Robot shaft hole assembling method based on demonstration learning

Abstract

The invention discloses a demonstration learning-based robot shaft hole assembly method, which comprises the following steps of 1, demonstrating an insertion action for multiple times in a dragging demonstration mode, and simultaneously collecting the stress of a shaft and the posture deviation of the shaft hole in a demonstration process; 2, obtaining a mapping relation between the stress of the shaft and the attitude deviation of the shaft hole through Gaussian mixture regression; 3, enabling the tail end of the robot to approach the estimated position of the hole through admittance control; fourthly, constant force control is conducted in the inclination direction of the shaft and the vertical direction of the plane of the hole part to complete the hole searching stage; and 5, detecting the attitude deviation of the shaft hole in real time, and obtaining expected forces in all directions according to Gaussian mixture regression, thereby performing force control on the shaft to complete the insertion stage.

Description

[0001] The present invention relates to the field of robot imitation learning technology, and in particular to a robot shaft hole assembly method based on demonstration learning. Background Art

[0002] Assembly is a key link in industrial production, accounting for 50% and 30% of the total time and total cost of the entire production cycle, respectively. Due to the particularity of the environment and the strict requirements on quality, assembly is a labor-intensive process with high labor intensity and low work efficiency. Robots are regarded as an ideal solution for assembly operations, but currently assembly robots face many challenges, especially in some cases with small gaps, often with low assembly success rate, poor efficiency, and difficulty in adapting to environmental changes. These problems limit the widespread application of assembly robots. Therefore, it is of great significance to study how to achieve precision assembly in uncertain environments.

[0003] The current assembly strategies can be roughly divided into two categories, passive compliance and active compliance. The typical representative of passive compliance is RCC, i.e. Remote Compliance Center. Although RCC is easy to use, it has poor versatility. Often, an RCC device can only be used on one type of part to be assembled, which is difficult to meet the changing operation requirements.

[0004] Active compliance control assembly requires the installation of force, vision and other sensors, and then estimates the current state based on the feedback information, adjusts the installation action and completes the assembly. As mentioned above, these active compliance controls require a lot of time to build complex geometric and mechanical models, and uncertainties such as friction and geometric errors will greatly affect the performance of the model. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some embodiments. The purpose of the abstract and the invention title in this section and the present application is not to be ambiguous, and such simplification or omission cannot be used to limit the scope of the present invention.

[0006] In view of the above problems, the present invention proposes a robot shaft-hole assembly method based on demonstration learning. The demonstrator demonstrates the insertion action multiple times by dragging and teaching, and collects the force and posture information of the shaft in the process, and obtains the mapping relationship between the force of the shaft and the shaft-hole posture deviation through Gaussian mixture model and Gaussian mixture regression. Then, the robot end is made to reach the estimated position of the hole through admittance control, and then the hole-finding stage is completed by constant force control of the shaft in the appropriate direction. Finally, the shaft is force-controlled according to the shaft-hole posture deviation and Gaussian mixture regression model to complete the insertion stage.

[0007] The technical solution adopted by the present invention to solve its technical problems is: a robot shaft-hole assembly method based on demonstration learning, comprising the following steps: step one: demonstrating the insertion action multiple times by dragging and teaching, and collecting the shaft force and shaft-hole posture deviation during the demonstration; step two: obtaining the mapping relationship between the shaft force and the shaft-hole posture deviation by Gaussian mixture regression; step three: making the robot end close to the estimated position of the hole by admittance control; step four: performing constant force control in the inclination direction of the shaft and the vertical direction of the hole part plane to complete the hole-finding stage; step five: detecting the shaft-hole posture deviation in real time, and obtaining the expected force in each direction according to Gaussian mixture regression, so as to perform force control on the shaft to complete the insertion stage.

[0008] Furthermore, step 1 includes: the demonstrator drags the robot in the drag teaching mode to demonstrate the insertion action for multiple times, and collects the posture and the end force. According to the posture, we can obtain the posture deviation θ. Then, the force F in the tilt direction of the axis s Force F in the direction perpendicular to the plane of the hole part v , and then combine the two and θ into a two-dimensional vector and Where N is the number of demonstrations.

