Robot teaching method, and teaching system

Abstract

To reduce amounts of work by an operator in robot teaching.SOLUTION: In a robot teaching method, a computer reads-in a plurality of teaching points (t1)-(t13) for a robot 2 that is controlled using a correction mode 41 for an absolute position error to perform work to work-pieces 31 and 32, which are given to the robot by offline teaching, and the computer extracts the teaching point that is corrected using the correction model, out of the plurality of teaching points, on the basis of similarity of respective motion of the robot at the plurality of teaching points, and then the computer outputs the extracted teaching point.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The technology disclosed herein relates to a robot teaching method and a teaching system. [Background technology]

[0002] Patent Document 1 describes a conventional offline teaching device. The conventional offline teaching device uses a neural network to suppress absolute position errors in a six-axis vertical articulated robot. The conventional offline teaching device corrects teaching points given to the robot using a mathematical model. The conventional offline teaching device also inputs the teaching points corrected using the mathematical model into a neural network, and obtains, as an output from the neural network, a correction amount for further correcting the position error of the corrected teaching points. The robot control device corrects the teaching points corrected using the mathematical model using the correction amount, which is the output from the neural network, and then controls the robot. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 6-114769 Summary of the Invention [Problem to be solved by the invention]

[0004] The absolute position error of a robot is mainly due to individual differences between robots. To suppress the absolute position error, the absolute position error is determined for each robot. In offline teaching of a robot, not only in the conventional offline teaching device described above, but also in general, an operator performs the task of determining the absolute position error by actually operating the robot at each teaching point. The more teaching points given to a robot, the greater the amount of work the operator has to do.

[0005] Furthermore, if the type of workpiece differs, the teaching points given to the robot will also change, and every time a new type of workpiece is replaced with a new type of workpiece, the operator must perform a huge amount of work to determine the absolute position error for each teaching point. [Means for solving the problem]

[0006] The technology disclosed herein is a computer reads in a plurality of teaching points that are given to the robot by offline teaching, the teaching points being used by a robot controlled using a correction model of absolute position error to perform an operation on a workpiece; the computer extracts, from the plurality of teaching points, teaching points for which correction by the correction model is to be adjusted, based on a similarity between the movements of the robot at each of the plurality of teaching points; In this robot teaching method, the extracted teaching points are output by the computer. [Effects of the Invention]

[0007] The robot teaching method can reduce the amount of work required to adjust the correction using the correction model. [Brief explanation of the drawings]

[0008] [Figure 1] Figure 1 shows a robot teaching system. [Figure 2] The upper diagram in Figure 2 shows the procedure for correcting absolute position errors using a mechanism model, and the lower diagram shows the procedure for correcting absolute position errors using a black box model. [Figure 3] Figure 3 shows a method for extracting teaching points from the teaching points of the second workpiece. [Figure 4] Figure 4 shows the relationship between the similarity of the teaching points and the magnitude of the absolute position error of the robot controlled using the learned correction model. [Figure 5] FIG. 5 is a flowchart showing a procedure for extracting teaching points, which is executed by the teaching device. [Figure 6] FIG. 6 is a flowchart showing a procedure for learning a correction model, which is executed by the teaching device and the controller. [Figure 7] Figure 7 shows an image that shows the difference between the correction model before and after learning. [Figure 8] FIG. 8 shows an example in which learned teaching points are classified into multiple groups. [Figure 9] FIG. 9 shows a method for extracting teaching points to be used for learning from the teaching points of a workpiece during the initial learning of a correction model. [Figure 10] FIG. 10 shows an example in which the similarity is determined according to the information about the robot's motion and task. [Figure 11] FIG. 11 shows a method for extracting a group of teaching points to be used for learning from a group of teaching points that includes multiple consecutive teaching points during the initial learning of a correction model, based on the similarity of the group of teaching points. [Figure 12] FIG. 12 shows a method for extracting a group of teaching points for adjusting the correction by the correction model from a group of newly given teaching points when the correction model has already been learned. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of a robot teaching method and teaching system will be described with reference to the drawings. The robot teaching method and teaching system described here are merely examples.

[0010] (Robot teaching system) 1 shows a robot teaching system 1. The teaching system 1 teaches a robot 2 offline.

[0011] The teaching system 1 is equipped with a robot 2. The robot 2 is an industrial robot that performs work on a workpiece 3. The work performed by the robot 2 includes a variety of tasks. The robot 2 may, for example, perform spot welding on the workpiece 3. The robot 2 may, for example, perform laser welding on the workpiece 3. The robot 2 may, for example, perform milling on the workpiece 3. The robot 2 may, for example, perform AFP (Automated Fiber Placement) on the workpiece 3. An end effector according to the work performed by the robot 2 is attached to the robot 2. The robot 2 in the illustrated example spot welds on the workpiece 3. A welding gun 21 serving as an end effector is attached to the robot 2.

[0012] The workpiece 3 in the illustration is a car body. The illustration conceptually shows a car body. The workpiece 3 is not limited to a car body.

