JP2022128680A - Method for adjusting operation parameter and robot system - Google Patents

Abstract

To provide a technique for adjusting an operation parameter with consideration given to the influence degree of the operation parameter on a performance index value.SOLUTION: This method includes the steps of: (a) causing a robot to execute an operation by using a plurality of operation conditions and calculating a performance index value by using the measuring result of a currently operated sensor; (b) constructing a first auto-regressive model with a plurality of operation parameters set as inputs and the performance index value set as an output; (c) calculating the influence degree of each operation parameter on the performance index value by using the first auto-regressive model; (d) selecting one or more to be adjusted from among the operation parameters in accordance with the influence degree; (e) constructing a second auto-regressive model with the operation parameters to be adjusted as inputs and the performance index value set as an output; and (f) adjusting the parameters to be adjusted so as to improve the performance index value by using the second auto-regressive model.SELECTED DRAWING: Figure 3

Description

The present disclosure relates to methods for adjusting operating parameters and robotic systems.

Japanese Patent Laid-Open No. 2002-200001 describes an automatic blur adjustment device for a robot that automatically adjusts "shake", which is one of the performance index values of the robot, by optimizing the operating parameters of the robot.

WO2016/125204

Individual operating parameters of the robot have different effects on performance index values. However, in the prior art, since optimization is not performed with consideration given to such influences, there is a problem that it may not be possible to adjust operating parameters so as to obtain sufficiently good performance index values.

SUMMARY According to a first aspect of the present disclosure, a method is provided for adjusting multiple operational parameters for a robotic system having a robot and sensors used to determine performance indicators of the robot. This method includes (a) causing the robot to perform an action using a plurality of operating conditions obtained by combining candidate values for the plurality of operating parameters, and calculating the performance index value using the measurement result of the sensor during the action. and (b) using the performance index values obtained in step (a), constructing a first regression model having the parameter values of the plurality of operating parameters as inputs and the performance index values as outputs. (c) using the first regression model to determine the degree of influence of each operating parameter on the performance index value; and (d) selecting one or more of the plurality of operating parameters according to the degree of influence. and (e) using the performance index value obtained in the step (a), a second step of inputting the parameter value of the adjustment target parameter and outputting the performance index value and (f) using the second regression model to adjust the parameter to be adjusted such that the performance index value is improved.

According to a second aspect of the present disclosure, there is provided a robot system that includes a robot, sensors used to obtain a performance index value of the robot, and a controller that controls the robot. (a) causing the robot to perform an action using a plurality of operating conditions obtained by combining candidate values for the plurality of operating parameters; and (b) constructing a first regression model having the parameter values of the plurality of operating parameters as inputs and the performance index values as outputs, using the performance index values obtained in the process (a). (c) using the first regression model to determine the degree of influence of each operating parameter on the performance index value; (d) selecting one of the plurality of operating parameters according to the degree of influence; and (e) using the performance index value obtained in the processing (a), the parameter value of the adjustment target parameter is input and the performance index value is output. A process of constructing a two-regression model; and (f) a process of adjusting the parameter to be adjusted such that the performance index value is improved using the second regression model.

1 is an explanatory diagram of a configuration example of a robot system according to an embodiment; FIG. FIG. 2 is a functional block diagram of an information processing device; 4 is a flowchart showing a procedure for adjusting operating parameters; Explanatory drawing which shows the example of operation|movement for adjustment. 4 is a flowchart showing a detailed procedure of step S10; Explanatory drawing which shows an example of a 1st regression model. Graph showing the degree of influence of operating parameters on takt time. Graph showing changes in tact time when operating parameters are changed. Explanatory drawing which shows an example of a 2nd regression model. Explanatory drawing which shows the search range of step S70, S80. The flowchart which shows the procedure of step S50 in 2nd Embodiment.

A. 1. First Embodiment FIG. 1 is an explanatory diagram showing an example of a robot system according to a first embodiment. This robot system includes a robot 100 , a control device 200 that controls the robot 100 , and an information processing device 300 . The robot 100 is installed on a pedestal 500 . The information processing device 300 is, for example, a personal computer.

The robot 100 has a base 110 and a robot arm 120 . An end effector 150 is attached to the tip of the robot arm 120 . A vibration sensor 410 is installed at the tip of the robot arm 120 . In the present disclosure, “hand end of robot arm 120 ” means a portion from the tip of robot arm 120 to end effector 150 . Note that the vibration sensor 410 may be installed at a position other than the tip of the robot arm 120 . For example, a vibration sensor may be installed on the gantry 500 when the vibration of the gantry 500 serves as a performance index value of the robot system.

