage optimization, a refined trajectory is fitted, and further a centered skeleton node is generated, and finally the pose trajectory of the bent rigid tube is obtained.

[0015] A robot flexible sleeve assembly system based on adaptive variable impedance as described above, the controller uses one-dimensional moving least squares method to fit and generate the refined trajectory; the controller combines the cross-section cutting and ellipse fitting technology of the bent rigid tube to reposition the geometric center to generate the centered skeleton node.

[0016] A robot flexible sleeve assembly system based on adaptive variable impedance as described above, wherein the controller uses a second-order dynamic model to define the dynamic characteristics of the robot. The controller follows the pose trajectory of the bent rigid tube and calculates the pose correction amount through an impedance model according to the contact force sent by the force sensor, and superimposes the calculated pose correction amount onto the pose trajectory of the bent rigid tube to form a dynamically adjusted pose, and then constructs the variable impedance parameter learning network.

[0017] A robot flexible sleeve assembly system based on adaptive variable impedance as described above, wherein the controller defines a variable impedance parameter learning network, inputs the current assembly position of the flexible sleeve and the contact force and torque information received by the flexible sleeve during the sleeving process, and defines the dynamic impedance control parameters as the dynamic stiffness parameter and the dynamic damping parameter during the sleeving process.

[0018] A robot flexible sleeve assembly system based on adaptive variable impedance as described above, wherein the controller dynamically adjusts the stiffness and damping parameters through a reinforcement learning algorithm, including the following contents:

[0019] Formula definition and determination of the reinforcement learning algorithm framework;

[0020] Reward function design;

[0021] Training process of the reinforcement learning algorithm network, repeating the reward function design and training of the reinforcement learning algorithm until the loop terminates.

[0022] A robot flexible sleeve assembly system based on adaptive variable impedance as described above, wherein the formula definition includes the determination of the maximum total reward formula, and the maximum total reward formula is determined by the entropy regularization coefficient, the entropy value, the defined current assembly position of the flexible sleeve and the contact force and torque information received by the flexible sleeve during the sleeving process, and the dynamic impedance control parameters.

[0023] A robot flexible sleeve assembly system based on adaptive variable impedance as described above designs the reward function through multiple parameter and contact force constraints, torque constraints and terminal rewards.

[0024] In a second aspect, the present invention also provides a robot flexible sleeve assembly method based on adaptive variable impedance, including the following contents:

[0025] The first camera captures an image of the flexible sleeve, and the second camera captures an image of the bent rigid tube. The first camera and the second camera send the captured images to the controller. The controller identifies the pose of the flexible sleeve according to the flexible sleeve image and captures the pose trajectory of the bent rigid tube according to the bent rigid tube image;

[0026] The controller grasps the flexible sleeve according to the pose of the flexible sleeve;

[0027] The force sensor obtains the magnitude of the contact force when the robotic manipulator gripper grasps the flexible sleeve and sends it to the controller;

[0028] The controller constructs a variable impedance parameter learning network based on the rigid tube pose trajectory data and the contact force dynamic data. The controller analyzes the contact force state in real time according to the variable impedance parameter learning network and outputs the stiffness parameter and damping parameter that match the current state;

[0029] The controller generates an end trajectory correction instruction according to the output stiffness parameter and damping parameter, drives the robot to perform compliant adjustment, and realizes the installation of the flexible sleeve at the bent rigid tube.

[0030] The beneficial effects of the present invention are as follows:

[0031] 1) The robotic flexible sleeve assembly system provided by the present invention takes into account the spatial inflection point of the bent rigid tube, can obtain the rigid tube pose trajectory data, fully considers the change of the contact force caused by the non-linear elastic deformation of the flexible sleeve, obtains the contact force of the robotic manipulator gripper grasping the flexible sleeve through the force sensor, constructs a variable impedance parameter learning network according to the rigid tube pose trajectory data and the contact force dynamic data sent by the force sensor. In this way, considering the change of the impedance parameters during the sleeving process, the stiffness parameter and damping parameter that match the current state are output to drive the robot to perform compliant adjustment, not only realizing the sleeving of the flexible sleeve on the bent rigid tube, but also enabling the robot to adaptively adjust the compliance during the assembly process, effectively avoiding excessive deformation of the flexible sleeve such as a foam tube.

[0032] 2) In the present invention, the positions of the first camera and the second camera are reasonably installed. The first camera is reasonably set to obtain the pose of the flexible sleeve so that the robotic manipulator gripper can grasp the flexible sleeve. The position of the second camera is reasonably set, and the controller can obtain the pose trajectory of the bent rigid tube.