[0013] In the illustrated example, the robot 2 is an articulated robot. The robot 2 is not limited to an articulated robot. More precisely, the robot 2 is a vertical articulated robot. The robot 2 is not limited to a vertical articulated robot. The robot 2 has multiple joints. Specifically, the robot 2 is a six-axis robot having joint JT1, joint JT2, joint JT3, joint JT4, joint JT5, and joint JT6. The number of axes of the robot 2 is not limited to six. The number of axes of the robot 2 may be less than six or more than six. Note that the robot 2 may include an external device in addition to the robot that performs the work. The external device is, for example, a positioner or a turntable, or both a positioner and a turntable.

[0014] Each of the joints JT1-JT6 connects adjacent links. Each of the joints JT1-JT6 rotates adjacent links relative to each other around a rotation axis. Each of the joints JT1-JT6 has an actuator. The actuator is a power source that moves the joints JT1-JT6. The actuator is, for example, an electric motor. The electric motor is, for example, a servo motor. Note that the joints of the robot 2 are not limited to joints with rotation axes. The joints of the robot 2 may also be joints with linear axes.

[0015] The teaching system 1 includes a controller 4. The controller 4 controls the robot 2. The controller 4 is a computer including at least a processor, a memory, and an interface. The actuators of the joints JT1-JT6 are electrically connected to the controller 4. The controller 4 outputs control signals to the actuators. The actuators operate in accordance with the control signals. The operation of the actuators causes the robot 2 to perform work on the workpiece 3. In this example, the robot 2 spot welds to the body of an automobile.

[0016] The teaching system 1 includes a teaching device 5. The teaching device 5 performs offline teaching of the robot 2. The teaching device 5 is a computer including at least a processor, a memory, and an interface. The teaching device 5 is connected to a controller 4. The teaching device 5 outputs teaching points to the controller 4. The controller 4 controls the robot 2 in accordance with the teaching points.

[0017] An input device 51 is connected to the teaching device 5. Note that the input device 51 is not an essential element of the teaching system 1. The input device 51 inputs data to the teaching device 5. Examples of data input to the teaching device 5 through the input device 51 include three-dimensional model data of the workpiece 3 and the robot 2. The three-dimensional model data may be so-called CAD (Computer Aided Design) data. The teaching device 5 sets points on the workpiece 3 to be spot-welded by the robot 2, i.e., teaching points, based on the three-dimensional model data of the workpiece 3 and the robot 2. The teaching device 5 sets multiple teaching points for one workpiece 3. In the example of FIG. 1, the teaching device 5 sets ten teaching points (t1), (t2), (t3), (t4), (t5), (t6), (t7), (t8), (t9), and (t10) for one workpiece 3. The number of teaching points is not limited to a specific number. Information about the teaching point may be input to the teaching device 5 through the input device 51.

[0018] An output device 52 is connected to the teaching device 5. Note that the output device 52 is not an essential element of the teaching system 1. The output device 52 provides various information to an operator of the teaching device 5. The output device 52 is, for example, a display. The output device 52 may also be, for example, a printer. As will be described later, the output device 52 outputs to the operator information on a teaching point that is extracted from a plurality of teaching points and that adjusts the correction by the correction model 41.

[0019] (Robot position correction) As described above, the robot 2 operates according to the teaching points set by the teaching device 5. When the robot 2 operates, the actual position of the tip of the robot 2 deviates from the teaching points, i.e., the ideal position. This position deviation is called an absolute position error. The cause of the absolute position error is not limited to a specific factor, but examples include differences in link length and individual actuators, or deflection of the link.

[0020] The controller 4 has a function of correcting the absolute position error. Specifically, the controller 4 has a correction model 41. The correction model 41 corrects the command value to be output to the robot 2 so as to eliminate the absolute position error of the robot 2 with respect to the input teaching point. The controller 4 is not limited to having one correction model 41. As will be described later, the controller 4 may have multiple correction models 41.

[0021] 2 shows steps 42 and 43 of position correction by the controller 4 using a correction model 41. The correction model 41 includes two types of models: a mechanism model 411 and a black box model 412. The controller 4 has the mechanism model 411 or the black box model 412.

[0022] The mechanism model 411 is a model that is set in advance based on the mechanism of the robot 2. The black box model 412 is a model based on a neural network or machine learning. The black box model 412 is not a model based on the mechanism of the robot 2. Note that the black box model 412 is not limited to a neural network or machine learning.

[0023] The position correction procedures 42 and 43 performed by the controller 4 include a learning phase and a correction phase.

[0024] In the learning phase, the controller 4 causes the correction model 41 to learn. Specifically, the robot 2 is actually operated according to the teaching points set by the teaching device 5, and the amount of deviation between the tip position of the robot 2 and the ideal position, or the amount of correction from the tip position to the ideal position, is obtained. The amount of deviation or the amount of correction is obtained by measuring the distance or teaching the robot 2. An operator may measure the distance or teach the robot 2. Alternatively, a measuring device may measure the distance without the intervention of an operator. The measuring device may measure the amount of deviation between the actual tip position of the robot 2 and the ideal position, for example, using a camera.