The robot arm 120 is sequentially connected by four joints J1 to J4. A "joint" is also called an "axis". A TCP (Tool Center Point) as a control point of the robot 100 is set near the tip of the robot arm 120 . A “control point” is a reference point for controlling the robot arm 120 . TCP can be set at any position. Controlling the robot 100 means controlling the position and orientation of this TCP. In this embodiment, a four-axis robot in which the robot arm 120 has four joints J1 to J4 is exemplified, but a robot having an arbitrary arm mechanism having one or more joints can be used. Further, the robot 100 of this embodiment is a horizontal articulated robot, but a vertical articulated robot may be used.

Vibration sensor 410 is a sensor capable of detecting vibration of robot 100 . In general, vibration can be detected by measuring time-series changes in any of position, velocity, and acceleration. As the vibration sensor 410, for example, a position sensor, a gyro sensor, an acceleration sensor, an inertial measurement unit, etc. can be used.

FIG. 2 is a block diagram showing functions of the information processing apparatus 300. As shown in FIG. The information processing apparatus 300 has a processor 310 , a memory 320 , an interface circuit 330 , an input device 340 and a display section 350 connected to the interface circuit 330 . The interface circuit 330 is also connected to the controller 200 . Measurement results of the sensors 400 of the robot 100 are supplied to the information processing device 300 via the control device 200 .

The sensors 400 include, in addition to the vibration sensor 410 described with reference to FIG. I'm in. Position sensor 430 is, for example, an encoder. At least some of these sensors 400 are used when obtaining performance index values used in the optimization process of operation parameters, which will be described later. However, some of these sensors 400 can be omitted. The sensors 400 correspond to the "sensor" of the present disclosure.

The processor 310 includes a performance index value calculation unit 312 that calculates a performance index value from the measurement results of the sensors 400, a regression model construction unit 314 that constructs a regression model, a parameter selection unit 316 that selects parameters to be adjusted, and an adjustment Each function of the parameter adjustment unit 318 that performs adjustment regarding the target parameter is executed. These units 312 , 314 , 316 , 318 are implemented by processor 310 executing a computer program stored in memory 320 . However, some or all of these units 312, 314, 316, and 318 may be realized by hardware circuits. Processor 310 corresponds to the “controller” of the present disclosure.

The memory 320 stores various data and files as follows.
(1) Parameter file PF
The parameter file PF contains initial values for a plurality of operating parameters for the robot system. Further, when the adjustment of the operating parameters is completed, the optimized operating parameters after adjustment are stored.
(2) Operation program RP
The operation program RP is composed of a plurality of instructions for operating the robot 100 . The process of obtaining performance index values using a plurality of operating conditions is executed by operating the robot 100 using this operating program RP.
(3) Operation log file ML
The operation log file ML is a file in which operation conditions and performance index values used when the performance index value calculation unit 312 executes processing for obtaining performance index values are recorded.
(4) First regression model RM1
The first regression model RM1 is a regression model used when the parameter selection unit 316 calculates the degree of influence of performance index values on operating parameters. The first regression model RM1 receives parameter values of a plurality of operating parameters as inputs and outputs performance index values. As the first regression model RM1, for example, linear regression or quadratic regression can be used, but it is preferable to use a machine learning model using a decision tree algorithm such as Random Forest or XGBoost. . A machine learning model using a decision tree algorithm computes the degree of importance that indicates which variable influences the result, and can use the importance as the "influence" of the operating parameter.
(5) Second regression model RM2
The second regression model RM2 is used by the parameter adjustment unit 318 when adjusting the adjustment target parameter. The second regression model RM2 is a regression model in which the parameter value of the parameter to be adjusted is input and the performance index value is output. As the second regression model RM2, for example, linear regression or quadratic regression can be used, and machine learning models such as Gaussian process regression, neural network, and random forest can also be used.

