Humanoid robot joint motor overload protection method, device and equipment

Abstract

The application provides a kind of humanoid robot joint motor overload protection method, comprising the following steps: obtaining the current of joint motor;Linear processing is carried out to the current of joint motor, and first current processing result is obtained;The first current processing result is predicted by current prediction module, and current prediction value is obtained;Current prediction value is compared with the overload threshold of the current of joint motor, and residual value is calculated;Residual value is processed by compensation value calculation module, and current compensation value is obtained;The control amount of current loop of joint motor is compensated according to current compensation value, the prediction value based on prior state of kalman filter can be used to predict the change of current in advance, so that the system can respond to the possible current overload in advance;At the same time, the currents of multiple motors are integrated using multilayer perceptron, which can associate the judgment of one current with other currents.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robot control, in particular to a human-robot joint motor overload protection method, device and equipment. BACKGROUND

[0002] In recent years, with the rapid development of robot technology, robots need to work in human-friendly environments or natural environments in the wild, and need to work in uncertain environments and uncertain tasks, such as walking on uneven ground, walking on complex road conditions, and completing a series of tasks such as screwing, carrying, and even rescue after reaching the destination. Because the model cannot completely simulate the situation of uncertain environment, the human-robot will face the unexpected situation of motor overload during action planning and execution.

[0003] Some existing protection schemes (such as dynamic load monitoring) may not respond to sudden overload in real time, thereby reducing the protection effect to some extent. Software protection and dynamic load monitoring rely on real-time measurement and calculation of joint state, and measurement error and calculation error may affect the accuracy of overload protection. And the conservative scheme may need to limit the performance of the motor, such as reducing torque and speed, which may cause the motion performance of the robot to decline. SUMMARY

[0004] The purpose of the present application is to provide a human-robot joint motor overload protection method, device and equipment, which realizes effective protection of the joint motor of the robot.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme: a human-robot joint motor overload protection method, comprising the following steps: obtaining the current of the joint motor; linearly processing the current of the joint motor to obtain a first current processing result; predicting the first current processing result through a current prediction module to obtain a current prediction value; comparing the current prediction value with an overload threshold of the joint motor current to calculate a residual value; processing the residual value through a compensation value calculation module to obtain a current compensation value; and compensating the control amount of the current loop of the joint motor according to the current compensation value.

[0006] Further, the linear processing of the current of the joint motor comprises:

[0007] training a first multi-layer perception machine;

[0008] inputting the joint motor current into the first multi-layer perception machine;

[0009] processing the joint motor current through the first multi-layer perception machine without using an activation function, and outputting a first current processing result.

[0010] Further, the current prediction module comprises a Kalman filter, a system state equation of the Kalman filter is:

[0011] x k =φ k x k-1 +w k (12)

[0012] An observation equation of the Kalman filter is:

[0013] y k =h k x k +v k (13)

[0014] Wherein, x k is a current value of the joint motor at k, w k is a system Gaussian noise at k, v k is an observed Gaussian noise at k, φ k is a linear transfer change ratio of the current value x k of the joint, h k is a proportional error when measuring the current value of the joint motor at k, y k is an observed value of the joint motor current at k.

[0015] Further, a state updating step of the Kalman filter is as follows:

[0016] x k- =φ k-1 x (k-1)+ (14)

[0017]

[0018]

[0019] p k+ =(1-K k h k )p k- (17)

[0020] x k+ =x k- +K k (y k -h k x k- ) (18)

[0021] I″=h k x k+ (19)

[0022] wherein p k- is a prior estimate of the system variance, p k+ is a posterior estimate of the system variance, a linear transfer variation of the current value x (k-1 , p (k-1) is a posterior estimate of the system variance at k-1, K k is a Kalman filter coefficient at k, and I" is the current prediction value.

[0023] Further, the current prediction value is compared with an overload threshold of the joint motor current by the following formula:

[0024] I"'= sgn(I") |I max -I" | (20)

[0025] wherein I"'is a residual value, I max is the overload threshold of the joint motor current.