[0025] The controller 4 learns the correction model 41 based on the teaching points and the acquired deviation amount or correction amount. The learning of the correction model 41 can be performed by a known method.

[0026] Once the correction model 41 is trained, the controller 4 controls the robot 2 using the trained correction model 41 in the correction phase, i.e., when operating the robot 2. The controller 4 inputs teaching points to the correction model 41 and outputs command values ​​corrected by the correction model 41 to the robot 2. The absolute position error of the robot 2 is eliminated or reduced.

[0027] For the initial learning of the correction model 41, the deviation amount or correction amount is acquired for each teaching point given to the controller 4 by the teaching device 5. The greater the number of teaching points given to the workpiece 3, the greater the workload for acquiring the deviation amount or correction amount.

[0028] The workpiece 3 on which the robot 2 works is not limited to one type, but the robot 2 works on each of multiple types of workpieces 3. In the example of spot welding on an automobile body, the multiple types of workpieces 3 correspond to bodies of different car models, for example.

[0029] The multiple teaching points (t1)-(t10) of the first workpiece 31 shown in Fig. 3 are not the same as the multiple teaching points (t11), (t12), and (t13) of the second workpiece 32, which is a different type from the first workpiece 31. Note that in the example of Fig. 3, the number of teaching points of the second workpiece 32 is intentionally reduced to facilitate understanding. In reality, the number of teaching points of the second workpiece 32 is approximately the same as the number of teaching points of the first workpiece 31.

[0030] Even if the correction model 41 has been learned for each of the multiple teaching points (t1)-(t10) of the first workpiece 31, it does not necessarily mean that the absolute position error can be accurately suppressed for the multiple teaching points (t11), (t12), and (t13) of the second workpiece 32. In order to adjust the correction by the correction model 41 for the second workpiece 32, the operator must once again obtain the amount of deviation or correction for each of the multiple teaching points of the second workpiece 32. Every time the number of types of workpieces 3 increases, the operator has to perform a huge amount of work to correct the absolute position error.

[0031] The teaching system 1 and the teaching method for the robot 2 disclosed herein can reduce the amount of work required to adjust the correction using the correction model 41.

[0032] The teaching system 1 and teaching method disclosed herein are based on the following finding of the inventors: that is, a correction model 41 learned for at least one of a plurality of teaching points at which the robot 2 behaves similarly can accurately suppress absolute position errors for the other teaching points in the plurality of teaching points, in the same way as for the one teaching point. Hereinafter, extraction of teaching points based on the similarity of the robot 2's behavior will be described with reference to the drawings.

[0033] Example 1 3 shows an example of extracting teaching points for adjusting the correction by the correction model 41 from among multiple teaching points of the second workpiece 32 when the correction model 41 has already been learned for the first workpiece 31. Table 61 in FIG. 3 shows the motion of the robot 2 at each of teaching points (t1), (t2), (t3), (t4), (t5), (t6), (t7), (t8), (t9), and (t10) given to the robot 2 for the first workpiece 31. Table 62 in FIG. 3 shows the motion of the robot 2 at each of teaching points (t11), (t12), and (t13) given to the robot 2 for the second workpiece 32. The motion of the robot 2 is represented by the axis values ​​of the joints JT1, JT2, JT3, JT4, JT5, and JT6. In the case of joints JT1, JT2, JT3, JT4, JT5, and JT6 having rotation axes, the axis values ​​may be the angles of the joints JT1, JT2, JT3, JT4, JT5, and JT6. Note that the axis values ​​of the joints are shown in obscured letters in Figure 3.

[0034] Graph 63 in Figure 3 is a two-dimensional coordinate plot of only the axis values ​​of joints JT1 and JT2 out of the six joints to visually show the movement of robot 2 at teaching points (t1)-(t13).

[0035] The correction model 41 has been trained for all of the teaching points (t1)-(t10) of the first workpiece 31. In the graph 63, the trained correction model 41 is visualized by a closed curve that includes the teaching points (t1)-(t10). It is assumed that the trained correction model 41 can accurately correct the absolute position error of the robot 2 for the teaching points within the closed curve.

[0036] The teaching device 5 receives the teaching points (t1)-(t10) of the first workpiece 31 and the teaching points (t11), (t12), and (t13) of the second workpiece 32. The teaching device 5 determines the similarity of each of the teaching points (t11), (t12), and (t13) of the second workpiece 32 to the teaching points (t1)-(t10) of the first workpiece 31. More precisely, the teaching device 5 determines the similarity of the behavior of the robot 2 at each of the teaching points (t11), (t12), and (t13) of the second workpiece 32 to the behavior of the robot 2 at the teaching points (t1)-(t10) of the first workpiece 31. Note that hereinafter, the similarity of the behavior of the robot 2 at the teaching points will simply be referred to as the similarity of the teaching points.