[0009] Furthermore, step 2 includes: converting the demonstration data ξ of step 1 s and v It is constructed as a Gaussian mixture model (GMM) containing K Gaussian basis functions.

[0010]

[0011] The model parameter solution uses the k-means algorithm to obtain the initial value of the model parameter, and iterates through the expectation maximization algorithm (EM) until convergence to establish the specific parameters of the Gaussian mixture model. Then, the mapping relationship between the inclined force of the axis and the vertical force of the hole part plane and the axis-hole posture deviation is obtained by Gaussian mixture regression (GMR).

[0012] Furthermore, step three includes: controlling the robot end to reach the estimated position of the hole, using admittance control in the process to avoid rigid contact between the shaft hole to damage the robot arm, and detecting the end force in real time. When the force reaches a certain threshold, it is determined that the shaft hole has been in contact, and the approaching stage ends.

[0013] Furthermore, step four includes: performing constant force control on the tilting direction of the robot axis and the vertical direction of the hole part plane, and judging that the hole-finding phase ends when the forces in the tilting direction of the axis and the vertical direction of the hole part plane reach a threshold and the end position of the robot no longer changes.

[0014] Furthermore, step five includes: real-time detection of the error posture of the shaft hole, and then obtaining the expected force in the tilt direction of the shaft and the vertical direction of the hole part plane through the Gaussian mixture regression model in step one, so as to control the force value of the robot in the tilt direction of the shaft and the vertical direction of the hole part plane, and when the insertion depth reaches a certain threshold, it is judged that the insertion is successful.

[0015] The beneficial effect of the present invention is that a robot shaft hole assembly method based on demonstration learning is provided, which avoids complex physical modeling, thereby enabling the robot to master shaft hole assembly skills directly from demonstration. At the same time, the method has high robustness to positioning errors and nail hole gap changes, and has certain practical significance for solving robot shaft hole assembly problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] Figure 1 It is a flow chart of the present invention;

[0018] Figure 2 It is a diagram of the dragging teaching process of the present invention;

[0019] Figure 3 This is a mapping curve between the shaft hole inclination angle and the forces in the x-direction and the z-direction obtained by the present invention through Gaussian mixture regression.

[0020] Figure 4 It is the force control direction diagram of the hole-finding stage of the present invention;

[0021] Figure 5 It is a force data diagram during the assembly process of the present invention; DETAILED DESCRIPTION

[0022] Reference Figure 1 A robot shaft-hole assembly method based on demonstration learning includes the following steps: Step 1: Demonstrate the insertion action multiple times by dragging and teaching, and collect the shaft force and shaft-hole posture deviation during the demonstration; Step 2: Obtain the mapping relationship between the shaft force and the shaft-hole posture deviation by Gaussian mixture regression; Step 3: Make the robot end approach the estimated position of the hole by admittance control; Step 4: Perform constant force control in the inclination direction of the shaft and the vertical direction of the hole part plane to complete the hole-finding stage; Step 5: Detect the shaft-hole posture deviation in real time, and obtain the expected force in each direction according to Gaussian mixture regression, so as to perform force control on the shaft to complete the insertion stage.

[0023] Reference Figure 2 , step 1 includes: the research object of the present invention is the problem of shaft hole assembly by robot, there is a cylinder clamp at the end of the robot, the clamp grabs the shaft, the demonstrator drags the robot to demonstrate the insertion action for many times in the drag teaching mode, and collects the posture and the end force. According to the posture, we can get the posture deviation θ. In this experiment, the shaft is only tilted around y at a certain angle, so it is necessary to collect the x-direction force F x and the force F in the z direction z , and then combine the two and θ into a two-dimensional vector and Where N is the number of demonstrations.

[0024] Reference Figure 3 , step 2 includes: converting the demonstration data ξ of step 1 x and z It is constructed as a Gaussian mixture model (GMM) containing K Gaussian basis functions.

[0025]

[0026] The model parameter solution uses the k-means algorithm to obtain the initial value of the model parameter, and iterates through the expectation maximization algorithm (EM) until convergence to establish the specific parameters of the Gaussian mixture model. Then, the mapping relationship between the inclined force of the axis and the vertical force of the hole part plane and the axis-hole posture deviation is obtained by Gaussian mixture regression (GMR).

