Method for suppressing residual vibration of robot arm, control device, storage medium, and robot arm component

Abstract

The present invention provides a method for suppressing residual vibration of a robotic arm, a control device, a storage medium, and a robotic arm component. The method for suppressing residual vibration of the robotic arm includes: obtaining the dynamic characteristics of the robotic arm under a predetermined position and orientation and a predetermined load, and constructing a data set in which the dynamic characteristics correspond to the predetermined position and orientation and the predetermined load; training a deep neural network model based on the data set; predicting the target dynamic characteristics of the robotic arm using the trained deep neural network model based on the target position and orientation and the target load of the robotic arm; designing a vibration suppressor based on the target dynamic characteristics, and cooperating with a motion controller to control the movement of the robotic arm to the target position and orientation while suppressing residual vibration. By realizing online real-time prediction of dynamic characteristics at a low computational cost and adaptively designing a vibration suppressor, vibration suppression in an open work scenario can be realized.

Description

Technical Field

[0001] The present invention relates to the field of motion control technology, and particularly to a method for suppressing residual vibration of a robot arm, a control device, a storage medium, and a robot arm component.

Background Art

[0002] Residual vibration is caused by the elasticity of the joints of the robot arm and the structural inertial force, and is a typical step vibration phenomenon of a dynamic system. The vibration frequency is the natural frequency of the system, and the maximum amplitude usually has a positive correlation with the operating acceleration of the robot arm. That is, sudden startup or braking of the robot arm causes large vibrations at the end effector. In scenarios where efficiency is important, residual vibration limits the operating speed of the robot arm, cannot fully utilize the capabilities of the actuator, and wastes a lot of time waiting for the vibration to disappear. In scenarios where high precision is required for advanced task end effectors such as cooperative surgery, overshoot of the position of the end effector due to vibration is not allowed.

[0003] However, most of the current methods for suppressing vibration of robot arms only correspond to a single frequency. Changes in the posture of the robot arm itself and the load at the tip directly affect the inertial distribution of the system. Therefore, the dynamic characteristics of the system change over time, and there is a strong non-linear relationship between the inertial distribution of the robot arm and the posture angles of each joint. It is difficult to accurately represent the mapping relationship between the two in a simple functional form. During the real-time operation of the robot arm, continuously formulating and solving complex dynamic equations for robot arms with different loads will significantly increase the requirements for the hardware of the controller, generate a large amount of computational workload, and the computational time consumed by the posture may vary greatly, which may have an adverse effect on the control of the robot arm. Therefore, in order to solve the above problems, a new method for suppressing vibration of robot arms is urgently required.

Summary of the Invention

[0004] The present invention addresses the drawbacks of related technologies and provides a method for suppressing residual vibration of a robotic arm, a control device, a storage medium, and a robotic arm component for solving the problems in related technologies that vibration suppression is only possible for a single frequency or the adaptability of the frequency corresponding to vibration suppression is low.

[0005] The present invention obtains the dynamic characteristics of the robotic arm under a predetermined position and orientation and a predetermined load, and constructs a data set corresponding to the dynamic characteristics, the predetermined position and orientation, and the predetermined load; trains a deep neural network model based on the data set; predicts the target dynamic characteristics of the robotic arm using the trained deep neural network model based on the target position and orientation and the target load of the robotic arm; designs a vibration suppressor based on the target dynamic characteristics, and cooperates with a motion controller to control the movement of the robotic arm to the target position and orientation while suppressing residual vibration. A method for suppressing vibration of a robotic arm is provided.

[0006] According to the above embodiment, the present invention constructs a data set in which dynamic characteristics correspond to a predetermined position and orientation and a predetermined load, trains a deep neural network model based on the data set, and uses the trained deep neural network model to perform online prediction of the dynamic characteristics of a robot arm system with time-varying characteristics. By combining with a vibration suppressor, the residual vibration at the tip of the robot arm can be adaptively suppressed at a very low computational cost. The present invention replaces the complex process of obtaining the characteristic frequencies of the dynamics of the robot arm using a deep neural network (DNN) model, and realizes the rapid online prediction of the time-varying frequencies of the robot arm. Here, the data set used for training the deep neural network model in the present invention is obtained by simulation analysis of dynamic parameters without relying on a physical prototype. On the other hand, the number of samples in the sample space of the training data set can be determined according to requirements such as the design range of each parameter of the robot arm and the vibration suppression effect, so the operation is simple and easy to implement.

[0007] Furthermore, by combining the design of a pre-trained deep neural network model and a vibration suppressor, the target dynamic characteristics of the robot arm can be quickly predicted based on the target position and orientation and the target load of the robot arm, and the robot arm can be smoothly moved to the target position. The required computational cost is extremely low, and the requirements for rapid online execution of the robot arm can be fully met.