[0037] The teaching device 5 may determine the similarity using a statistical method. The teaching device 5 may determine the similarity using a known method for determining outliers. Specific examples of known methods for determining outliers include (1) Isolation Forest, (2) Local Outlier Factor, (3) One Class Support Vector Machine, and (4) Hierarchical Clustering.

[0038] In graph 63, the similarity of the teaching points is indicated by the distance between the teaching points. In other words, if the distance between the teaching points is short, they are similar, and if the distance between the teaching points is long, they are dissimilar. Note that the judgment of the similarity of the teaching points is actually based on the movement of the robot 2 represented by the axis values ​​of the six joints JT1-JT6. In graph 63, even if the coordinates of the teaching points of the first workpiece 31 and the second workpiece are close to each other, it does not necessarily mean that the two teaching points are similar.

[0039] In the example of Figure 3, the teaching points (t11) and (t12) of the second workpiece 32 are close to the teaching points (t1), (t3), (t5), and (t8) of the first workpiece 31. The teaching points (t11) and (t12) of the second workpiece 32 are similar to the teaching points (t1), (t3), (t5), and (t8) of the first workpiece 31. The teaching point (t13) of the second workpiece 32 is far from the teaching points (t1)-(t10) of the first workpiece 31. The teaching point (t13) of the second workpiece 32 is dissimilar to the teaching points (t1)-(t10) of the first workpiece 31.

[0040] For the teaching points (t11) and (t12) of the second workpiece 32, the correction model 41 can accurately suppress the absolute position error of the robot 2. As shown in the graph 63 of FIG. 3, the teaching points (t11) and (t12) are included in the correction model 41. For the teaching point (t13) of the second workpiece 32, the correction model 41 cannot accurately suppress the absolute position error of the robot 2. As shown in the graph 63 of FIG. 3, the teaching point (t13) is not included in the correction model 41.

[0041] Here, with reference to FIG. 4, the relationship between the similarity of the taught points and the magnitude of the absolute position error of the robot 2 controlled using the correction model 41 will be explained. The vertical axis in FIG. 4 represents the similarity of the taught points. The similarity of the taught points is calculated by Isolation Forest based on the motion of the robot 2 represented by the axis values ​​of the joints JT1-JT6. Note that the axis values ​​of the joints JT1-JT6 are angles here.

[0042] The horizontal axis in Figure 4 represents the magnitude of the absolute position error. As can be seen from the figure, there is a correlation between the similarity of the taught points and the magnitude of the absolute position error; the higher the similarity of the taught points, the smaller the absolute position error. In other words, for the taught points of the second workpiece 32 that are similar to the taught points of the first workpiece 31, the absolute position error of the robot 2 can be accurately suppressed using the correction model 41. On the other hand, for the taught points of the second workpiece 32 that are not similar to the taught points of the first workpiece 31, the absolute position error of the robot 2 cannot be accurately suppressed using the correction model 41. For taught points that are not similar to the taught points of the first workpiece 31, it is better to suppress the absolute position error, in other words, adjust the correction using the correction model 41.

[0043] A tolerance E for the absolute position error may be set in advance. A threshold value for determining whether the points are similar or dissimilar may be set based on the tolerance E. The teaching device 5 determines whether the given teaching points are similar or dissimilar based on the set threshold value.

[0044] The teaching device 5 extracts teaching points for adjusting the correction by the correction model 41 from among the multiple teaching points of the second workpiece 32 based on the aforementioned similarity. The teaching device 5 also provides information on the extracted teaching points to an operator via the output device 52. The operator teaches, for example, the robot 2, only the provided teaching points. By teaching the robot 2, the absolute position error at the teaching points is suppressed. Of the multiple teaching points of the second workpiece 32, teaching points that were not provided via the output device 52 do not need to be taught. Because the operator does not need to teach the robot 2 about all teaching points of the second workpiece 32, the amount of operator work is reduced compared to conventional methods. Teaching the robot 2 is an example of adjusting the correction by the correction model 41.

[0045] 5 shows a procedure for extracting teaching points executed by the teaching device 5. In step S11, the teaching device 5 reads teaching points through the input device 51. In the example of FIG. 3, the teaching device 5 reads teaching points (t1)-(t10) for the first workpiece 31 and multiple teaching points (t11), (t12), and (t13) for the second workpiece 32.

[0046] In the next step S12, the teaching device 5 determines the similarity of the read teaching points. In the example of Fig. 3, the teaching device 5 determines the similarity of each of the multiple teaching points (t11), (t12), and (t13) for the second workpiece 32 to the teaching points (t1)-(t10) for the first workpiece 31.

[0047] In step S13, the teaching device 5 outputs the teaching points extracted based on the similarity determination in step S12 through the output device 52. In the example of Fig. 3, the teaching device 5 outputs only the teaching point (t13) that is not similar to the teaching point of the first workpiece 31, among the teaching points (t11), (t12), and (t13) of the second workpiece 32.

