A method and device for establishing a robot flexible collision detection model

Abstract

The present disclosure provides a method and device for establishing a robot flexible collision detection model, wherein the method comprises: establishing a robot dynamics model, identifying dynamics parameters in the robot motion process using the robot dynamics model; separately establishing a neural network model for compensating joint friction torque in the dynamics model at each joint of the robot, saving information of the neural network model after training and testing the neural network model, optimizing the dynamics model through the neural network model to obtain an optimized dynamics model; establishing a flexible contact force model according to environmental stiffness, indentation depth, contact surface geometry and hysteresis damping factor; combining the optimized dynamics model with a generalized momentum expression to perform formula derivation, designing a sliding mode momentum observer, estimating joint torque of the flexible contact force model using the sliding mode momentum observer, and obtaining a final robot flexible collision detection model.

Description

TECHNICAL FIELD

[0001] The present document relates to the technical field of computer technology, and particularly relates to a method and device for establishing a robot flexible collision detection model. BACKGROUND

[0002] With the rapid development of science and technology, robots are increasingly widely applied in medical education, industrial production and other fields, but in many cases, there is a possibility of collision between the robot and the surrounding environment, which brings certain safety hazards, so robot collision detection is needed, that is, the force change when the robot contacts the environment is analyzed.

[0003] In the related art, since the non-dynamics model needs to install sensors on the robot body, resulting in a huge increase in cost, the dynamics model without the aid of sensors is more widely used, and the dynamics model usually uses a generalized momentum observer to judge the difference between the actual input torque and the reference input torque, thereby obtaining the collision detection result. The above technical solution based on the dynamics model has the following defects: the external joint torque is highly sensitive to the friction parameters in the motor, and the friction parameters have complex and nonlinear characteristics, resulting in inaccurate establishment of the friction model and the dynamic model.

[0004] Based on the above analysis of the development status of the technical field, there is a need for a dynamics model that can handle the nonlinearity of the friction parameters in the existing technical solutions. SUMMARY

[0005] The purpose of the present application is to provide a method and device for establishing a robot flexible collision detection model, which aims to solve the above problems in the prior art.

[0006] According to a first aspect of the embodiments of the present disclosure, a method for establishing a robot flexible collision detection model is provided, comprising:

[0007] establishing a robot dynamics model, using the robot dynamics model to identify the dynamics parameters in the robot motion process;

[0008] establishing a neural network model for compensating the joint friction torque in the dynamics model at each joint of the robot, after training and testing the neural network model, saving the information of the neural network model, optimizing the dynamics model through the neural network model, and obtaining an optimized dynamics model;

[0009] establishing a flexible contact force model according to the environmental stiffness, the indentation depth, the contact surface geometry and the hysteresis damping factor, wherein the hysteresis damping factor is used to represent the influence of different materials on the contact force;

[0010] The formula derivation is combined with the optimized dynamics model and the generalized momentum expression, a sliding mode momentum observer is designed, the joint torque of the flexible contact force model is estimated by using the sliding mode momentum observer, and a final robot flexible collision detection model is obtained.

[0011] According to a second aspect of the embodiments of the present disclosure, a robot flexible collision detection model establishment device is provided, comprising:

[0012] The dynamics model establishment module is configured to establish a robot dynamics model, and identify dynamics parameters in a robot motion process by using the robot dynamics model.

[0013] The joint friction torque compensation module is configured to separately establish a neural network model for compensating the joint friction torque in the dynamics model at each joint of the robot, save information of the neural network model after training and testing the neural network model, and optimize the dynamics model by using the neural network model to obtain an optimized dynamics model.

[0014] The flexible contact force model establishment module is configured to establish a flexible contact force model according to an environmental stiffness, an indentation depth, a contact surface geometry, and a hysteresis damping factor, wherein the hysteresis damping factor is used to represent an influence of different materials on the contact force.

[0015] The sliding mode momentum observer generation module is configured to combine the optimized dynamics model with the formula derivation of the generalized momentum expression, design a sliding mode momentum observer, estimate the joint torque of the flexible contact force model by using the sliding mode momentum observer, and obtain a final robot flexible collision detection model.

[0016] The technical solutions provided by the embodiments of the present disclosure can have the following beneficial effects: the non-linear characteristics of the joint friction torque are captured by using the neural network, the dynamics model can more accurately represent the force conditions of the robot, and the sliding mode momentum observer is designed to replace the sensor to improve the accuracy of the estimated joint torque.