As the operation parameters of the robot system, for example, the following operation parameters of the robot 100 can be used.
(1) Max Acceleration
The maximum acceleration is set for each axis of the robot 100 when the moment of inertia of the portion on the tip side of the axis with respect to the axis is maximum, that is, when the robot arm 120 is extended. is the maximum acceleration to be applied.
(2) Max Deceleration
The maximum deceleration is the maximum value of deceleration set for each axis of the robot 100 when the moment of inertia of the portion on the tip side of the axis with respect to the axis is at its maximum. .
(3) Acceleration Limit
The limit acceleration is set for each axis of the robot 100 when the moment of inertia of the portion on the tip side of the axis relative to the axis is minimum, that is, when the robot arm 120 is folded. is the maximum acceleration to be applied.
(4) Deceleration Limit
The critical deceleration is the maximum value of deceleration set for each axis of the robot 100 when the moment of inertia of the portion on the tip side of the axis with respect to the axis is at its minimum. .
(5) Weight Correction Coefficient Weight correction coefficient is a coefficient for adjusting the acceleration in the motion of each axis according to the load weight held by the robot 100 .
Note that any operation parameter other than these can be used as the operation parameter to be adjusted in this embodiment.

ここで、Ｎabは平均出力回転速度、Ｃは基本動定格荷重、ｆｗは荷重係数、Ｐｃは動等価ラジアル荷重である。荷重係数ｆｗの値は、振動の大きさに応じて１～３の値に設定される。上記（１）式による寿命の計算方法は周知である。（１）式のパラメーターのうち、平均出力回転速度Ｎabは、位置センサー４３０で計測される回転数から算出できる。荷重係数ｆｗは振動の大きさによって変わるので、振動センサー４１０の計測結果から得られる振動の大きさに応じて荷重係数ｆｗの値が調整される。振動の大きさと荷重係数ｆｗとの関係は、予め設定されている。
（４）各軸のモーターの過負荷率
モーターの過負荷率は、電流センサー４２０で計測されるモーターの電流値から算出できる。
（５）各軸のモーターのピークトルク
モーターのピークトルクも、電流センサー４２０で計測されるモーターの電流値から算出できる。 As the performance index value of the robot, for example, the following indexes can be used.
(1) Work tact time The work tact time is the time required for the robot 100 to work on one work. In the work using the robot 100, the tact time is often the same as the cycle time.
(2) Vibration of Specific Parts of Robot System As vibrations of specific parts of the robot system, for example, vibration of the robot 100, vibration of the pedestal 500, and the like can be used as performance index values. Vibration of the robot 100 is obtained from the measurement result of the vibration sensor 410 . As the vibration of the robot 100, for example, the amount of positional overshoot when reaching the target position, the amount of change in acceleration when reaching the target position, and the like can be used. The motion position of each joint measured by the position sensor 430 is also used when obtaining the overshoot amount. A vibration sensor is installed on the pedestal 500 when the vibration of the pedestal 500 is used as the performance index value.
(3) Life expectancy value of specific part of robot system As the life of the specific part of the robot system, for example, the life of the speed reducer provided for each axis can be used. For example, in the following formula, it is assumed that a strain wave gear reducer is used, and the service life of the ball bearing is calculated as the service life L of the speed reducer.

Here, Nab is the average output rotational speed, C is the basic dynamic load rating, fw is the load factor, and Pc is the dynamic equivalent radial load. The value of the load factor fw is set to a value between 1 and 3 according to the magnitude of vibration. The method of calculating the service life using the formula (1) is well known. Among the parameters of equation (1), the average output rotation speed Nab can be calculated from the rotation speed measured by the position sensor 430 . Since the load coefficient fw varies depending on the magnitude of vibration, the value of the load coefficient fw is adjusted according to the magnitude of vibration obtained from the measurement result of the vibration sensor 410 . The relationship between the magnitude of vibration and the load factor fw is set in advance.
(4) Motor Overload Rate of Each Axis The motor overload rate can be calculated from the current value of the motor measured by the current sensor 420 .
(5) Peak Torque of Motor for Each Axis The peak torque of the motor can also be calculated from the current value of the motor measured by the current sensor 420 .