[0048] (Action and effect) As described above, when the correction model 41 has been learned for the first workpiece 31, the teaching system 1 extracts teaching points for teaching the robot 2 from among the multiple teaching points of the second workpiece 32. The extraction of teaching points is based on the similarity of the operation of the robot 2 between the teaching points of the first workpiece 31 and the teaching points of the second workpiece 32. The number of teaching points of the second workpiece 32 at which the robot 2 is actually taught is reduced. The teaching system 1 and teaching method can reduce the workload of the operator.

[0049] Example 2 In the first embodiment, the operator instructs the robot 2 based on the teaching points provided through the output device 52. Instead of teaching the robot 2, the correction model 41 may be learned. Learning the correction model 41 is an example of adjusting the correction by the correction model 41.

[0050] 6 shows a procedure for learning the correction model 41, which is executed by the teaching device 5 and the controller 4. In step S21, the teaching device 5 reads teaching points through the input device 51. In this example, similar to the example in FIG. 3, the teaching device 5 reads a plurality of teaching points (t1)-(t10) for the first workpiece 31 and a plurality of teaching points (t11), (t12), and (t13) for the second workpiece 32.

[0051] In the next step S22, the teaching device 5 determines the similarity of the read teaching points. In this example, the teaching device 5 determines the similarity of each of the multiple teaching points (t11), (t12), and (t13) for the second workpiece 32 to the teaching points (t1)-(t10) for the first workpiece 31 (see FIG. 3).

[0052] In step S23, the teaching device 5 outputs the teaching points extracted based on the similarity determination in step S12 through the output device 52. In this example, the teaching device 5 outputs the teaching point (t13) that is not similar to the teaching point of the first workpiece 31, from among the teaching points (t11), (t12), and (t13) of the second workpiece 32.

[0053] The operator actually operates the robot 2 in accordance with the provided teaching points, and obtains the amount of deviation between the tip position of the robot 2 and the ideal position, or the amount of correction from the tip position to the ideal position. The operator inputs the obtained amount of deviation or amount of correction to the controller 4 via the input device 51 and the teaching device 5. Instead of the operator inputting, the teaching device 5 may output dissimilar teaching points to the controller 4 in step S23, and the controller 4 may actually operate the robot 2 in accordance with the provided teaching points, and use a measuring device to obtain the amount of deviation between the tip position of the robot 2 and the ideal position, or the amount of correction from the tip position to the ideal position, without the intervention of the operator.

[0054] In step S24, the controller 4 determines whether or not actual data for the extracted teaching point, i.e., the actual amount of deviation or amount of correction, has been acquired. If the actual data has been acquired, the controller 4 trains the correction model 41 in the following step S25. The training of the correction model 41 is performed in accordance with the training phase of FIG.

[0055] (Action and effect) As described above, the teaching system 1 and teaching method can reduce the workload of the operator because the number of teaching points used to learn the correction model 41 is reduced among the teaching points of the second workpiece 32. Furthermore, even when the correction model 41 is learned without the intervention of an operator through cooperation between the teaching device 5 and the controller 4, the burden on the operator is reduced because the number of teaching points used to learn is reduced.

[0056] FIG. 7 is an image illustrating the effect of training the correction model 41. The correction model 41 is trained for taught points (t1)-(t10) of the first workpiece 31. Assume that the correction model 41 is trained based on a dissimilar taught point (t13) of the second workpiece 32. In graph 64 of FIG. 7, the correction model 41 is expanded to include taught point (t13). For example, if a taught point (t14) for a new third workpiece is given that is similar to taught point (t13), the correction model 41 before training may not be able to suppress the absolute position error of taught point (t14), but the correction model 41 after training can suppress the absolute position error of taught point (t14).

[0057] Learning the correction model 41 can reduce the workload of the operator when the number of types of workpieces increases.

[0058] Example 3 3, the correction model 41 learned for the teaching points of the first workpiece 31 includes all the teaching points. The number of correction models 41 possessed by the controller 4 is not limited to one. The teaching points (t1)-(t10) of the first workpiece 31 may be divided into one or more groups based on their mutual similarities, and a correction model 41 may be learned for each group.

[0059] 8 shows an example in which the teaching points (t1)-(t10) of the first workpiece 31 are classified into a first group, a second group, and a third group based on the similarity. A correction model 41 is trained separately for each of the first group, the second group, and the third group. The controller 4 essentially has a plurality of correction models 41.

[0060] When the teaching points (t11), (t12), and (t13) of the second workpiece 32 are given to the robot 2, the teaching device 5 determines the similarity of each of the teaching points (t11), (t12), and (t13) of the second workpiece 32 with the teaching points (t1)-(t10) of the first workpiece 31, as described above. In the example of FIG. 8, the teaching point (t11) is included in the first group, and the teaching point (t12) is included in the second group. There is no need to teach the robot 2 or learn the correction model 41 for the teaching points (t11) and (t12).