[0017] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of one or more embodiments of the present disclosure or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor.

[0019] Figure 1is a flow chart of a method for establishing a robot flexible collision detection model according to an embodiment of the present application;

[0020] Figure 2 is a schematic diagram of a neural network model according to an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of torque estimation of a flexible contact force model according to an embodiment of the present application;

[0022] Figure 4 is a schematic diagram of torque estimation of a sliding mode momentum observer according to an embodiment of the present application;

[0023] Figure 5 is a schematic diagram of a process for establishing a robot flexible collision detection model according to an embodiment of the present application;

[0024] Figure 6 is a schematic diagram of an apparatus for establishing a robot flexible collision detection model according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the personnel in the technical field better understand the technical solutions in one or more embodiments of the present specification, the technical solutions in one or more embodiments of the present specification will be described clearly and completely in conjunction with the drawings in one or more embodiments of the present specification. Obviously, the described embodiments are only a part of the embodiments of the present specification, not all. Based on one or more embodiments of the present specification, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present document.

[0026] Method embodiments

[0027] According to an embodiment of the present application, a method for establishing a robot flexible collision detection model is provided, Figure 1 is a flow chart of a method for establishing a robot flexible collision detection model according to an embodiment of the present application, as Figure 1 shown, the method for establishing a robot flexible collision detection model according to an embodiment of the present application specifically includes:

[0028] In step S110, a robot dynamics model is established, and the robot dynamics model is used to identify the dynamics parameters in the robot motion process. Specifically, it includes:

[0029] The robot dynamics model is established by using Newton-Lagrange method, wherein the robot dynamics model contains the relationship among joint driving torque, rigid body torque and joint friction torque, and formula 1 represents the dynamics model established by using Newton-Lagrange method:

[0030]

[0031] wherein, τ denotes the joint driving torque, denotes the rigid body torque τ b , τ f denotes the joint friction torque, wherein, denotes the inertia matrix, q denotes the joint rotation angle, denotes the joint velocity, denotes the joint acceleration, denotes the centrifugal force and Coriolis force matrix, G(q) denotes the gravity torque, denotes the joint friction torque vector, is a real number set.

[0032] In step S110, a linear regression equation of a robot dynamic model is established, an excitation trajectory is designed using a Fourier series to obtain a trajectory point position, and a controller is used to issue a control point position to the robot to obtain motion data of the robot in a motion process. The motion data is brought into the linear regression equation based on a weighted least squares method, and the dynamics parameters in the motion process of the robot are identified through the linear regression equation.

[0033] In step S120, a neural network model of the joint friction torque in the compensation dynamics model is separately established at each joint of the robot. After training and testing the neural network model, information of the neural network model is saved, the dynamics model is optimized through the neural network model, and an optimized dynamics model is obtained. Specifically, it includes:

[0034] The joint driving torque is the sum of the rigid body torque and the joint friction torque. Therefore, the joint friction torque can be calculated by subtracting the joint driving torque from the rigid body torque in the case where the joint driving torque and the rigid body torque are known. The joint friction torque is calculated through formula 2:

[0035]

[0036] wherein, are respectively the identification value of the joint motion data and the joint driving torque τ, the joint friction torque τ f is calculated, and the joint friction torque τ f is taken as the target output of the neural network.

[0037] A single joint modeling method is used to fit the joint friction torque of a single joint, that is, a deep neural network is separately established for each joint of the robot. In this embodiment, a neural network model containing three input neurons and one output neuron is established at each joint of the robot. The mapping relationship from the joint motion data to the joint friction torque is obtained through the neural network model, wherein the input neurons respectively represent the rotation angle, velocity and acceleration of the joint i, that is, The output neuron is a neuron representing the i joint friction torque τ f,i . Figure 2 is a schematic diagram of the neural network model of an embodiment of the present application, as Figure 2 shown, showing the structure of the neural network model used to fit the joint friction torque, the connection parameters of each neural layer are updated by the back propagation algorithm, when the loss function of the training set and the test set converges to the minimum value, stop training and save the information of the neural network model at this time.