[0033] 3) In the present invention, the controller calculates the pose correction amount through the impedance model, superimposes the calculated pose correction amount on the rigid tube pose trajectory, forms a dynamic adjustment and then constructs a variable impedance parameter learning network. The controller defines the variable impedance parameter learning network, inputs the current assembly position of the flexible sleeve and the contact force and torque information of the flexible sleeve during the sleeving process, defines the dynamic impedance control parameters as the dynamic stiffness parameter and dynamic damping parameter during the sleeving process, and considers reasonable factors to realize the establishment of the variable impedance parameter learning network.

[0034] 4) In the present invention, the controller dynamically adjusts the stiffness and damping parameters through a reinforcement learning algorithm. The reinforcement learning algorithm continuously updates the reward function to facilitate the output of optimal dynamic stiffness and dynamic damping parameters, which is beneficial to balancing the trajectory tracking accuracy of the bending hard tube and the compliance of force control, overcomes the limitation of traditional impedance control relying on manually preset parameters, and also improves the generalization ability of the robot.

[0035] 5) In the present invention, the controller dynamically generates a matching impedance parameter combination based on the pose trajectory data and real-time contact force information of the bending hard tube, enabling the robot to adaptively adjust the compliance during the assembly process. This method breaks through the contradiction between the fixed parameters of traditional impedance control and complex assembly working conditions, and improves the generalization ability to the random deformation of the flexible sleeve through the reinforcement learning algorithm. In this way, the robot flexible sleeve assembly technology realizes high-precision automated operation of sleeve assembly under complex working conditions, providing key process guarantees for the intelligent upgrading of new energy vehicle air conditioner manufacturing or other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0037] Figure 1 is a flowchart of a robot flexible sleeve assembly system based on adaptive variable impedance according to one or more embodiments of the present invention.

[0038] Figure 2 is a working principle diagram of a robot flexible sleeve assembly system based on adaptive variable impedance according to one or more embodiments of the present invention.

[0039] Figure 3 is a flowchart of constructing a variable impedance parameter learning network by the controller of a robot flexible sleeve assembly system based on adaptive variable impedance according to one or more embodiments of the present invention.

[0040] Figure 4 is a flowchart of training the SAC learning network in a robot flexible sleeve assembly system based on adaptive variable impedance according to one or more embodiments of the present invention.

[0041] In the figure: The distances or sizes between each part are exaggerated for showing the positions of each part, and the schematic diagram is only for illustration. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0043] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the present invention clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof;

[0044] As introduced in the background art, in the prior art, robots are unable to sleeve a flexible sleeve onto a bent hard pipe. To solve the above technical problems, the present invention proposes a robot flexible sleeve assembly system based on adaptive variable impedance.

[0045] Embodiment 1

[0046] In a typical embodiment of the present invention, referring to Figure 1 and Figure 2 as shown, a robot flexible sleeve assembly system based on adaptive variable impedance, which can realize sleeving a flexible sleeve such as a foam tube onto a bent hard pipe such as a metal bent pipe, includes:

[0047] A robot, which is provided with a mechanical gripper to grasp the flexible sleeve;

[0048] A vision module, including a first camera for collecting images of the flexible sleeve and a second camera for collecting images of the bent hard pipe;

[0049] A force sensor, which is arranged at the mechanical gripper, and the force sensor is a six-dimensional force sensor;

[0050] A controller, the robot, the first camera, the second camera, and the force sensor are respectively and independently connected to the controller. The controller identifies the pose of the flexible sleeve according to the flexible sleeve image so that the robot mechanical gripper can grasp the flexible sleeve, and captures the pose trajectory of the bent hard pipe according to the bent hard pipe image. The controller constructs a variable impedance parameter learning network based on the hard pipe pose trajectory data and the contact force dynamic data sent by the force sensor. The controller performs real-time analysis on the contact force state according to the variable impedance parameter learning network, and outputs stiffness parameters and damping parameters matching the current state. The controller generates an end trajectory correction instruction according to the output stiffness parameters and damping parameters, and drives the operation of the robot to perform compliant adjustment to realize the installation of the flexible sleeve at the bent hard pipe.

[0051] Among them, it should be noted that the first camera is installed at the end of the robotic arm of the robot so that the first camera can face the flexible sleeve to be grasped by the robotic gripper. The first camera is an existing two-dimensional camera, and the second camera is fixed obliquely above the bent hard tube. The second camera is a binocular depth camera; the controller is the controller of the robot, and the controller can be a PLC controller or other types of controllers.