When adjusting the operating parameters of the robot 100, one or more of the plurality of performance index values described above can be used. Either of the following two methods can be adopted for optimization processing when using a plurality of performance index values.
<Optimization Processing Using Objective Function and Constraints>
In this optimization process, one of a plurality of performance index values is set as an objective function and the other is set as a constraint condition. Constraints are conditions that require performance index values to fall within an acceptable range. For example, if the tact time is the objective function and the amount of overshoot and the life of the reducer are set as constraints, the operating parameters are set to , is adjusted so that the takt time is the best value, that is, the smallest value. In this embodiment, the case of using this optimization process will be described.
<Optimization processing using objective function including multiple performance index values>
In this optimization process, a single objective function is set by weighting and adding a plurality of performance index values. For example, when the tact time E1, the overshoot amount E2, and the reduction gear life E3 are used as performance index values, the objective function Et can be expressed by the following equation.
Et=k1*E1+k2*E2+k3/E3 (1)
where k1 to k3 are positive coefficients.
At this time, the operating parameters are adjusted so that the objective function Et, which is the comprehensive performance index value, has the best value, that is, the smallest value.

FIG. 3 is a flow chart showing the procedure for adjusting operating parameters. This processing is executed by the processor 310 of the information processing device 300 .

In step S10, the performance index value calculation unit 312 causes the robot 100 to perform a motion using a plurality of motion conditions obtained by combining candidate values for a plurality of motion parameters, and uses the measurement results of the sensors 400 during the motion. Execute the process for obtaining the performance index value. In this process, adjustment operations are used to adjust operation parameters.

FIG. 4 is an explanatory diagram showing an example of the adjustment operation. The horizontal axis of FIG. 4 indicates time change, and the vertical axis indicates simplified position change of the robot 100 . Here, a trajectory is drawn that sequentially traces a plurality of teaching points from the first teaching point P1 to the teaching point P4 during the period from time t0 to time t7, and finally returns to the position of the first teaching point P1. . After returning to the position of the first taught point P1 at time t7, it is in a standby state for the next work until time t8. At this time, the time from time t0 to time t8 corresponds to takt time or cycle time. Operation commands for such adjustment operations are created as an operation program RP and stored in the memory 350 .

FIG. 5 is a flow chart showing the detailed procedure of step S10. In step S110, the performance index value calculator 312 reads the initial values of the operation parameters from the parameter file PF in the memory 320, and sets them as candidate values of the operation parameters.

In step S120, the performance index value calculation unit 312 causes the robot 100 to perform a motion using the motion parameter candidate values. In step S130, the performance index value calculator 312 obtains the performance index value from the measurement results of the sensors 400 in operation. In this embodiment, the tact time of the work, the overshoot amount of the vibration of the robot 100, and the life of the reduction gear of each axis are used as the performance index values.

In step S140, the performance index value calculation unit 312 updates the candidate values of the operation parameters by executing the optimization process of the operation parameters using the optimization algorithm. In this embodiment, optimization is performed with the takt time as the objective function and with the amount of vibration overshoot and the life of the reducer of each axis as the constraint conditions. That is, under the constraint that the amount of vibration overshoot generated by the operation of the robot 100 is less than or equal to a predetermined vibration threshold, and that the life of the reduction gear of each axis is greater than or equal to a predetermined life threshold, Candidate values for operation parameters are searched for and updated so that the takt time of work is shortened as much as possible. Various algorithms such as random search, CMA-ES (Covariance Matrix Adaptation Evolution Strategy), Nelder-Mead method, and Bayesian optimization can be used as the optimization algorithm. Also, in order to prevent the optimization processing time from becoming excessively long, it is preferable to use an optimization algorithm other than the grid search.

In step S150, the performance index value calculation unit 312 determines whether the optimization processing has converged. If not converged, the process returns to step S120, and the processes of steps S120 to S150 are repeated until converged.

The process shown in FIG. 5 is a process of operating the robot under a plurality of operating conditions and obtaining a performance index value using the sensor measurement results during the operation. As described above, in this process, it is preferable to use an optimization algorithm other than the grid search to optimize the candidate values of the plurality of operating parameters using the performance index value as the objective function. However, even when such optimization is performed, the objective function does not always have the optimum value. This is because, for example, the optimization process may fall into a local optimum and a global optimum may not be obtained. Therefore, in the process of step S70, which will be described later, by further adjusting the adjustment target parameter with a high degree of influence, even if the optimization process using the optimization algorithm in step S10 is insufficient, sufficiently good performance can be obtained. Operational parameter values can be adjusted to provide index values. However, the process of step S10 may be executed without using the optimization algorithm. In this case, the user may preset a combination of candidate values of the operating parameters to be used under a plurality of operating conditions. In this case as well, it is preferable to search for candidate values of operating parameters that give the best performance index value.