[0038] In step S130, a flexible contact force model is established according to the environmental stiffness, the indentation depth, the contact surface geometry and the hysteresis damping factor, wherein the hysteresis damping factor is used to represent the influence of different materials on the contact force, formula 3 represents the original robot contact force model, formula 4 represents the expression of the hysteresis damping factor, that is, the energy dissipation problem in the contact process is considered, and the final form of the flexible contact force model is obtained by combining formula 3 and formula 4, that is, formula 5:

[0039] F e =kδ n Formula 3;

[0040]

[0041]

[0042] In the formula, F e represents the contact force between the robot and the environment, k represents the environmental stiffness, n represents the topological index of the contact surface geometry, δ represents the indentation depth caused by the contact deformation between the robot and the external environment, that is, δ=X e -X m , wherein X e represents the position information of the external environment, and X m represents the actual position information of the robot.

[0043] γ represents the hysteresis damping factor, represents the deformation velocity, and Cr represents the material recovery coefficient, represents the initial contact relative velocity, a, b represent constants in the interval [0, 1], d represents a constant less than 0, g represents a constant in the interval [0, 0.5], wherein the hysteresis damping factor γ is related to the environmental stiffness, the material recovery coefficient and the initial contact relative velocity, and is set between 0-1, the hysteresis damping factor of the rigid material is usually close to 1, the hysteresis damping factor of the flexible material is usually less than 1, indicating that the kinetic energy loss after collision is large, the material recovery coefficient Cr refers to the ratio of the relative velocity after the collision of two objects to the relative velocity before the collision, reflecting the recovery ability of the deformation of the object, and is usually between 0-1, the recovery coefficient of the rigid material is usually close to 1, indicating that the kinetic energy loss after collision is very small, and the recovery coefficient of the flexible material is usually less than 1, indicating that the kinetic energy loss after collision is large.

[0044] In the embodiment, the simulation results of the contact force between the robot and the environment in the flexible contact force model are obtained by using MATLAB / Simulink, Figure 3 is a schematic diagram of torque estimation of the flexible contact force model of the embodiment of the application, as Figure 3 shown, showing the change curve of the robot collision simulation external force.

[0045] In step S140, formula derivation is performed in combination with the optimized dynamics model and the generalized momentum expression, a sliding mode momentum observer is designed, the joint torque of the flexible contact force model is estimated using the sliding mode momentum observer, and a final robot flexible collision detection model is obtained. Specifically, it includes:

[0046] A first-order momentum dynamics equation is established according to the optimized dynamics model and the generalized momentum expression, and the first-order momentum dynamics equation is further optimized according to the momentum deviation to obtain a linear first-order low-pass filter, and the expression of the sliding mode momentum observer is obtained through the linear first-order low-pass filter. Formula 6 represents the generalized momentum expression and its differential expression, in combination with the optimized dynamics model, as formula 1, but the joint friction torque has been compensated and optimized by the neural network model, to obtain the first-order momentum dynamics equation represented by formula 7, formula 8 represents a linear first-order low-pass filter obtained by using the generalized momentum torque estimation method for the first-order momentum dynamics equation, and the external collision torque is estimated according to the proportional multiplier of the momentum deviation, and formula 9 represents the expression of the sliding mode momentum observer, so that the sliding mode control has the characteristics of strong robustness and fast convergence:

[0047]

[0048]

[0049]

[0050]

[0051] In the formula, p and respectively represent generalized momentum and its differential, τ ext represents the external moment generated when colliding with the contact, represents the external moment of the robot r, that is, K∈Rn×1>0 is a vector gain, and the estimation convergence speed can be improved by increasing the gain K. High gain will reduce the noise resistance of the estimation, and at the same time cannot cope with nonlinear changes, and it is difficult to achieve a good balance between speed and accuracy, assuming τ ext has a known Lipschitz constant L, the signal is not smooth but its derivative almost exists everywhere, and sgn() represents the sign function, represents σ represents the estimation of τ ext in a limited time, and the derivative of σ is where S1 and S2 represent positive definite diagonal matrices.

[0052] The joint torque estimation in the Cartesian space of the flexible contact force model is carried out by the sliding mode momentum observer, the joint torques estimated at different time points are connected, the change of the joint torque in the robot motion process is obtained, that is, the change of the torque of the end joint in the six-degree-of-freedom robot arm, and in the embodiment, Adam / Simulink joint simulation is adopted, Figure 4 is a schematic diagram of the torque estimation of the sliding mode momentum observer of the embodiment of the present application, as Figure 4 shown, is the simulation result of the torque estimation of the sliding mode momentum observer.