[0008] In one embodiment, the step of obtaining the dynamic characteristics of the robot arm under a predetermined position and orientation and a predetermined load, and constructing a data set in which the dynamic characteristics correspond to the predetermined position and orientation and the predetermined load is as follows: The step of obtaining the dynamic parameters of the robot arm joints; The step of constructing a robot arm dynamics model based on the structural design scheme of the robot arm and the dynamic parameters of the robot arm joints; Analyzing and obtaining the dynamic characteristics of the robot arm under a predetermined position and orientation and a predetermined load based on the robot arm dynamics model, and constructing a dataset in which the dynamic characteristics correspond to the predetermined position and orientation and the predetermined load.

[0009] In one embodiment, the step of obtaining the dynamic parameters of the robot arm joints includes: Obtaining the impulse response curve of the robot arm joints; Processing the impulse response curve to obtain the equivalent stiffness, equivalent damping, and equivalent inertia information of each joint of the robot arm.

[0010] In one embodiment, the step of constructing a robot arm dynamics model based on the dynamic parameters of the robot arm joints includes: Combining the dynamic parameters of the robot arm joints with the three-dimensional model of the robot arm to construct the robot arm dynamics model.

[0011] In one embodiment, the step of analyzing and obtaining the dynamic characteristics of the robot arm under a predetermined position and orientation and a predetermined load based on the robot arm dynamics model, and constructing a dataset in which the dynamic characteristics correspond to the predetermined position and orientation and the predetermined load includes: Constructing an independent variable dataset with the predetermined position and orientation and the predetermined load of the robot arm as independent variables; Using the robot dynamics model to obtain the dynamic characteristics of the robot arm under the conditions corresponding to each combination of independent variables in the independent variable dataset, and constructing a response dataset.

[0012] In one embodiment, the step of training a deep neural network model based on the dataset includes: Training neural networks with different numbers of layers to determine the optimal number of layers; Training the neural network by different training set selection methods to obtain the mapping relationship of the dataset corresponding to the dynamic characteristics, the predetermined position and orientation, and the predetermined load.

[0013] In one embodiment, the step of predicting the target dynamic characteristics of the robotic arm using the trained deep neural network model based on the target position and orientation and the target load of the robotic arm is as follows: Introducing the trained deep neural network model into the control device of the robotic arm, inputting the information of the target position and orientation and the target load of the robotic arm into the trained deep neural network model, and determining the target dynamic characteristics of the robotic arm.

[0014] In one embodiment, the robotic arm dynamics model is a robotic arm simulation dynamics model. Based on the robotic arm dynamics model, the step of analyzing and obtaining the dynamic characteristics of the robotic arm under a predetermined position and orientation and a predetermined load is as follows: Obtaining the actual dynamic characteristics of the robotic arm prototype under a predetermined position and orientation and a predetermined load through experiments. Based on the robotic arm simulation dynamics model, obtaining the simulation dynamic characteristics of the robotic arm in the same state as the experiment under a predetermined position and orientation and a predetermined load. Including the step of modifying the robotic arm simulation dynamics model to determine the dynamic characteristics.

[0015] In one embodiment, the step of obtaining the actual dynamic characteristics of the robotic arm prototype under a predetermined position and orientation and a predetermined load through experiments is as follows: Including the step of determining the natural frequency, mode decay ratio, and vibration mode of the robotic arm prototype under multiple positions and orientations and multiple loads through experiments.

[0016] In one embodiment, the step of modifying the robot arm simulation dynamics model to determine the dynamic characteristics includes comparing the actual dynamic characteristics with the simulated dynamic characteristics, determining sensitive parameters by sensitivity analysis, and modifying the sensitive parameters to determine the dynamic characteristics.

[0017] The present invention further provides a robot arm component including an arm body and a controller. Here, the controller is connected to the arm body, controls the movement of the arm body to a target position and orientation, and implements the vibration suppression method of the robot arm described above. characterized in that

[0018] The present invention further provides a control device including a processor, a memory, and at least one program. Here, the memory is communicably connected to the processor, the at least one program is stored in the memory, and when the at least one program is executed by the processor, the vibration suppression method of the robot arm described above is implemented.

[0019] The present invention further provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a control device, the vibration suppression method of the robot arm described above is implemented.

[0020] Additional aspects and advantages of the present invention will be partially shown in the following description, and these will become apparent from the following description or through the implementation of the present invention.

Brief Description of the Drawings

[0021] The accompanying drawings are incorporated herein and constitute a part of this specification, show embodiments consistent with the present invention, and are used to explain the principles of the present invention together with this specification.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Embodiments for Carrying Out the Invention

[0022] Exemplary embodiments will now be described in detail, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise specified, the same numbers in different drawings refer to the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of apparatuses, devices, and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0023] The terms used in the present invention are for the purpose of describing particular embodiments only and are not intended to limit the present invention. The singular forms "a", "said", and "the" used in the present invention and the appended claims are intended to include the plural forms as well, unless the context clearly indicates otherwise. Also, the term "and / or" as used herein refers to any and all possible combinations of one or more of the associated listed items and is to be understood as inclusive.