[0026] Further, the processing of the residual value by the compensation value calculation module comprises:

[0027] training a second multi-layer perception;

[0028] inputting the residual value into the first multi-layer perception;

[0029] processing the residual value by the second multi-layer perception using an activation function and outputting a residual value processing result;

[0030] setting a current correction coefficient, and obtaining the current compensation value by the current correction coefficient and the residual value processing result.

[0031] Further, the formula for the first multi-layer perception processing the joint motor current is:

[0032]

[0033] wherein I1-I n is the joint motor current, and I'1-I' is the first current processing result. n

[0034] Further, the activation function of the second multi-layer perception is a tanh activation function, and the formula for the second multi-layer perception obtaining the current compensation value by the current correction coefficient and the residual value is:

[0035]

[0036] ​Wherein, P is a current correction constant, tanh is a function of calculating the corresponding hyperbolic tangent value of each element, e1-e n is a current compensation value.

[0037] In another aspect, a human-imitating robot joint motor overload protection device is provided, comprising: a current acquisition module, acquiring the current of a joint motor; a first processing module, linearly processing the current of the joint motor to obtain a first current processing result; a first prediction module, predicting the first current processing result through a current prediction module to obtain a current prediction value; a comparison module, comparing the current prediction value with an overload threshold of the current of the joint motor to calculate a residual value; a residual value processing module, processing the residual value through a compensation value calculation module to obtain a current compensation value; and a compensation module, compensating the control quantity of the current loop of the joint motor according to the current compensation value.

[0038] In another aspect, an electronic device is provided, comprising: a processor and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the human-imitating robot joint motor overload protection method described above.

[0039] It can be analyzed that the human-imitating robot joint motor overload protection method, device and equipment disclosed in the present application can predict the change of the current in advance through the prediction value of the Kalman filter based on the prior state, so that the system can respond to the possible current overload in advance; at the same time, the current of multiple motors is comprehensively processed using a multi-layer perception machine, so that the judgment of one current is associated with other currents. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings, which form a part of the present application, are included to provide a further understanding of the application, and are incorporated herein for purposes of explanation and are not intended to limit the application. In the drawings:

[0041] Figure 1 Flow chart of an embodiment of the present application. DETAILED DESCRIPTION

[0042] The present application will be described in detail below with reference to the accompanying drawings and embodiments. Each example is provided by way of explanation of the present application rather than limiting the present application. In fact, those skilled in the art will appreciate that modifications and variations can be made in the present application without departing from the scope or spirit of the present application. For example, features shown or described as part of one embodiment can be used in another embodiment to produce yet another embodiment. Therefore, it is intended that the present application encompass such modifications and variations as fall within the scope of the appended claims and their equivalents.

[0043] One or more examples of the present application are illustrated in the accompanying drawings. The detailed description uses numerical and letter designations to refer to features in the drawings. Like or similar designations in the drawings and description have been used to refer to like or similar parts of the application. As used herein, the terms "first", "second", "third", and "fourth" and the like can be used interchangeably to distinguish one component from another and are not intended to signify location or importance of the individual components.

[0044] As shown in the drawings, according to an embodiment of the present application, a human-robot joint motor overload protection method is provided, comprising the following steps: Figure 1

[0045] Step S101, obtaining joint motor current;

[0046] Step S102, linearly processing the current of the joint motor to obtain a first current processing result;

[0047] The above linear processing of the current of the joint motor includes: training a first multi-layer perception machine; inputting the joint motor current into the first multi-layer perception machine; processing the joint motor current through the first multi-layer perception machine without using an activation function, and outputting the first current processing result.

[0048] Specifically, the formula for the first multi-layer perception machine to process the joint motor current is:

[0049]

[0050] wherein, I1-I n is the joint motor current, and I′1-I′ n is the first current processing result.