[0061] The teaching point (t13) is not included in any of the first, second, and third groups. For the teaching point (t13), the robot 2 is taught or the correction model 41 is learned. The teaching device 5 outputs information about the teaching point for teaching the robot 2 or learning the correction model 41 through the output device 52. In the third embodiment, the teaching device 5 outputs the information about the teaching point in accordance with the flowchart of FIG.

[0062] As in the first or second embodiment described above, the operator teaches the robot 2 about the provided teaching points, or makes the correction model 41 learn.

[0063] When the correction model 41 is to be learned, the teaching device 5 and the controller 4 learn the correction model 41 in accordance with the flowchart of Fig. 6. Of the correction models 41 of the first group, the second group, and the third group, the correction model 41 of the group that is most similar to the teaching point (t13) is learned. In the example of Fig. 8, the correction model 41 of the second group is learned based on the teaching point (t13). Note that a new fourth group may be provided for the teaching point (t13), and the correction model 41 of the fourth group may be learned based on the teaching point (t13). (Action and effect) If the correction model 41 is individually trained for each of the multiple groups classified based on similarity, the training accuracy of each correction model 41 is improved. The correction model 41 can accurately suppress the absolute position error for the teaching points of the corresponding group.

[0064] In addition, when teaching points for a new workpiece are given, the teaching device 5 determines whether or not teaching of the robot 2 or learning of the correction model 41 is necessary based on the similarity between the given teaching points and each of the multiple groups, thereby reducing the workload of the operator.

[0065] Example 4 The above-described first, second and third embodiments are all examples of reducing the workload of the operator when the correction model 41 has already been learned for the first workpiece 31 and teaching points for the second workpiece 32 are given.

[0066] The technique disclosed herein can reduce the workload of the operator even when teaching points are given for the first workpiece 31 and the correction model 41 is initially learned based on the teaching points.

[0067] 9 shows an example of extracting teaching points to be used for training the correction model 41 based on similarity when a plurality of teaching points (t1)-(t10) are given for the first workpiece 31. Similar to FIG. 3, a table 61 in FIG. 9 shows the motion of the robot 2 at each of the teaching points (t1), (t2), (t3), (t4), (t5), (t6), (t7), (t8), (t9), and (t10) given to the robot 2 for the first workpiece 31. A graph 66 in FIG. 9 plots the axis values ​​of the joints JT1 and JT2 at the teaching points (t1)-(t10) in two-dimensional coordinates.

[0068] When the teaching points (t1)-(t10) of the first workpiece 31 are given, the teaching device 5 determines the mutual similarity of the teaching points (t1)-(t10). The teaching device 5 classifies the teaching points (t1)-(t10) into one or more groups based on the similarity. In the example of FIG. 9, as shown in graph 67, the teaching points (t1)-(t10) are classified into a first group including teaching points (t1), (t2), (t3), and (t4), a second group including teaching points (t5), (t6), (t7), and (t8), and a third group including teaching points (t9) and (t10).

[0069] The teaching device 5 extracts a teaching point that represents the group. The teaching device 5 may extract a teaching point that represents the group based on an outlier of the teaching points included in the group. In the example of FIG. 9, (t1) is extracted as a teaching point that represents the first group, (t5) is extracted as a teaching point that represents the second group, and (t9) is extracted as a teaching point that represents the third group. Information on the extracted teaching points is provided to the operator via the output device 52.

[0070] The operator actually operates the robot 2 according to the provided teaching points, and acquires the amount of deviation between the tip position of the robot 2 and the ideal position, or the amount of correction from the tip position to the ideal position. The acquired amount of deviation or amount of correction is provided to the controller 4. In the example of FIG. 9, the operator only needs to acquire the amount of deviation or amount of correction for teaching points (t1), (t5), and (t9). The controller 4 trains a correction model 41 based on the teaching points and the acquired amount of deviation or amount of correction. The training of the correction model 41 is performed in accordance with the learning phase of FIG. 2.

[0071] In this example, the procedure for learning the correction model 41 conforms to Fig. 6. In step S21, the teaching device 5 reads teaching points for the first workpiece 31. In step S22, the teaching device 5 determines the similarity of the read teaching points for the first workpiece 31 and classifies the teaching points into one or more groups. In the following step S23, the teaching device 5 extracts a teaching point that represents the group and outputs information about the extracted teaching point. In steps S24 and S25, the controller 4 performs initial learning of the correction model 41.

[0072] The controller 4 may learn the correction model 41 for each group as shown in the third embodiment (see FIG. 8).

[0073] (Action and effect) For multiple teaching points where the robot 2 behaves similarly, if the correction model 41 is learned for one of the multiple teaching points, the correction model 41 can accurately suppress absolute position errors for the other teaching points as well. When the correction model 41 is first learned, it is not necessary to learn the correction model 41 for all given teaching points; if learning is performed only for teaching points extracted based on similarity, the correction model 41 can suppress absolute position errors for all given teaching points. The teaching system 1 and teaching method can realize efficient learning of the correction model 41. Furthermore, the teaching system 1 and teaching method can reduce the workload of the operator.