[0024] The dynamic model of the residual vibration phenomenon represented by a robotic arm with flexible joints can be simplified into a typical multi-degree-of-freedom mass-spring-damper system in structural dynamics. It has been found that the natural frequency of the system is jointly determined by the spring stiffness, damping, and mass distribution, and the main frequency of the residual vibration is the lower-order natural frequency of the system. According to related research, the speed reducer within the joint is the main factor for the elastic and damping characteristics, and its equivalent stiffness and damping can be regarded as constants. However, the changes in the posture of the robotic arm and the tip load directly affect the inertia distribution of the system, resulting in the time-varying dynamic characteristics of the system. The relationship between the inertia distribution of the robotic arm and the posture angles of each joint has strong non-linearity based on the sequential multiplication of the coordinate transformation matrix. If the posture angles of the joints and the tip load inertia are defined as the domain, and the natural frequency of the robotic arm is defined as the range, it is difficult to express the mapping between the two in a simple functional form. In related research, vibration suppression is carried out by analyzing the response data to extract system characteristics. However, such a method has natural hysteresis, and the application effect only shortens the decay time of the vibration. The suppression of the maximum overshoot is relatively limited and cannot handle delicate operations. As another solution, a model can be introduced for prediction, and based on the prediction results, a controller (feedback or feedforward) can be designed in real time to solve the hysteresis of extracting system characteristics based on the response. However, due to the high hardware cost, it cannot be applied to high-speed process systems, and there is still no better solution. During the real-time operation of the robotic arm, continuously establishing and solving complex dynamic equations for robotic arms with different loads will significantly increase the requirements for the hardware of the controller, generate a large amount of computational workload, and the computational time consumed by the posture may vary greatly, which may have an adverse impact on the control of the robotic arm.

[0025] The present invention provides a method for suppressing residual vibration of a robotic arm and a control device based on a deep neural network, aiming to solve the above technical problems of the related art.

[0026] Hereinafter, a method for suppressing residual vibration of a robot arm based on a deep neural network according to an embodiment of the present invention and a control device will be described in detail with reference to the accompanying drawings. In the following embodiments, the features can be complemented or combined with each other as long as they do not conflict.

[0027] A schematic diagram showing the flow of the method for suppressing residual vibration of a robot arm provided in an embodiment of the present invention is shown in FIG. 1 and includes the following steps S100 to S400.

[0028] In S100, the dynamic characteristics of the robot arm under a predetermined position and posture and a predetermined load are acquired, and a dataset in which the dynamic characteristics correspond to the predetermined position and posture and the predetermined load is constructed.

[0029] The dynamic characteristics in this embodiment are the natural frequency f i , the mode damping ratio ζ i , and the vibration mode φ i . The predetermined position and posture are represented by each joint angle coordinate q j (j = 1,..., 7), the predetermined load is m, and a dataset {(q j , m) | (f i , ζ i , φ i )} in which the dynamic characteristics correspond to the predetermined position and posture and the predetermined load is constructed.

[0030] In some embodiments, the dynamic characteristics may include only the natural frequency f i and the mode damping ratio ζ i . Those skilled in the art can set it flexibly.

[0031] The robot arm and the integrated joints in this embodiment can be applied to, for example, industrial robots, service robots, or special robots. As an example, the robot arm can be applied to a piano-playing robot.

[0032] The degree of freedom of the robotic arm in this embodiment is n, where n is a positive integer. As an example, the degree of freedom of the robotic arm is 7.

[0033] Note that, in this embodiment, the method for obtaining the dynamic characteristics of the robotic arm under a predetermined position and orientation and a predetermined load may use conventional calculation methods such as analysis methods and dynamic modeling, and the present invention does not make specific limitations.

[0034] In S200, a deep neural network model is trained based on the dataset.

[0035] This embodiment uses the dataset {(q j , m)|(f i , ζ i , φ i )} in step S100 to train a deep neural network, constructs a domain {q j , m} with a predetermined position and orientation and a predetermined load as design factors, constructs a value range {f i , ζ i , φ i} with the natural frequency f i , the mode decay ratio ζ i , and the vibration mode φ i as response parameters, and obtains the mapping relationship between the design factors (q j , m) and the dynamic characteristic parameters (f i , ζ i , φ i ). This embodiment can perform online prediction of the dynamic characteristics of a time-varying robotic arm system by means of a pre-trained deep neural network model, avoid the computational cost of solving the dynamics of the robotic arm by online simulation, is more advantageous for the online adaptive design of a vibration suppressor, and can fully meet the needs of the online rapid execution of the robotic arm.

[0036] In S300, based on the target position and orientation and the target load of the robotic arm, the target dynamic characteristics of the robotic arm are predicted using the trained deep neural network model.

[0037] In this embodiment, a trained deep neural network model is introduced into the control device of the robotic arm. After receiving the information on the target position and orientation of the robotic arm and the target load, the information on the target position and orientation of the robotic arm and the target load is input into the trained deep neural network model in order to predict and determine the target dynamic characteristic parameters {f i , ζ i , φ i}.