[0051] Step S103, predicting the first current processing result through a current prediction module to obtain a current prediction value;

[0052] The current prediction module includes a Kalman filter, and the system state equation of the Kalman filter is:

[0053] x k = φ k x k-1 + w k (24)

[0054] The observation equation of the Kalman filter is:

[0055] y k = h k x k + v k (25)

[0056] wherein, x k ​is the current value of the joint motor at time k, w k is the system Gaussian noise at time k, when is the current value of the joint motor at time k, w k is Gaussian distributed, v k is the observed Gaussian noise at time k, when is Gaussian distributed, v k is Gaussian distributed, φ k is the current value of the joint motor x k is the proportion of linear transfer variation of the current value of the joint motor x k is the proportional error when measuring the current value of the joint motor at time k, φ k and h k are both 1, y k is the observed value of the current of the joint motor at time k, which is the actual measured value.

[0057] Preferably, the state update step of the Kalman filter is as follows:

[0058] x k- = φ k-1 x (k-1)+ (26)

[0059]

[0060]

[0061] p k+ = (1 - K k h k ) p k- (29)

[0062] x k+ = x k - + K k (y k - h k x k- ) (30)

[0063] I″= h k x k+ (31)

[0064] where p k- is the prior estimate of the system variance, p k+ is the posterior estimate of the system variance, is the proportion of linear transfer variation of the current value of the joint motor x (k-1) , p (k-1) is the posterior estimate of the system variance at time k-1, K k is the Kalman filter coefficient at time k, which is different for each input and can change or not change at each time k. It is assumed that the initial state initial posterior x 0+ = x0, p 0+ = p0, then I(k) is assumed to be the corresponding current observation value y k I" is the current prediction value.

[0065] The statistical posterior value of the current value of the joint motor is estimated by formula (26) above. Formula (27) realizes the prior estimation of the variance value of the system. Formula (28) realizes the calculation of the Kalman filter coefficient. Formula (29) realizes the posterior estimation of the variance value of the system. Formula (30) realizes the calculation of the current statistical posterior value.

[0066] Step S104, compare the current prediction value with the overload threshold of the current of the joint motor, and calculate the residual value;

[0067] The current prediction value and the overload threshold of the current of the joint motor are compared by the following formula:

[0068] I"'= sgn(I") |I max -I" | (32)

[0069] Where I"'is the residual value, I max is the current overload threshold of the joint motor, and sgn is a sign function that returns 1 if >= 0 and -1 if < 0.

[0070] Step S105, process the residual value through the compensation value calculation module to obtain the current compensation value;

[0071] The processing of the residual value through the compensation value calculation module includes: training a second multilayer perceptron; inputting the residual value into the first multilayer perceptron; using an activation function and processing the residual value through the second multilayer perceptron, and outputting the residual value processing result; setting a current correction constant, and obtaining the current compensation value through the current correction constant and the residual value processing result.

[0072] The second multilayer perceptron is trained using a supervised learning paradigm, that is, there is a series of correct answers of "input-output", and then if the output of the second multilayer perceptron reasoning does not conform to the correct answer, the error will be back propagated to update the weight parameters. Repeat this back propagation process to complete the training.

[0073] The activation function of the second multilayer perceptron is a tanh activation function, and the formula for the second multilayer perceptron to obtain the current compensation value through the current correction constant and the residual value is:

[0074]

[0075] wherein P is a current correction constant, tanh is a function for calculating a hyperbolic tangent value corresponding to each element, e 1- e n is a current compensation value.

[0076] In step S106, the control quantity of the current loop of the joint motor is compensated according to the current compensation value.

[0077] The control quantity of the current loop of the joint motor is compensated by the current compensation value, so that the motor avoids overload during operation.

[0078] The application further discloses a human-simulating robot joint motor overload protection device, which comprises a current acquisition module, a first processing module, a first prediction module, a comparison module and a compensation value calculation module.

[0079] The application further discloses an electronic device, which comprises a processor and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the human-simulating robot joint motor overload protection method.