[0074] This example is effective when the correction model 41 is a black box model 412. This is because the trained black box model 412 can accurately suppress absolute position errors for multiple teaching points that are similar to each other. Also, the black box model 412 can achieve a high learning effect even if the number of teaching points used for learning is small.

[0075] Example 5 The similarity of the teaching points can be determined not only based on the movement of the robot 2 with respect to the teaching points, but also based on the movement of the robot 2 and information related to the work.

[0076] FIG. 10 shows an example of determining the similarity of teaching points based on the operation of the robot 2 and information about the task. Table 68 in FIG. 10 sets the axis values ​​of joints JT1-JT6, task type, task condition 1,..., task condition N for each of teaching points (t1)-(t10). Examples of task types include spot welding and laser welding. Examples of task conditions include whether welding is performed or not and the classification number of the workpiece. The task types and task conditions are quantified so that the teaching device 5 can determine the similarity. Note that the task types and task conditions are examples of task-related information.

[0077] By determining the similarity of the taught points based on the information about the robot 2's motion and the task, the teaching device 5 can determine the similarity in accordance with the tendency of the robot 2 to generate absolute position errors. For example, there are workpieces that tend to contract and workpieces that tend to expand. Assume that the first workpiece is a workpiece that tends to contract and the second workpiece is a workpiece that tends to expand. Even if the coordinates of the taught points for the first workpiece and the coordinates of the taught points for the second workpiece are the same and the motion of the robot 2, i.e., the axis values ​​of the joints JT1-JT6, are the same, the direction of the absolute position error of the robot 2 relative to the first workpiece that tends to contract will differ from the direction of the absolute position error of the robot 2 relative to the second workpiece that tends to expand. Determining the similarity based on the information about the robot 2's motion and the task allows for a more accurate determination of the similarity in accordance with the tendency of the robot 2 to generate absolute position errors.

[0078] Once teaching points are extracted based on the similarity determined based on the information regarding the movement of the robot 2 and the work, teaching of the robot 2 or learning of the correction model 4 is performed according to the above-mentioned embodiment 1, 2, 3 or 4.

[0079] (Action and effect) The teaching device 5 can more accurately determine the degree of similarity based on information about the movement and task of the robot 2. The teaching system 1 or teaching method can accurately suppress the absolute position error of the robot 2 while reducing the workload of the operator.

[0080] Example 6 In the above example, the teaching points of the robot 2 are given as independent points. As described above, the teaching points given as independent points are teaching points given when the robot 2 performs spot welding.

[0081] When the task of the robot 2 is to move the end effector of the robot 2 continuously in a line relative to the workpiece 3, a teaching point cloud including a plurality of consecutive teaching points is given to the robot 2. The technology disclosed herein is also effective when a teaching point cloud including a plurality of consecutive teaching points is given to the robot 2.

[0082] Examples of tasks in which the end effector of the robot 2 is moved continuously linearly relative to the workpiece 3 include laser welding, milling, and AFP.

[0083] FIG. 11 shows an example of extracting a teaching point group to be used for the initial learning of the correction model 41 when multiple teaching point groups (t1), (t2), (t3), (t4), (t5), (t6), (t7), (t8), (t9) and (t10) are given.

[0084] The teaching device 5 determines the similarity between the teaching point groups based on the shape of the trajectory of the end effector of the robot 2 moving in accordance with the teaching point groups. Graph 69 in Fig. 11 is an image of the trajectory shapes of the teaching point groups (t1)-(t10) plotted in two-dimensional coordinates. In the two-dimensional coordinates in Fig. 11, teaching point groups with similar trajectory shapes are close to each other, and teaching point groups with dissimilar trajectory shapes are far from each other.

[0085] The shape of the trajectory can be acquired, for example, by an image sensor capturing a 2D image or a 3D image. The teaching device 5 reads the shape of the trajectory, for example, through the input device 51. The shape of the trajectory can also be acquired, for example, by simulating a three-dimensional model of the robot 2. The teaching device 5 may acquire the shape of the trajectory by simulating the read three-dimensional model of the robot 2.

[0086] The teaching device 5 also determines the similarity between multiple teaching point groups using a known image classification method based on the shape of the trajectory. Similar to the example of FIG. 9, the teaching device 5 classifies the teaching point groups (t1)-(t10) into one or more groups based on the similarity. In the graph 610 of FIG. 11, the teaching point groups (t1)-(t10) are classified into a first group including teaching point groups (t1), (t2), (t3), and (t4), a second group including teaching point groups (t5), (t6), (t7), and (t8), and a third group including teaching points (t9) and (t10). The first group is a group of similar circular trajectory shapes. The second group is a group of similar rectangular trajectory shapes. The third group is a group of similar triangular trajectory shapes.

[0087] The teaching device 5 extracts a teaching point group representative of the group. As described above, the teaching device 5 may extract a teaching point group representative of the group based on outliers of the teaching point groups included in the group. Information on the extracted teaching point group is provided to the operator via the output device 52.