[0038] In S400, a vibration suppressor is designed based on the target dynamic characteristics, and in cooperation with the motion controller, it controls the movement of the robotic arm to the target position and orientation while suppressing the residual vibration.

[0039] In this embodiment, the vibration suppressor may select an input shaper that performs input shaping on the initial motion control command input to the robotic arm so as to suppress the residual vibration at the tip of the robotic arm during the process of controlling the movement of the robotic arm to the target position and orientation. The input shaper may be designed based on the target dynamic characteristic parameters {f i , ζ i , φ i}. Specifically, the input parameters of each joint are convolved with the corresponding input shaper to obtain the final input parameters of each joint, and in cooperation with the motion controller, the robotic arm is driven to complete the planned motion and smoothly reach the target position, thereby realizing the vibration suppression at the tip.

[0040] In some embodiments, the dynamic characteristics may only include the natural frequency f i , the mode damping ratio ζ i . Those skilled in the art can set them flexibly.

[0041] Note that different vibration suppressors may be used to achieve vibration suppression. In the present invention, an embodiment of the input shaper is shown, but it is not limited thereto. Those skilled in the art can set it flexibly according to the actual situation.

[0042] Furthermore, the methods for suppressing the residual vibration of a robotic arm by control in this field are roughly classified into two types: feedback control and feedforward control. Feedback control is represented by PID (Proportional-Integral-Derivative) control based on an inverse dynamics model, which is easy to implement and robust, but is sensitive to the dynamic parameters of the robotic arm, cannot be applied to variable loads or high-speed motion control, and the suppression effect on the maximum overshoot of residual vibration is also limited. In addition, a delay occurs in the feedback loop, and it is also necessary to equip a feedback controller and a sensor for detecting the position and orientation of the tip, so the complexity of the system and the hardware cost increase.

[0043] Common methods of feedforward control include optimal trajectory planning and input shaping. These methods do not require a feedback sensor, but input control cannot consider the variability of the system either. The trajectory planning of a robotic arm is currently based on a cubic polynomial or cubic B-spline curve interpolation, which generates a trajectory with a sharp change in the derivative value at the start and stop points, inducing vibration at startup and stop. On the other hand, a trajectory based on a NURBS (Non-uniform Rational B-spline) curve can better remove vibration, but the solution of the NURBS curve is complex and the computational amount is large, which is also a common problem in trajectory planning methods using high-order polynomials and other complex curves. Most of such algorithms are currently being studied at the algorithm level, and some of them have poor generality. When considering both vibration suppression and comprehensive optimization such as time and energy consumption, a method considering weight coefficients is generally adopted, but the selection of weight coefficients has a strong subjective element. In addition, the optimal trajectory conflicts in terms of operation time and the maximum output of the driving device.

[0044] Input shaping removes specific frequency components from the system input through feedforward control and suppresses the corresponding frequency components during the response. Achieving the suppression of specific frequency components in the system response through feedforward control essentially means filtering out these frequency components from the frequency characteristics of the system input. Compared with a complete notch filter, the input shaping technique utilizes only a plurality of pulses with specified delays and amplitude ratios to achieve filtering, and has advantages such as simple form, small delay, and no introduction of complex ripples. In the time domain, it has the effect of signal smoothing, and in the frequency domain, it has the same effect as a comb filter. The advantages of input shaping are very suitable for suppressing the residual vibration of a robotic arm. However, since the notch frequency of the pulse sequence is directly determined by the time interval between pulses, the problem of predicting the target frequency must be solved while the natural frequency of the robotic arm varies with time. Therefore, in the present invention, a deep neural network model is introduced into the vibration suppression method of the robotic arm and combined with an input shaper, so that the effect of accurately and quickly suppressing the residual vibration at the tip of the robotic arm can be obtained.

[0045] The present invention constructs a dataset in which dynamic characteristics correspond to a predetermined position and orientation and a predetermined load, trains a deep neural network model based on the dataset, and uses the trained deep neural network model to perform online prediction of the dynamic characteristics of a robotic arm system having time-varying characteristics. By combining with a vibration suppressor, the residual vibration at the tip of the robotic arm is adaptively suppressed at a very low computational cost. The present invention replaces a complex process of obtaining the characteristic frequencies of a robotic arm in dynamics using a deep neural network (DNN) model, and realizes the rapid online prediction of the time-varying frequencies of the robotic arm. Here, the dataset used for training the deep neural network model in the present invention is obtained by simulation and analysis of dynamic parameters without relying on a physical prototype. On the other hand, the number of samples in the sample space of the training dataset can be determined according to requirements such as the design range of each parameter of the robotic arm and the vibration suppression effect, so the operation is simple and the realization is easy.

[0046] Furthermore, by combining the design of a pre-trained deep neural network model and a vibration suppressor, the target dynamic characteristics of the robotic arm can be rapidly predicted based on the target position and orientation and the target load of the robotic arm, and the robotic arm can be smoothly moved to the target position. The required computational cost is extremely low, and the requirement for rapid online execution of the robotic arm can be fully satisfied.

[0047] In some embodiments, step S100 includes the following steps S110 to S130 as shown in FIG. 2.