In step S20 of FIG. 3, the performance index value calculator 312 deletes abnormal data from the data acquired in step S10. Abnormal data can be determined, for example, according to the presence or absence of an abnormality notification from the control device 200 of the robot 100 . Alternatively, Mahalanobis distances may be calculated for a plurality of data, and data with large Mahalanobis distances may be determined as abnormal data. Note that the process of step S20 may be omitted.

In step S30 of FIG. 3, the regression model construction unit 314 constructs the first regression model RM1 using the data obtained in steps S10 and S20. The first regression model RM1 is a regression model that inputs parameter values of a plurality of operating parameters and outputs performance index values. In this embodiment, a machine learning model using a decision tree algorithm is used as the first regression model RM1.

FIG. 6 is an explanatory diagram showing an example of the first regression model RM1. In this first regression model RM1, a total of 12 operations of maximum acceleration (Max Acceleration), maximum deceleration (Max Deceleration), limit acceleration (Acceleration Limit), and limit deceleration (Deceleration Limit) for three axes J1 to J3 It is a random forest that takes parameters as input and outputs takt time, overshoot amount, and speed reducer life of axes J1 to J3. Instead of building one regression model with multiple outputs, multiple regression models with only one output may be built. When using a machine learning model such as a random forest as the first regression model RM1, learning of the machine learning model is executed in step S30. In this learning, the data obtained in steps S10 and S20 are used as teacher data. As can be understood from this explanation, in the present disclosure, the meaning of "constructing a regression model" means making a state in which regression can be executed using the regression model.

In step S40 of FIG. 3, the parameter selection unit 316 uses the first regression model RM1 to calculate the degree of influence of the operating parameter on the performance index value. In this embodiment, a machine learning model using a decision tree algorithm is used as the first regression model RM1. It can be used as an influence of operating parameters. This importance is also called Variable Importance.

Instead of using the importance of the decision tree algorithm, the influence of the operation parameter may be calculated according to the following equation.
[Influence]=[Change in performance index value]/[Change in parameter value] (2)
The degree of influence using this formula (2) can be applied when using the first regression model RM1 that does not use the decision tree algorithm. In addition, it is preferable that the right side of the above equation (2) takes an absolute value. Moreover, it is preferable that the change in the performance index value is obtained as the change in the performance index value when the parameter value is changed from the optimum value of the operation parameter obtained in the optimization process of step S10.

FIG. 7 is a graph showing the degree of influence of operating parameters on takt time. The horizontal axis of the figure is the importance calculated by the first regression model RM1 constructed as a random forest. In this example, the maximum acceleration (Max Acceleration J1) of axis J1 has the greatest influence, the maximum deceleration (Max Deceleration J1) of axis J1 has the second greatest influence, and the deceleration limit of axis J3 (Deceleration Limit It can be seen that J3) has the third highest degree of influence. As described below, for operating parameters that have a high degree of influence on the performance index value, it is expected that the performance index value will be further improved when the parameter value is changed from the optimum value obtained in step S10. can.

FIG. 8 is a graph showing an example of changes in tact time when operating parameters are changed. Here, as operation parameters, the robot 100 is operated under conditions in which the parameter values are changed from the optimum values obtained in step S10 for the three operation parameters having the highest degrees of importance shown in FIG. The results are shown. As can be seen from this example, there is often room for further optimization of the optimum values of the operating parameters obtained in step S10. Therefore, as described below, some operation parameters that have a large influence on the takt time, which is the performance index value, are selected as adjustment target parameters, and the performance index value is further improved by readjusting them. be able to.

In step S50 of FIG. 3, the parameter selection unit 316 displays the degree of influence of each operation parameter as shown in FIG. When the user viewing this display selects one or more operating parameters having a high degree of influence as adjustment target parameters and issues a selection instruction, parameter selection section 316 accepts the selection instruction. According to this process, the user can select an operation parameter with a high degree of influence as a parameter to be adjusted. Selection is made so that the number of parameters to be adjusted is less than the number of operating parameters used in step S10.

In step S60, the regression model constructing unit 314 constructs a second regression model RM2 that receives the parameter value of the parameter to be adjusted and outputs the performance index value. As the data for constructing the second regression model RM2, the data obtained in step S10 can be reused. If the data obtained in step S10 is insufficient, the robot 100 may be operated to supplement the data.