[0053] The method further comprises: judging the flexible collision of the robot through the final robot flexible collision detection model.

[0054] In summary, the above technical scheme of the embodiment of the present application proposes a method for establishing a robot flexible collision detection model, adopts a neural network to capture the nonlinear characteristics of the joint friction torque, so that the dynamics model can more accurately represent the stress of the robot, establishes a flexible contact force model to accurately represent the change of the joint torque when the robot and the external environment have flexible contact, and designs a sliding mode momentum observer to replace the sensor to improve the accuracy of the estimated joint torque.

[0055] The above technical scheme of the embodiment of the present application is exemplified in combination with the following drawings.

[0056] Figure 5 is a schematic diagram of the process of establishing the robot flexible collision detection model of the embodiment of the present application, as Figure 5 shown, the complete process of establishing the robot flexible collision detection model is shown.

[0057] Device embodiment

[0058] According to the embodiment of the present application, a robot flexible collision detection model establishment device is provided, Figure 6 is a schematic diagram of the robot flexible collision detection model establishment device of the embodiment of the present application, as Figure 6 shown, the robot flexible collision detection model establishment device according to the embodiment of the present application specifically comprises:

[0059] The dynamics model establishment module 60 is configured to establish a robot dynamics model, and identify dynamics parameters in the robot motion process by using the robot dynamics model. Specifically, the dynamics model establishment module 60 is configured to:

[0060] The robot dynamics model is established by using the Newton-Lagrange method, which contains the relationship among the joint driving torque, the rigid body torque and the joint friction torque; a linear regression equation of the robot dynamics model is established, a Fourier series is used to design an excitation trajectory to obtain a trajectory point, and a controller is used to issue a control point to the robot to obtain motion data of the robot in the motion process; the motion data is brought into the linear regression equation based on the weighted least squares method, and the dynamics parameters in the robot motion process are identified by using the linear regression equation.

[0061] The joint friction torque compensation module 62 is configured to separately establish a neural network model of the joint friction torque in the compensation dynamics model at each joint of the robot. After the neural network model is trained and tested, the information of the neural network model is saved, the dynamics model is optimized by using the neural network model, and an optimized dynamics model is obtained. Specifically, the joint friction torque compensation module 62 is configured to:

[0062] In the embodiment, a neural network model containing three input neurons and one output neuron is established at each joint of the robot, and a mapping relationship from the joint motion data to the joint friction torque is obtained by using the neural network model, wherein the input neurons respectively represent the rotation angle, the speed and the acceleration of the joint, and the output neuron represents the joint friction torque.

[0063] The flexible contact force model establishment module 64 is configured to establish a flexible contact force model according to the environmental stiffness, the indentation depth, the contact surface geometry and the hysteresis damping factor, wherein the hysteresis damping factor is used to represent the influence of different materials on the contact force, the hysteresis damping factor is related to the environmental stiffness, the material recovery coefficient and the initial contact relative speed, and is set to be between 0 and 1.

[0064] The sliding mode momentum observer generation module 66 is configured to perform formula derivation by combining the optimized dynamics model and the generalized momentum expression, design a sliding mode momentum observer, estimate the joint torque of the flexible contact force model by using the sliding mode momentum observer, and obtain the final robot flexible collision detection model. Specifically, the sliding mode momentum observer generation module 66 is configured to:

[0065] According to the optimization of the kinetic model and the generalized momentum expression, a first-order momentum dynamic equation is established, and the first-order momentum dynamic equation is further optimized according to the momentum deviation, so as to obtain a linear first-order low-pass filter, and the expression of the sliding mode momentum observer is obtained through the linear first-order low-pass filter; the joint torque estimation in the Cartesian space of the flexible contact force model is carried out through the sliding mode momentum observer, and the joint torque estimated at different time points is connected to obtain the change of the joint torque in the robot motion process.

[0066] The device further comprises:

[0067] The flexible collision detection module 68 is used for judging the flexible collision of the robot through the final robot flexible collision detection model.

[0068] In summary, the above technical scheme of the embodiment of the application proposes a robot flexible collision detection model establishment device, adopts a neural network to capture the nonlinear characteristics of the joint friction torque, so that the dynamic model can more accurately represent the stress condition of the robot, the flexible contact force model is established to accurately represent the change of the joint torque when the robot and the external environment are in flexible contact, and the sliding mode momentum observer is designed to replace the sensor to improve the accuracy of the estimated joint torque.