[0048] In S110, the dynamic parameters of the robotic arm joints are obtained.

[0049] The dynamic parameters in this embodiment include equivalent stiffness, equivalent damping, and equivalent inertia of rotating parts.

[0050] In this embodiment, the equivalent stiffness, equivalent damping of each joint, and equivalent inertia of the rotating parts of robot arms (physical prototypes) of different models are measured through joint-specific dynamics experiments.

[0051] Note that the equivalent stiffness, equivalent damping, and equivalent inertia of the rotating parts in this embodiment are obtained by the measurement and calculation methods of conventional vibration theory, and the present invention does not make specific limitations.

[0052] In S120, based on the structural design scheme of the robot arm and the dynamic parameters of the robot arm joints, a robot arm dynamics model is constructed.

[0053] In this embodiment, based on the dynamic parameters of the robot arm and the three-dimensional model of the robot arm, specifically, a robot arm dynamics model, which is a dynamic simulation model of the robot arm, is constructed. Here, the structural design scheme includes the geometric information of the robot arm. Specifically, the geometric information includes the length of each arm segment between each joint of the robot arm, the relative position and angle of each joint.

[0054] In S130, based on the robot arm dynamics model, the dynamic characteristics of the robot arm under a predetermined position and posture and a predetermined load are analyzed and obtained, and a data set in which the dynamic characteristics correspond to the predetermined position and posture and the predetermined load is constructed.

[0055] In some embodiments, the robot arm dynamics model is constructed by MATLAB (registered trademark) / Simulink (or ADAMS) multi-body system dynamics simulation software. The structural model of each arm segment of the robot arm is introduced by a CAD model, and the dynamic parameters obtained in step 100 are introduced into the original dynamics model of the robot arm, and finally, a robot arm dynamics model with the same or similar end residual vibration behavior as the physical prototype is obtained through simplification.

[0056] In some embodiments, step S110 includes the following steps S111 to S112 as shown in FIG. 3.

[0057] In S111, an impulse response curve of the robot arm joint is obtained.

[0058] In this embodiment, a robot arm with a counterweight at the movable end of the joint is fixedly attached, an impulse input is applied to the joint, and the impulse response curve is obtained by measuring the response parameters of the counterweight.

[0059] In S112, the impulse response curve is processed to obtain the equivalent stiffness, equivalent damping, and equivalent inertia information of each joint of the robot arm.

[0060] In this embodiment, the dynamic parameters of each joint are obtained by processing the response signal of the impulse response curve.

[0061] In some embodiments, step S130 includes the following steps S131 to S132 as shown in FIG. 4.

[0062] In S131, an independent variable data set is constructed with the predetermined position and orientation of the robot arm and the predetermined load as independent variables.

[0063] In this embodiment, for each joint angle coordinate q j (j = 1,..., 7) and the tip load m, an 8-dimensional vector {q 1 ,..., q 7 , m} is constructed as an independent variable vector. Each element is normalized by the upper and lower limits of each angle and the range of the load limit, and a normalized DOE (Design Of Experiments) matrix of size n DOE ×8 is constructed using the Latin Hypercube method to construct the independent variable data set. Here, each row is a combination of random normalized independent variables, and n DOEThe value of must meet the requirements for training the deep neural network model in step S400. Here, each joint angle coordinate corresponds to the generalized displacement of each degree of freedom of the robotic arm, that is, it corresponds to the rotation angle of a revolute joint or the expansion and contraction of a sliding joint.

[0064] In S132, the dynamic characteristics of the robotic arm under the conditions corresponding to each combination of independent variables in the independent variable dataset are obtained using the robot dynamics model, and a response dataset is constructed.

[0065] In this embodiment, the combinations of independent variables are extracted row by row from the DOE matrix in step S310, they are denormalized, and input into the robot arm dynamics model as initial state parameters. A pulse signal of a cubic polynomial quasi-trapezoidal wave with an appropriate magnitude is applied to drive each joint to start and stop rapidly in sequence, and the length of the pulse interval needs to be such that the tip vibration decays to a sufficiently low level. The dynamic transfer function between each joint and the tip vibration response is analyzed for each segment, and the corresponding dynamic characteristics are extracted. A dynamic characteristic parameter matrix of the robotic arm is constructed, and a response dataset is constructed. Here, the dynamic characteristic parameters include {f i , ζ i , φ i}, where i is the mode order that needs to be noted.

[0066] In some embodiments, the dynamic characteristics may only include the natural frequency f i and the mode damping ratio ζ i . Those skilled in the art can set them flexibly.

[0067] In some embodiments, step S200 includes the following steps S210 to S220 as shown in FIG. 5.

[0068] In S210, neural networks with different numbers of layers are trained to determine the optimal number of layers.

[0069] The optimal number of layers in this embodiment is the minimum number of layers while satisfying the accuracy requirements.

[0070] In S220, the neural network is trained by different training set selection methods to obtain the mapping relationship of the dataset corresponding to the dynamic characteristics, a predetermined position and orientation, and a predetermined load.