[0052] It should be noted that the controller captures the pose trajectory of the bent hard tube through the following content:

[0053] The controller filters the noise of the obtained image of the bent hard tube, reconstructs the structure of the bent hard tube based on the curve skeleton extraction algorithm, generates candidate skeleton nodes with discrete distribution inside the bent hard tube, and after spatial filtering and shrinkage optimization, the controller uses the one-dimensional moving least squares method to fit and generate a refined trajectory, combines the cross-sectional cutting and ellipse fitting technology of the bent hard tube to reposition the geometric center, to generate the centered skeleton nodes, and finally obtains the pose trajectory X of the bent hard tube d =(p x ,p y ,p z ,r x ,r y ,r z ).

[0054] Furthermore, the controller adopts a second-order dynamic model (impedance model): Ms 2 +Bs+K to define the dynamic characteristics of the robot. The inertia matrix M is a set range value. The controller follows the pose trajectory X of the bent hard tube d , and according to the contact force F sent by the force sensor e =(F x ,F y ,F z ,T x ,T y ,T z ), calculates the pose correction amount e=(Δx,Δy,Δz,Δθ x ,Δθ y ,Δθ z ), where Δx, Δy, and Δz are the axial displacement increments at the end of the robotic arm, and Δθ x ,Δθ y ,Δθ z are the rotational angle increments of the end of the robotic arm around the axis. The calculated pose correction amount is superimposed on the pose trajectory of the bent hard tube, forms a dynamically adjusted and then constructs a variable impedance parameter learning network, and adaptively adjusts the damping matrix B and the stiffness matrix K through the SAC reinforcement learning algorithm to balance the trajectory tracking accuracy and the force control compliance.

[0055] ​Among them, the impedance model realizes compliant control by dynamically adjusting the end pose of the robot to respond to external contact forces. Its core consists of a damping matrix B and a stiffness matrix K, and the dynamic properties of the robot are changed by adjusting these two parameters.

[0056] Next, referring to Figure 3 as shown, the controller defines a variable impedance parameter learning network, with the input state s t =(s F , d t ), where s F =(F x , F y , F z , T x , T y , T z ) being the contact forces on the flexible casing during the casing process. Among them, F x , F y , F z are force information, T x , T y , T z are torque information, d t being the current assembly position of the flexible casing, d t =(d x , d y , d z ), and defining the dynamic impedance control parameter a t =(B, K), the dynamic stiffness parameter B and the dynamic damping parameter K during the casing process.

[0057] Finally, the controller dynamically adjusts the stiffness and damping parameters through a reinforcement learning algorithm, including the following:

[0058] 1) Defining the formula and determining the reinforcement learning algorithm (SAC) framework:

[0059] By optimizing the stochastic policy, the SAC algorithm can obtain high cumulative rewards and enhance the exploration ability of the agent by maximizing entropy:

[0060]

[0061] H(π(·|s t )) = E[-logπ(·|s t )] (Formula 2)

[0062] where π * is the optimal policy, representing the policy found through optimization that can maximize the sum of the cumulative reward and the entropy-weighted sum; π is used to update the policy that has found the maximum total reward; γ t is the discount factor, used to weigh the importance of the current and future rewards; is the expected value, the state s generated based on the policy π t and the action a t The trajectory calculation of; r(s t , a t ) is the state s t The action a is executed under t The reward of; α is the entropy regularization coefficient, used to control the importance of entropy; H(π(·|s t ) represents the entropy value. The larger the entropy value, the greater the exploration degree of the agent in the environment, enabling the agent to find a more efficient policy, which helps to accelerate subsequent policy learning.

[0063] Reference Figure 4 As shown, in the proposed SAC framework, it contains an Actor network π θ 、two Critic networks and as well as two target Critic networks and θ, and are the weight parameters of the neural network respectively. The target Critic network Copies parameters from the Critic network in the following way:

[0064] The Critic network Is updated by minimizing the mean square Bellman error, while the Actor network π θ Is updated by minimizing the KL divergence:

[0065]