FIG. 9 is an explanatory diagram showing an example of the second regression model RM2. In this example, in step S50, the maximum acceleration of axis J1, the maximum deceleration of axis J1, and the limit deceleration of axis J3 are selected as parameters to be adjusted, and these three operating parameters are used in the second regression model RM2. is the input of The outputs of the second regression model RM2 are the tact time, the amount of overshoot, and the reduction gear life of the axes J1 to J3, similarly to the first regression model RM1 shown in FIG. Alternatively, as with the first regression model RM1, a plurality of regression models having only one output may be constructed. Linear regression and quadratic regression can be used as the second regression model RM2, and machine learning models such as Gaussian process regression, neural networks, and random forests can also be used. Since the second regression model RM2 has fewer inputs than the first regression model RM1, it is possible to perform regression with higher accuracy.

In steps S70 and S80 of FIG. 3, the parameter adjustment unit 318 uses the second regression model RM2 to adjust the adjustment target parameter. This adjustment is performed so as to improve the performance index value, which is the objective function. In this embodiment, the takt time is the objective function, and the amount of overshoot and the life of the speed reducers of the axes J1 to J3 are the constraints. The parameter value of the target parameter is searched. A grid search, for example, can be used for this search.

FIG. 10 is an explanatory diagram showing an example of the search range in steps S70 and S80. Here, for convenience of explanation, an example is shown in which two parameters to be adjusted are the maximum acceleration of the axis J1 and the maximum deceleration of the axis J1. The black circle in the center indicates the position of the optimum value obtained in step S10. In this example, a grid search is used, a range of ±5% from the optimum value is set as a search range, and a search point is set for each minute search step. The same applies when there are three or more parameters to be adjusted. The search range is set wide enough to obtain a global optimal solution. However, since the wider the search range, the longer the processing time, the narrower the search range, the better, in order not to excessively increase the processing time. Considering these points, the search range for each parameter to be adjusted is preferably set to a range of ±10% or less when the optimum value obtained in step S10 is 100%. % or less is more preferable.

In steps S70 and S80, performance index values are estimated using the second regression model RM2 for all search points within the search range, and the operating parameter values that yield the best performance index values are determined. By adjusting the parameters to be adjusted using grid search in this way, it is possible to increase the possibility of obtaining a global optimal solution without falling into a local optimal solution. In addition, in the processing of steps S70 and S80, the parameters to be adjusted are limited to only some of all the operation parameters, so even if grid search is used, it is possible to prevent the search from taking an excessive amount of time. However, the parameter to be adjusted may be adjusted using a search method other than the grid search. The parameter values of the adjustment target parameters adjusted in this way are saved in the parameter file PF and used for the work of the robot 100 .

As described above, in the first embodiment, the first regression model RM1 is used to examine the operating parameters that have a high degree of influence on the performance index value, and the second regression model RM2 is used to adjust the operating parameters that have a high degree of influence. Therefore, the operating parameter values can be adjusted to obtain a sufficiently good performance index value.

B. Second Embodiment FIG. 11 is a flow chart showing the procedure of step S50 in the second embodiment. The difference of the second embodiment from the first embodiment is that step S50 of FIG. 3 is executed according to the procedure of FIG. be. In the procedure of FIG. 11, the adjustment target parameter is automatically selected by the parameter selection unit 316 .

In step S510, the parameter selection unit 316 adds one of the operating parameters having a high degree of influence on the performance index value as a parameter to be adjusted. As described in the first embodiment, in step S40 of FIG. 3, the degree of influence of each operating parameter on the performance index value is obtained as illustrated in FIG. The parameter selection unit 316 refers to this result, selects one operation parameter in descending order of influence, and adds it as an adjustment target parameter in step S510.

In step S<b>520 , the regression model construction unit 314 constructs a regression model that receives the parameter values of the adjustment target parameters that have been added so far and outputs the performance index values. As the data for constructing the regression model, the data obtained in step S10 is reused. As this regression model, it is possible to use linear regression or quadratic regression in the same way as the second regression model RM2, and it is also possible to use machine learning models such as Gaussian process regression, neural networks, and random forests. is also possible.