[0071] In this embodiment, after determining the optimal number of layers in step S210, the neural network is trained by different training set selection methods to obtain the mapping relationship of the dataset corresponding to the dynamic characteristics, a predetermined position and orientation, and a predetermined load. The obtained deep neural network has the ability to predict the dynamic characteristic parameters of the robotic arm based on the combination of independent variables.

[0072] In some embodiments, the robotic arm dynamics model is a robotic arm simulation dynamics model. As shown in FIG. 6, in step S130, the step of analyzing and obtaining the dynamic characteristics of the robotic arm under a predetermined position and orientation and a predetermined load based on the robotic arm dynamics model includes the following steps S1301 to S1303.

[0073] In S1301, through experiments, the actual dynamic characteristics of the robotic arm prototype under a predetermined position and orientation and a predetermined load are obtained.

[0074] In this embodiment, the physical prototype of the robotic arm is tested to obtain the actual dynamic characteristics of the robotic arm prototype. Specifically, the dynamic characteristics of the physical prototype under multiple positions and orientations and multiple loads are tested. The dynamic characteristics of the robotic arm prototype are the natural frequency f i , the mode decay ratio ζ i , and the vibration mode φ iIt includes the low-order mode parameters of the robotic arm. In one example, considering the complex and variable working conditions in the actual operation of the robotic arm, several typical positions and postures are selected and combined with multiple different loads to form test states, and the dynamic characteristics of the robotic arm are tested in each test state.

[0075] In S1302, based on the robotic arm simulation dynamics model, the simulation dynamic characteristics of the robotic arm in the same state as the experiment under a predetermined position and posture and a predetermined load are obtained.

[0076] In this embodiment, in step S1301, using the initialization parameters of each test state, based on the robotic arm dynamics model obtained in step S120, the simulation dynamic characteristics of the robotic arm in the corresponding state are obtained.

[0077] In S1303, the robotic arm simulation dynamics model is modified to determine the dynamic characteristics.

[0078] It should be noted that steps S1301 - S1303 of the present invention may be before step S131, or may be after step S131 and before step S132. The present invention is not particularly limited.

[0079] In this embodiment, the actual dynamic characteristics and the simulation dynamic characteristics are compared, sensitive parameters are determined by sensitivity analysis, and the sensitive parameters are modified to determine the dynamic characteristics. Here, the sensitive parameter is a simulation parameter that greatly affects the deviation between the two as a result of comparing the actual dynamic characteristics and the simulation dynamic characteristics. By modifying the sensitive parameters within a reasonable range, the dynamics model can simulate the dynamic operation of the physical prototype more accurately and can be used for the analysis of the dynamic characteristics of the robotic arm.

[0080] In this embodiment, by combining the physical prototype of the robotic arm with the three-dimensional model, the dynamic model is further corrected. Since the finally obtained dynamic characteristics are closer to the motion behavior of the physical prototype, they have higher accuracy.

[0081] Note that the correction of the dynamic model in the embodiment of the present invention depends on the physical prototype. If there is no physical prototype, this step can be skipped without affecting the implementation of the whole method.

[0082] To facilitate understanding, the present invention shows a specific application scenario. For example, when a piano-playing robot performs piano playing, when the 7-degree-of-freedom bionic arm of the piano-playing robot moves between different keys, its rapid start and stop induce hand vibration. Since the rigidity of the bionic arm is large, this phenomenon is a typical residual vibration problem of a robotic arm with flexible joints. Such a problem can be simplified as a typical vibration phenomenon formed by a multi-degree-of-freedom mass-spring-damper system under step excitation in structural dynamics. In actual applications, the requirement for the vibration state of the tip of the robotic arm is usually that the end effector such as the hand or tool can reach the target position stably and stop without causing large overshoot or vibration.

[0083] To address the above problems, it is necessary to consider two aspects. The first is the analysis of the dynamic behavior of the robotic arm system. The frequency of the step vibration of the system, which is the natural frequency of the system, is co-determined by the spring stiffness, damping, and mass distribution. According to the research in relevant literature, the speed reducer within the joint is the main factor for elastic and damping characteristics, and its equivalent stiffness and damping can be regarded as constants. However, the changes in the posture of the robotic arm and the tip load directly affect the inertia distribution of the system, resulting in the time-varying vibration characteristics of the system. Moreover, the non-linearity of the relationship between the inertia distribution and the posture angles of each joint is strong, and it is difficult to express the mapping from the joint posture angles and the tip load to the natural frequency of the robotic arm in a simple functional form. The complexity of the mapping relationship is one of the difficulties in suppressing residual vibration. The second is that when a fast and stable stop is required after the tip reaches the target position, it is obvious that the introduction of feed-forward control is a more suitable method, and other vibration suppression methods based on feedback control or response measurement cannot avoid hysteresis. Combining these two, a residual vibration suppression method with a certain degree of generality needs to have the ability to more accurately predict the time-varying natural frequency of the robotic arm. Therefore, the present invention can solve the problems arising from the above application scenarios by providing a vibration suppression method for the robotic arm that can more accurately predict the time-varying natural frequency of the robotic arm.