[0080] From the above description, it can be seen that the above-mentioned embodiments of the application achieve the following technical effects: the shallow multi-layer perceptron and Kalman filter used in the application have high calculation efficiency; through the prediction value of the Kalman filter based on the prior state, the change of the current can be predicted in advance, so that the system can respond to the possible current overload in advance; and the multi-layer perceptron is used to comprehensively process the currents of multiple motors, so that the judgment of one current is associated with other currents.

[0081] The above only describes the preferred embodiments of the application and is not used to limit the application. For those skilled in the art, the application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for overload protection of a humanoid robot joint motor, characterized in that, The method comprises the following steps: obtaining the current of the joint motor; linearly processing the current of the joint motor to obtain a first current processing result; predicting the first current processing result through a current prediction module to obtain a current prediction value; comparing the current prediction value with an overload threshold of the current of the joint motor to calculate a residual value; processing the residual value through a compensation value calculation module to obtain a current compensation value; compensating the control amount of the current loop of the joint motor according to the current compensation value; The current prediction module comprises a Kalman filter, and the system state equation of the Kalman filter is: x k = φ k x k-1 + w k (1) The observation equation of the Kalman filter is: y k = h k x k + v k (2) wherein x k is the current value of the joint motor at time k, w k is the system Gaussian noise at time k, v k is the observed Gaussian noise at time k, φ k is the linear transfer variation proportion of the current value x k of the joint, h k is the proportional error when measuring the current value of the joint motor at time k, y k is the observed value of the joint motor current at time k; The state updating step of the Kalman filter is as follows: x k- = φ k-1 x (k-1)+ (3) p k+ = (1 - K k h k )p k- (6) x k+ = x k- + K k (y k - h k x k- ) (7) I" = h k x k+ (8) wherein p k- is a prior estimate of the system variance, p k+ is a posterior estimate of the system variance, a linear transfer change of the current value x (k-1) , p (k-1) is a posterior estimate of the system variance at k-1, K k is a Kalman filter coefficient at k, and I" is the current prediction value; The current prediction value and the overload threshold of the current of the joint motor are compared through the following formula: I'''= sgn (I' ') |Imax-I'' | (9) where I'" is the residual value, I max is the current overload threshold for the joint motor; The processing of the residual value through the compensation value calculation module comprises: training a second multilayer perceptron; inputting the residual value into the first multilayer perceptron; processing the residual value through the second multilayer perceptron using an activation function and outputting a residual value processing result; setting a current correction coefficient, and obtaining the current compensation value through the current correction coefficient and the processing result of the residual value.

2. The human-imitative robot joint motor overload protection method according to claim 1, characterized in that, The linear processing of the current of the joint motor comprises: training a first multilayer perceptron; inputting the current of the joint motor into the first multilayer perceptron; processing the current of the joint motor through the first multilayer perceptron without using an activation function and outputting a first current processing result.

3. The method of claim 1, wherein the method further comprises: The formula for the first multilayer perceptron to process the current of the joint motor is: I1-I n I1-I n I1-I 4. The method of claim 3, wherein the method further comprises: The activation function of the second multilayer perceptron is a tanh activation function, and the formula for the second multilayer perceptron to obtain the current compensation value through the current correction coefficient and the residual value is: where P is a current correction constant, tanh is a function for calculating a hyperbolic tangent value of each element, e1-e n is a current compensation value.

5. A human-robot joint motor overload protection device, the overload protection device is configured to perform the overload protection method of any one of claims 1-4, characterized in that, comprise: a current acquisition module that acquires the current of the joint motor; a first processing module that linearly processes the current of the joint motor to obtain a first current processing result; a first prediction module that predicts the first current processing result through a current prediction module to obtain a current prediction value; a comparison module that compares the current prediction value with an overload threshold of the current of the joint motor to calculate a residual value; a residual value processing module that processes the residual value through a compensation value calculation module to obtain a current compensation value; a compensation module that compensates the control amount of the current loop of the joint motor according to the current compensation value.

6. An electronic device, comprising: The electronic device comprises a processor and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the joint motor overload protection method of any one of claims 1-4.

Images