In step S530, the parameter selection unit 316 predicts the performance index value using the regression model constructed in step S520, and calculates the error from the actual measurement value of the performance index value obtained in step S10 of FIG. . When predicting the performance index value, the performance index value is obtained by inputting the operating parameter values under the operating conditions used in step S10 into the regression model. In step S540, the parameter selection unit 316 determines whether or not the prediction accuracy of the regression model has deteriorated compared to before adding one adjustment target parameter in step S510. Since step S540 is irrelevant when the first parameter to be adjusted is added, the process returns to step S510, and the next parameter to be adjusted is added in order of influence. On the other hand, in step S540, if the prediction accuracy of the regression model has deteriorated, in step S550, the last added parameter to be adjusted is removed, and the parameter to be adjusted that has been used up to that point is changed to the parameter shown in FIG. is selected as a parameter to be adjusted to be used in the processing from step S60 onward. Whether or not the prediction error has deteriorated may be determined based on whether or not the amount of increase in the prediction error is equal to or greater than a threshold.

The adjustment target parameter selected in step S550 may be displayed on the display unit 350 to notify the user. Alternatively, the regression model constructed in step S520 may be used as it is as the second regression model RM2.

As described above, in the second embodiment, the parameter to be adjusted is automatically selected according to the degree of influence of each operating parameter on the performance index value. Therefore, the parameter to be adjusted is selected without requesting user operation. It is possible to

・Other embodiments:
The present disclosure is not limited to the embodiments described above, and can be implemented in various forms without departing from the scope of the present disclosure. For example, the present disclosure can also be implemented in the following aspects. The technical features in the above embodiments corresponding to the technical features in each form described below are used to solve some or all of the problems of the present disclosure, or to achieve some or all of the effects of the present disclosure. In order to achieve the above, it is possible to appropriately replace or combine them. Also, if the technical features are not described as essential in this specification, they can be deleted as appropriate.

(1) According to a first aspect of the present disclosure, there is provided a method of adjusting multiple operating parameters for a robot system having a robot and sensors used to determine performance index values of the robot. This method includes (a) causing the robot to perform an action using a plurality of operating conditions obtained by combining candidate values for the plurality of operating parameters, and calculating the performance index value using the measurement result of the sensor during the action. and (b) using the performance index values obtained in step (a), constructing a first regression model having the parameter values of the plurality of operating parameters as inputs and the performance index values as outputs. (c) using the first regression model to determine the degree of influence of each operating parameter on the performance index value; and (d) selecting one or more of the plurality of operating parameters according to the degree of influence. and (e) using the performance index value obtained in the step (a), a second step of inputting the parameter value of the adjustment target parameter and outputting the performance index value and (f) using the second regression model to adjust the parameter to be adjusted such that the performance index value is improved.
According to this method, the first regression model is used to examine the operating parameters that have a high impact on the performance index value, and the second regression model is used to perform adjustments for the high impact operating parameters, so that the performance is sufficiently good. Operating parameter values can be adjusted to provide performance index values.

(2) In the above method, the first regression model may be a machine learning model using a decision tree algorithm, and the degree of influence may be a degree of importance calculated by the machine learning model.
According to this method, the importance calculated by the decision tree machine learning model is used as the influence of each operation parameter on the performance index value, so that the operation parameter with high influence can be easily selected as the parameter to be adjusted. can.

(3) In the above method, the step (a) includes the step of optimizing the candidate values for the plurality of operating parameters using an optimization algorithm other than grid search, with the performance index value as an objective function, The step (f) may include adjusting the parameter to be adjusted so that the performance index value is optimized using grid search.
According to this method, after optimizing candidate values for multiple operating parameters using an optimization algorithm, the parameters to be adjusted that have a high degree of influence are adjusted by grid search, so optimization by the optimization algorithm is insufficient. Even in these cases, the operating parameter values can be adjusted to obtain sufficiently good performance index values.

(4) In the above method, the step (d) includes displaying the degree of influence of each operating parameter on the performance index value, and selecting the parameter to be adjusted from among the plurality of operating parameters. and receiving instructions.
According to this method, the user can select an operating parameter with a high degree of influence as a parameter to be adjusted.

(5) In the above method, the performance index value includes at least one of a takt time of work by the robot, vibration of a specific part of the robot system, and a life expectancy value of a specific part of the robot system. It can be a thing.
According to this method, the operating parameters can be adjusted using the takt time, vibration of specific parts of the robot system, predicted life expectancy of specific parts of the robot system, etc. as performance index values.