[0084] The robot arm simulation model of the piano-playing robot in the above application scenario is shown in Fig. 7. The robot arm simulation model in Fig. 7 is a 7-degree-of-freedom robot arm, which is composed of a first joint 1, a second joint 2, a third joint 3, a fourth joint 4, a fifth joint 5, a sixth joint 6, and a seventh joint 7 that are rotationally connected in sequence. In a certain posture, trapezoidal wave pulses are sequentially applied to the seven joints (keeping the magnitude of jerk the same during acceleration and deceleration), and the response spectrum of the tip vibration is shown in Fig. 8. It can be seen from the response spectrum that there are at least two main vibration frequencies in the arm. The embodiment of the present invention can solve the problems caused by the residual vibration at the tip of the robot arm of the piano-playing robot and can meet the high requirements for the posture adjustment time of the piano-playing task.

[0085] As an example, the residual vibration of the robot arm is tested according to the above application scenario. As shown in Fig. 9, it is a comparison diagram of the test results of the robot arm dynamics model before and after vibration suppression according to this embodiment. Comparing before and after vibration suppression, among 1000 groups of test cases, the average reduction rate is about 82.5%, the total vibration amplitude is reduced by about 86.8%, and the amplitude reduction rate is over 95% in 107 groups, 90% - 95% in 273 groups, and 80% - 90% in 377 groups. It can be seen that the vibration suppression method of the robot arm provided by the embodiment of the present invention can greatly improve the vibration suppression effect.

[0086] Based on the same inventive concept, the present invention further provides a robot arm component including an arm body and a controller. Here, the controller is connected to the arm body, controls the movement of the arm body to the target position and posture, and implements the above-mentioned vibration suppression method of the robot arm.

[0087] Based on the same inventive concept, the present invention further provides a control device 500 including a processor 510 as shown in FIG. 10, a memory 520, and at least one program 530. Here, the memory 520 is communicably connected to the processor 510, the at least one program 530 is stored in the memory and executed by the processor 510, and the at least one program 530 is arranged to implement the vibration suppression method of the robot arm described above.

[0088] Based on the same inventive concept, the present invention further provides a computer-readable storage medium storing a computer program, where when the computer program is executed by a control device, the vibration suppression method of the robot arm described above is implemented.

[0089] The foregoing embodiments of the present invention can complement each other as long as there is no contradiction.

[0090] As can be understood by those skilled in the art, the steps, means, and manners in the various operations, methods, and flows already discussed in the present invention may be alternated, changed, combined, or deleted. Further, other steps, means, and manners having the same features as those in the various operations, methods, and flows already discussed in the present invention may also be alternated, changed, rearranged, decomposed, combined, or deleted. Further, the steps, means, and manners in the related technologies having the same features as those in the various operations, methods, and flows disclosed in the present application may also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0091] In the description of the present invention, terms such as "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate the orientation or positional relationship based on that shown in the accompanying drawings, and are merely for facilitating the description of the present invention and simplifying the description, and do not indicate or imply that the mentioned device or element must be configured and operated in a specific orientation. Therefore, it should be understood that it is not to be construed as a limitation of the description of the present invention.

[0092] The terms "first" and "second" are used only for descriptive purposes and are not to be understood as indicating or implying relative importance, nor do they implicitly specify the number of the indicated technical features. Therefore, the features defined by the terms "first" and "second" may possibly include one or more such features explicitly or implicitly. In the description of the present invention, unless otherwise specified, "a plurality" means two or more.

[0093] In the description of the present invention, unless specifically specified and limited otherwise, the terms "mounted", "connected", and "coupled" should be broadly understood as, for example, fixed connection, detachable connection, integral connection, direct connection, indirect connection through an intermediate medium, or connection inside two elements. A person skilled in the art can understand the specific meaning of the above terms in the present invention based on specific cases.

[0094] In the description of this specification, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0095] It should be understood that although the individual steps in the flowchart of the accompanying drawings are sequentially shown as indicated by the arrows, the steps are not necessarily executed sequentially in the order indicated by the arrows. Unless otherwise specified in this specification, there is no strict order restriction on the execution of these steps, and they may be executed in other orders. Furthermore, at least some of the steps in the flowchart of the accompanying drawings may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed until completion at the same time, and may be executed at different times, and the execution order thereof is not necessarily sequential, and may be executed in sequence or alternately with at least some of other steps or sub-steps or stages of other steps.

[0096] The above is only a part of the embodiments of the present invention, and it should be noted that those skilled in the art can make a plurality of improvements and modifications without departing from the principles of the present invention, and these are also considered to be within the protection scope of the present invention.