[0069] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, but not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A method for establishing a robot flexible collision detection model, characterized in that, The method comprises the steps of: establishing a robot dynamics model, identifying dynamics parameters in the robot motion process by using the robot dynamics model; establishing a neural network model for compensating joint friction torque in the dynamics model at each joint of the robot, saving information of the neural network model after training and testing the neural network model, optimizing the dynamics model by using the neural network model, and obtaining an optimized dynamics model; establishing a flexible contact force model according to environmental stiffness, indentation depth, contact surface geometry and hysteresis damping factor, wherein the hysteresis damping factor is used to represent the influence of different materials on contact force; deriving a formula by combining the optimized dynamics model and a generalized momentum expression, designing a sliding mode momentum observer, estimating joint torque of the flexible contact force model by using the sliding mode momentum observer, and obtaining a final robot flexible collision detection model; specifically comprising: establishing a first-order momentum dynamics equation according to the optimized dynamics model and the generalized momentum expression, further optimizing the first-order momentum dynamics equation according to momentum deviation, obtaining a linear first-order low-pass filter, and obtaining an expression of the sliding mode momentum observer through the linear first-order low-pass filter; estimating joint torque in Cartesian space of the flexible contact force model by using the sliding mode momentum observer, connecting the joint torque estimated at different time points, and obtaining the change of joint torque in the robot motion process.

2. The method of claim 1, wherein, The method further comprises judging the situation of the robot flexible collision by using the final robot flexible collision detection model.

3. The method of claim 1, wherein, The method of establishing a robot dynamics model and identifying dynamics parameters in the robot motion process specifically comprises: establishing the robot dynamics model by using Newton-Lagrange method, wherein the robot dynamics model contains the relationship among joint driving torque, rigid body torque and joint friction torque; establishing a linear regression equation of the robot dynamics model, using Fourier series to design an excitation trajectory to obtain trajectory points, issuing control points to the robot through a controller to obtain motion data of the robot in the motion process, bringing the motion data into the linear regression equation based on the weighted least squares method, and identifying the dynamics parameters in the robot motion process by using the linear regression equation.

4. The method of claim 1, wherein, The hysteresis damping factor is related to environmental stiffness, material recovery coefficient and initial contact relative speed, and is set to be between 0 and 1.

5. An apparatus for establishing a robot flexible collision detection model, characterized by, The method comprises the steps of: a dynamics model establishing module, configured to establish a robot dynamics model and identify dynamics parameters in the robot motion process by using the robot dynamics model; a joint friction torque compensation module, configured to establish a neural network model for compensating joint friction torque in the dynamics model at each joint of the robot, save information of the neural network model after training and testing the neural network model, optimize the dynamics model by using the neural network model, and obtain an optimized dynamics model; The flexible contact force model establishing module is configured to establish a flexible contact force model according to an environmental stiffness, an indentation depth, a contact surface geometry, and a hysteresis damping factor, wherein the hysteresis damping factor is used to represent an influence of different materials on a contact force; The sliding mode momentum observer generating module is configured to derive a sliding mode momentum observer by combining the optimized dynamic model and a generalized momentum expression, and to estimate joint torques of the flexible contact force model by using the sliding mode momentum observer to obtain a final robot flexible collision detection model.

6. The apparatus of claim 5, wherein, The device further includes: The flexible collision detection module is configured to determine a flexible collision of the robot by using the final robot flexible collision detection model.

7. The device of claim 5, wherein The dynamic model establishing module is specifically configured to establish the robot dynamic model by using a Newton-Lagrange method, wherein the robot dynamic model includes relationships among joint driving torques, rigid body torques, and joint friction torques; to establish a linear regression equation of the robot dynamic model; to obtain trajectory points by using a Fourier series to design an excitation trajectory; to obtain motion data of the robot in a motion process by issuing control points to the robot through a controller; and to identify dynamic parameters in the motion process of the robot by using a weighted least square method to bring the motion data into the linear regression equation.

8. The apparatus of claim 5, wherein, The hysteresis damping factor is related to an environmental stiffness, a material recovery coefficient, and an initial contact relative speed, and is set to be between 0 and 1.

Images