[0027] Step three includes: controlling the robot end to reach the estimated position of the hole, using admittance control in the process to avoid rigid contact between the shaft and the hole to damage the robot arm, and detecting the end force in real time. When the force reaches a certain threshold, it is determined that the shaft and the hole are in contact, and the approaching stage ends.

[0028] Reference Figure 4 ,Step 4 includes: by controlling the robot with constant force in the x-direction and the z-direction, the axis will be sucked into the hole under the action of the combined force, and the hole-finding phase is judged to be over when the forces in the x-direction and the z-direction both reach the threshold and the end position of the robot no longer changes.

[0029] Step 5 includes: real-time detection of the error posture of the shaft hole, and then obtaining the expected force in the x-direction and z-direction through the Gaussian mixture regression model in step 1, so as to control the robot's force in the x-direction and z-direction. When the insertion depth reaches a certain threshold, the insertion is judged to be successful. The force of the shaft during the entire assembly process is referenced Figure 5 .

[0030] The above is only a specific embodiment of the present invention, and cannot be used to limit the scope of the invention. Therefore, the replacement of equivalent components, or equivalent changes and modifications made according to the scope of protection of the patent of the present invention, should still fall within the scope of this patent. In addition, the technical features of the present invention can be freely combined with each other, with each other, and with each other.

Claims

1. A robot shaft hole assembly method based on demonstration learning, characterized in that: The method comprises the following steps: step 1: demonstrating the insertion action for multiple times by dragging and teaching, and collecting the force on the shaft and the shaft-hole posture deviation during the demonstration; step 2: obtaining the mapping relationship between the force on the shaft and the shaft-hole posture deviation by Gaussian mixture regression; step 3: making the end of the robot close to the estimated position of the hole by admittance control; step 4: performing constant force control in the inclination direction of the shaft and the vertical direction of the plane of the hole part to complete the hole-finding stage; step 5: detecting the shaft-hole posture deviation in real time, and obtaining the expected force in each direction according to Gaussian mixture regression, thereby performing force control on the shaft to complete the insertion stage.

2. The robot shaft hole assembly method based on demonstration learning according to claim 1 is characterized in that: Step 1 includes: the demonstrator drags the robot in the drag teaching mode to demonstrate the insertion action for multiple times, and collects the posture and end force. According to the posture, we can get the posture deviation θ. In this experiment, the axis is only tilted around the y direction at a certain angle, so it is necessary to collect the x-direction force F x and the force F in the z direction z , and then combine the two and θ into a two-dimensional vector and Where N is the number of demonstrations.

3. The robot shaft hole assembly method based on demonstration learning according to claim 1 is characterized in that: Step 2 includes: converting the demonstration data ξ of step 1 x and z It is constructed as a Gaussian mixture model (GMM) containing K Gaussian basis functions. The model parameter solution uses the k-means algorithm to obtain the initial value of the model parameter, and iterates through the expectation maximization algorithm (EM) until convergence to establish the specific parameters of the Gaussian mixture model. Then, the mapping relationship between the inclined force of the axis and the vertical force of the hole part plane and the axis-hole posture deviation is obtained by Gaussian mixture regression (GMR).

4. The robot shaft hole assembly method based on demonstration learning according to claim 1 is characterized in that: Step three includes: controlling the robot end to reach the estimated position of the hole, using admittance control in the process to avoid rigid contact between the shaft and the hole to damage the robot arm, and detecting the end force in real time. When the force reaches a certain threshold, it is determined that the shaft and the hole are in contact, and the approaching stage ends.

5. The robot shaft hole assembly method based on demonstration learning according to claim 1 is characterized in that: Step 4 includes: by controlling the robot with constant force in the x-direction and the z-direction, the axis will be sucked into the hole under the action of the combined force, and the hole-finding phase is judged to be over when the forces in the x-direction and the z-direction both reach the threshold and the end position of the robot no longer changes.

6. The robot shaft hole assembly method based on demonstration learning according to claim 1 is characterized in that: Step five includes: detecting the error posture of the shaft hole in real time, and then obtaining the expected force in the x-direction and z-direction through the Gaussian mixture regression model in step one, so as to control the force value of the robot in the x-direction and z-direction. When the insertion depth reaches a certain threshold, the insertion is judged to be successful.

Images