Claims

1. Obtaining the dynamic characteristics of a robot arm under a predetermined position and orientation and a predetermined load, and constructing a data set in which the dynamic characteristics correspond to the predetermined position and orientation and the predetermined load; Training a deep neural network model based on the data set; Predicting the target dynamic characteristics of the robot arm using the trained deep neural network model based on the target position and orientation and the target load of the robot arm; Designing a vibration suppressor based on the target dynamic characteristics and cooperating with a motion controller to control the movement of the robot arm to the target position and orientation while suppressing residual vibration. A method for suppressing vibration of a robot arm, characterized by the above.

2. The step of obtaining the dynamic characteristics of a robot arm under a predetermined position and orientation and a predetermined load, and constructing a data set in which the dynamic characteristics correspond to the predetermined position and orientation and the predetermined load includes: Obtaining the dynamic parameters of the robot arm joints; Constructing a robot arm dynamics model based on the structural design scheme of the robot arm and the dynamic parameters of the robot arm joints; Analyzing and obtaining the dynamic characteristics of the robot arm under a predetermined position and orientation and a predetermined load based on the robot arm dynamics model, and constructing a data set in which the dynamic characteristics correspond to the predetermined position and orientation and the predetermined load. The method for suppressing vibration of a robot arm according to claim 1, characterized by the above.

3. The step of obtaining the dynamic parameters of the robot arm joints includes: Obtaining the impulse response curve of the robot arm joints; Processing the impulse response curve to obtain the equivalent stiffness, equivalent damping, and equivalent inertia information of each joint of the robot arm. The method for suppressing vibration of a robot arm according to claim 2, characterized by the above.

4. The step of constructing a robot arm dynamics model based on the dynamic parameters of the robot arm joints includes: Combining the dynamic parameters of the robot arm joints with the three-dimensional model of the robot arm to construct the robot arm dynamics model. The method for suppressing vibration of a robot arm according to claim 2, characterized by the above.

5. Based on the robot arm dynamics model, analyze and obtain the dynamics characteristics of the robot arm under a predetermined position and orientation and a predetermined load, and construct a data set in which the dynamics characteristics correspond to the predetermined position and orientation and the predetermined load. The steps are as follows: Construct an independent variable data set with the predetermined position and orientation and the predetermined load of the robot arm as independent variables. Use the robot dynamics model to obtain the dynamics characteristics of the robot arm under the conditions corresponding to each combination of independent variables in the independent variable data set, and construct a response data set. The steps include: The method for suppressing vibration of a robot arm according to claim 2, characterized in that.

6. The step of training a deep neural network model based on the data set includes: Training neural networks with different numbers of layers to determine the optimal number of layers. Training a neural network by different training set selection methods to obtain a mapping relationship of the data set in which the dynamics characteristics correspond to the predetermined position and orientation and the predetermined load. The steps include: The method for suppressing vibration of a robot arm according to claim 1, characterized in that.

7. Based on the target position and orientation and the target load of the robot arm, the step of predicting the target dynamics characteristics of the robot arm using the trained deep neural network model includes: Introduce the trained deep neural network model into the control device of the robot arm, input the information of the target position and orientation and the target load of the robot arm into the trained deep neural network model, and determine the target dynamics characteristics of the robot arm. The steps include: The method for suppressing vibration of a robot arm according to claim 1, characterized in that.

8. The robot arm dynamics model is a robot arm simulation dynamics model. Based on the robot arm dynamics model, the step of analyzing and obtaining the dynamics characteristics of the robot arm under a predetermined position and orientation and a predetermined load includes: Obtain the actual dynamics characteristics of the robot arm prototype under a predetermined position and orientation and a predetermined load through experiments. Based on the robot arm simulation dynamics model, obtaining the simulation dynamics characteristics of the robot arm in the same state as the experiment under a predetermined position and orientation and a predetermined load; modifying the robot arm simulation dynamics model to determine the dynamics characteristics; The vibration suppression method of the robot arm according to claim 2, characterized in that.

9. The step of obtaining the actual dynamics characteristics of the robot arm prototype under a predetermined position and orientation and a predetermined load by experiment is The step of determining the natural frequency, mode decay ratio, and vibration mode of the robot arm prototype under a plurality of positions and orientations and a plurality of loads by experiment, including The vibration suppression method of the robot arm according to claim 8, characterized in that.

10. The step of modifying the robot arm simulation dynamics model to determine the dynamics characteristics is comparing the actual dynamics characteristics with the simulation dynamics characteristics, determining sensitive parameters by sensitivity analysis, and modifying the sensitive parameters to determine the dynamics characteristics. The vibration suppression method of the robot arm according to claim 1, characterized in that.

11. An arm body; A controller connected to the arm body, controlling the movement of the arm body to a target position and orientation, and implementing the vibration suppression method of the robot arm according to any one of claims 1 to 10. A robot arm component, characterized in that.

12. A processor; A memory communicably connected to the processor; At least one program stored in the memory, which, when executed by the processor, implements the vibration suppression method of the robot arm according to any one of claims 1 to 10. A control device, characterized in that.

13. A computer-readable storage medium storing a computer program, which, when executed by a control device, implements the vibration suppression method of the robot arm according to any one of claims 1 to 10. A computer-readable storage medium, characterized in that.

Images