[0066] Among them, Is the loss function of the Critic network, used to update the Critic parameter Is the Q-value estimate of the i-th Critic network for the state s i and the action a i ; y is the target Q-value; D is the experience replay pool; Represents taking the smaller Q-value among the two Critic networks, reducing the overestimation bias; αlogπ θ Entropy regularization term, encouraging policy exploration; J π (θ) is the optimization objective of the Actor network, maximizing the entropy-adjusted Q-value; a i (s i ) Resamples the action through π θ (s i ). α is updated through the following objective function, where Is the negative value of the action dimension, ρ π Is the state steady-state distribution induced by the strategy:

[0067]

[0068] 2) Reward function design:

[0069] The total reward consists of the following parts with adjustable weights (balanced by hyperparameters λ1, λ2, λ3):

[0070] r t = λ1r force + λ2r torque + λ3r final

[0071] where r force is the contact force constraint, F i is the current three-dimensional contact force component, F max is the maximum allowable force; r torque is the torque constraint, T j is the current three-dimensional torque component, T max is the torque threshold; r final is the terminal reward, d k is the current assembly position, d target is the final assembly target point.

[0072] 3) Training the reinforcement learning algorithm network process, as shown in the reference figure, includes the following:

[0073] Step 1: Initialize the network and the experience pool, and initialize the parameters θ of the Actor network π θ and the parameters of the twin Critic networks and Copy the same parameters to initialize the target networks and and Create an empty experience replay pool D.

[0074] Step 2: Data collection and interaction with the environment, select the action a according to the current policy t = π θ (s t ). Execute the action a t , obtain the reward r t , and the environment state becomes s t+1 . Collect the experience tuple (s t , a t , r t , s t+1 ) and store it in the empty experience replay pool D;

[0075] Step 3: Network parameter update:

[0076] Randomly sample N tuples from the empty experience replay pool D, and calculate the target Q value for each tuple using the target network according to Equation 4. Update the two Critic networks by minimizing the loss function, i.e., Equation 3. Use reparameterization to sample actions, and then update the current Actor network with the minimized loss (Equation 5).

[0077] Then update α according to the policy entropy, according to Soft-update the target network.

[0078] Step 4: Repeat Steps 2 - 3 until the termination condition is met, and then the loop terminates.

[0079] The assembly system provided in this embodiment collects the pose trajectory points of the bent hard pipe and the dynamic data of the contact force / moment in real time. The variable impedance parameter learning network constructed based on the SAC reinforcement learning algorithm analyzes the contact force state in real time, outputs the stiffness parameter and damping parameter matching the current assembly state. The controller generates the end trajectory correction instruction according to the optimized impedance parameters (i.e., the stiffness parameter and damping parameter), and drives the robot to perform compliant adjustment. In this way, through the cyclic iteration of "environment perception - parameter optimization - instruction execution", the impedance parameters are autonomously updated with the change of the contact force, ensuring the dynamic adaptability in complex assembly scenarios and effectively avoiding excessive deformation of flexible sleeves such as foam pipes.

[0080] Embodiment 2

[0081] This embodiment provides a robot flexible sleeve assembly method based on adaptive variable impedance, including the following content:

[0082] The first camera captures the image of the flexible sleeve, and the second camera captures the image of the bent hard pipe. The first camera and the second camera send the captured images to the controller; the controller identifies the pose of the flexible sleeve according to the flexible sleeve image, and captures the pose trajectory of the bent hard pipe according to the bent hard pipe image;

[0083] The controller grasps the flexible sleeve according to the pose of the flexible sleeve;

[0084] The force sensor obtains the magnitude of the contact force when the robot gripper grasps the flexible sleeve and sends it to the controller;

[0085] The controller constructs a variable impedance parameter learning network based on the hard pipe pose trajectory data and the contact force dynamic data. The controller analyzes the contact force state in real time according to the variable impedance parameter learning network and outputs the stiffness parameter and damping parameter matching the current state;

[0086] The controller generates an end trajectory correction instruction according to the output stiffness parameter and damping parameter, drives the work of the robot to perform compliance adjustment, and realizes the installation of the flexible sleeve at the bent hard pipe.

[0087] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A robot flexible sleeve assembly system based on adaptive variable impedance, characterized in that, Comprising: A robot, which is provided with a mechanical gripper for grasping a flexible sleeve; A vision module, including a first camera for collecting images of the flexible sleeve and a second camera for collecting images of the bent rigid tube; A force sensor, which is disposed at the mechanical gripper; A controller, the robot, the first camera, the second camera and the force sensor are separately connected to the controller. The controller identifies the pose of the flexible sleeve according to the image of the flexible sleeve so that the mechanical gripper of the robot can grasp the flexible sleeve, and captures the pose trajectory of the bent rigid tube according to the image of the bent rigid tube. The controller constructs a variable impedance parameter learning network according to the pose trajectory data of the rigid tube and the dynamic contact force data sent by the force sensor. The controller analyzes the contact force state in real time according to the variable impedance parameter learning network, and outputs stiffness parameters and damping parameters matching the current state. The controller generates an end trajectory correction instruction according to the output stiffness parameters and damping parameters, drives the work of the robot to perform compliant adjustment, and realizes the installation of the flexible sleeve at the bent rigid tube.