(6) According to a second aspect of the present disclosure, there is provided a robot system including a robot, a sensor used to obtain a performance index value of the robot, and a control unit that controls the robot. (a) causing the robot to perform an action using a plurality of operating conditions obtained by combining candidate values for the plurality of operating parameters; and (b) constructing a first regression model having the parameter values of the plurality of operating parameters as inputs and the performance index values as outputs, using the performance index values obtained in the process (a). (c) using the first regression model to determine the degree of influence of each operating parameter on the performance index value; (d) selecting one of the plurality of operating parameters according to the degree of influence; and (e) using the performance index value obtained in the processing (a), the parameter value of the adjustment target parameter is input and the performance index value is output. A process of constructing a two-regression model; and (f) a process of adjusting the parameter to be adjusted such that the performance index value is improved using the second regression model.
According to this robot system, the first regression model is used to examine the operating parameters that have a high degree of influence on the performance index value, and the second regression model is used to perform adjustments regarding the operating parameters that have a high degree of influence. The operating parameter values can be adjusted to obtain a reasonable performance index value.

The present disclosure can also be implemented in various forms other than those described above. For example, realization in the form of a robot system comprising a robot and a robot control device, a computer program for realizing the functions of the robot control device, a non-transitory storage medium on which the computer program is recorded, etc. can do.

DESCRIPTION OF SYMBOLS 100... Robot, 110... Base, 120... Robot arm, 150... End effector, 200... Control apparatus, 300... Information processing apparatus, 310... Processor, 312... Performance index value calculation part, 314... Regression model construction part, 316 ... parameter selection unit, 318 ... parameter adjustment unit, 320 ... memory, 330 ... interface circuit, 340 ... input device, 350 ... display section, 400 ... sensors, 410 ... vibration sensor, 420 ... current sensor, 430 ... position sensor, 500... Mounting frame

Claims (6)

1. A method of adjusting a plurality of operating parameters for a robotic system having a robot and sensors used to determine a performance index value of said robot, comprising:
(a) causing the robot to perform a motion using a plurality of motion conditions obtained by combining candidate values for the plurality of motion parameters, and obtaining the performance index value using the measurement results of the sensor during the motion;
(b) using the performance index values obtained in step (a), constructing a first regression model having the parameter values of the plurality of operating parameters as inputs and the performance index values as outputs;
(c) determining the impact of each operating parameter on the performance index value using the first regression model;
(d) selecting one or more adjustment target parameters from among the plurality of operating parameters according to the degree of influence;
(e) using the performance index value obtained in step (a) to construct a second regression model having the parameter value of the parameter to be adjusted as an input and the performance index value as an output;
(f) using the second regression model to adjust the parameter to be adjusted so that the performance index value is improved;
A method, including 2. The method of claim 1, wherein
The first regression model is a machine learning model using a decision tree algorithm,
The method according to claim 1, wherein the influence is an importance calculated by the machine learning model. 3. A method according to claim 1 or 2,
The step (a) includes optimizing the candidate values for the plurality of operating parameters using an optimization algorithm other than a grid search, with the performance index value as an objective function;
The method, wherein the step (f) includes adjusting the parameter to be adjusted so that the performance index value is the best using a grid search. The method according to any one of claims 1 to 3,
The step (d) is
displaying the impact of each operating parameter on the performance index value;
receiving a user's selection instruction to select the parameter to be adjusted from among the plurality of operating parameters;
A method, including The method according to any one of claims 1 to 4,
The method, wherein the performance index value includes at least one of a takt time of work by the robot, vibration of a specific part of the robot system, and a life expectancy value of a specific part of the robot system. a robot system,
robot and
a sensor used to determine a performance index value of the robot;
a control unit that controls the robot;
with
The control unit
(a) a process of causing the robot to perform an action using a plurality of operating conditions obtained by combining candidate values for the plurality of operating parameters, and obtaining the performance index value using a measurement result of the sensor during the action;
(b) a process of using the performance index values obtained in the process (a) to construct a first regression model having the parameter values of the plurality of operating parameters as inputs and the performance index values as outputs;
(c) using the first regression model to determine the degree of influence of each operating parameter on the performance index value;
(d) selecting one or more adjustment target parameters from among the plurality of operating parameters according to the degree of influence;
(e) using the performance index value obtained in the processing (a) to construct a second regression model having the parameter value of the parameter to be adjusted as an input and the performance index value as an output;
(f) using the second regression model to adjust the parameter to be adjusted so that the performance index value is improved;
A robot system that executes

Images