2. The robotic flexible sleeve assembly system based on adaptive variable impedance according to claim 1, wherein The first camera is installed at the end of the robotic arm of the robot, and the second camera is fixed obliquely above the bent rigid tube.

3. The robotic flexible sleeve assembly system based on adaptive variable impedance according to claim 1, characterized in that, The controller captures the pose trajectory of the bent rigid tube through the following: The controller filters the noise of the obtained image of the bent rigid tube, reconstructs the structure of the bent rigid tube based on the curve skeleton extraction algorithm, generates candidate skeleton nodes distributed discretely inside the bent rigid tube. After spatial filtering and shrinkage optimization, a refined trajectory is fitted, and further a centered skeleton node is generated, and finally the pose trajectory of the bent rigid tube is obtained.

4. A robot flexible sleeve assembly system based on adaptive variable impedance according to claim 3, characterized in that, The controller uses one-dimensional moving least squares method to fit and generate the refined trajectory; the controller combines the cross-section cutting and ellipse fitting technology of the bent rigid tube to reposition the geometric center to generate the centered skeleton node.

5. A robot flexible sleeve assembly system based on adaptive variable impedance according to claim 1, characterized in that, The controller uses a second-order dynamic model to define the dynamic characteristics of the robot. The controller follows the pose trajectory of the bent rigid tube, and according to the contact force sent by the force sensor, calculates the pose correction amount through the impedance model, and superimposes the calculated pose correction amount on the pose trajectory of the bent rigid tube to form a dynamically adjusted one, and then constructs the variable impedance parameter learning network.

6. The robot flexible sleeve assembly system based on adaptive variable impedance according to claim 1, wherein, The controller defines a variable impedance parameter learning network, inputs the current assembly position of the flexible sleeve and the contact force and torque information received by the flexible sleeve during the sleeving process, and defines the dynamic impedance control parameters as the dynamic stiffness parameters and dynamic damping parameters during the sleeving process.

7. The robotic flexible sleeve assembly system based on adaptive variable impedance according to claim 6, wherein The controller dynamically adjusts the stiffness and damping parameters through a reinforcement learning algorithm, including the following: Formula definition and determination of the reinforcement learning algorithm framework; Reward function design; Training process of the reinforcement learning algorithm network, repeating the reward function design and training of the reinforcement learning algorithm until the loop terminates.

8. The robotic flexible sleeve assembly system based on adaptive variable impedance according to claim 7, wherein The formula definition includes the determination of the maximum total reward formula, which is determined by the entropy regularization coefficient, the entropy value, the defined current assembly position of the flexible sleeve and the contact force and torque information received by the flexible sleeve during the sleeving process, and the dynamic impedance control parameters.

9. The robotic flexible sleeve assembly system based on adaptive variable impedance according to claim 7, wherein The reward function is designed through multiple parameter and contact force constraints, torque constraints and terminal rewards.

10. A method for assembling a robot flexible sleeve based on adaptive variable impedance according to any one of claims 1-9, characterized in that, Including the following: The first camera collects images of the flexible sleeve, and the second camera collects images of the bent rigid tube. The first camera and the second camera send the collected images to the controller. The controller identifies the pose of the flexible sleeve based on the flexible sleeve images and captures the pose trajectory of the bent rigid tube based on the bent rigid tube images. The controller grasps the flexible sleeve according to the pose of the flexible sleeve. The force sensor obtains the magnitude of the contact force when the robot manipulator grasps the flexible sleeve and sends it to the controller. The controller constructs a variable impedance parameter learning network based on the rigid tube pose trajectory data and the contact force dynamic data. The controller performs real-time analysis of the contact force state according to the variable impedance parameter learning network and outputs the stiffness parameter and damping parameter matching the current state. The controller generates an end trajectory correction instruction according to the output stiffness parameter and damping parameter, drives the work of the robot to perform compliant adjustment, and realizes the installation of the flexible sleeve at the bent rigid tube.

Images