[0088] The operator actually operates the robot 2 according to the provided teaching point group, and acquires the amount of deviation between the tip position of the robot 2 and the ideal position, or the amount of correction from the tip position to the ideal position. The acquired amount of deviation or correction amount is provided to the controller 4. In the example of FIG. 11 , the operator only needs to acquire the amount of deviation or correction amount for the teaching point groups (t1), (t5), and (t9). The controller 4 trains a correction model 41 based on the teaching points and the acquired amount of deviation or correction amount. The training of the correction model 41 is performed in accordance with the training phase of FIG. 2. The teaching device 5 and the controller 4 train the correction model 41 in accordance with the flowchart of FIG. 6. Note that the controller 4 may train the correction model 41 for each group, as shown in the third embodiment (see FIG. 8 ).

[0089] 12 shows an example in which a correction model 41 has been trained for teaching point groups (t1), (t2), (t3), (t4), (t5), (t6), (t7), (t8), (t9), and (t10) including a plurality of consecutive teaching points, and teaching point groups (t11), (t12), and (t13) are newly assigned. Note that the correction model 41 is trained for each group.

[0090] The teaching device 5 determines the similarity of each of the given teaching point groups (t11), (t12), and (t13) to the teaching point group (t1)-(t10). The teaching device 5 determines the similarity based on the shape of the trajectory of the end effector of the robot 2.

[0091] In the graph 611 of FIG. 12, the teaching point cloud (t11) is included in the first group, and the teaching point cloud (t12) is included in the second group. For the teaching point cloud (t11) and the teaching point cloud (t12), there is no need to teach the robot 2 or learn the correction model 41. The teaching point cloud (t13) has a star-shaped trajectory, so it is not included in any of the first, second, or third groups. The operator performs work on the teaching point cloud (t13) to teach the robot 2 or learn the correction model 41. The teaching point cloud (t13) may be classified into a new fourth group, separate from the first, second, and third groups. The new fourth group is learned based on the teaching point cloud (t13).

[0092] (Action and effect) The teaching system 1 and teaching method can reduce the amount of work required for teaching the robot 2 or learning the corrected model 41, even when a teaching point cloud is given to the robot 2.

[0093] The similarity of the robot 2's motion related to the teaching point group can be determined based on the shape of the trajectory of the tip of the end effector. By determining the similarity based on the shape of the trajectory of the tip of the end effector, the teaching device 5 can accurately extract the teaching point group for teaching the robot 2 or learning the correction model 41. The teaching system 1 and teaching method simultaneously reduce the amount of work for the operator and suppress absolute position errors caused by the correction model 41.

[0094] (Variation) When determining the similarity of the movements of the robot 2 represented by the axis values ​​of the joints JT1-JT6, the teaching device 5 may represent the axis values ​​of the joints JT1-JT6 as vectors. For example, the angles of the joints JT1-JT6 of the robot 2, +180° and -180°, are significantly different in terms of numerical value. The teaching device 5 may determine that the axis values ​​of the joints JT1-JT6 of the robot 2 are dissimilar. However, the postures of the joints JT1-JT6 of the robot 2 are substantially the same when the angles of the joints JT1-JT6 are +180° and -180°.

[0095] The axis values ​​of the joints JT1-JT6 can be expressed as a vector with an angle θ and a length R, for example. When the axis values ​​of the joints JT1-JT6 are expressed as vectors, the teaching device 5 can determine that the angles of the joints JT1-JT6 at +180° and -180° are similar. The teaching device 5 can appropriately extract the teaching points in the above-described embodiment.

[0096] When determining the degree of similarity, the teaching device 5 may normalize the axis values ​​of the joints JT1-JT6 based on the range of motion of the joints JT1-JT6. Normalization allows the movement of the robot 2 to be represented more accurately. The teaching device 5 can determine the degree of similarity more accurately. The teaching device 5 can appropriately extract the teaching points in the above-described embodiment.

[0097] The teaching device 5 may change the weights among the multiple joints JT1-JT6 when determining the similarity. The weights may be changed based on the magnitude of the influence on the absolute position error of the robot 2. The weights may be increased for the joints JT1-JT6 that have a large influence on the absolute position error, and decreased for the joints JT1-JT6 that have a small influence on the absolute position error. The teaching device 5 can determine the similarity more accurately. The teaching device 5 can appropriately extract the teaching points in the above-mentioned embodiment.

[0098] The movement of the robot 2 related to the judgment of similarity of the teaching points may be expressed not only by the axis values ​​of the joints JT1-JT6 but also by the position and orientation (x, y, z, O, A, T) of the end effector. Note that (x, y, z) represent the position of the tip of the end effector, and (O, A, T) represent the orientation of the tip of the end effector. Furthermore, the movement of the robot 2 related to the judgment of similarity of the teaching points may also be expressed by the position (x, y, z) of the end effector.

[0099] The teaching device 5 may have a correction model 41 instead of the controller 4. The teaching device 5 outputs the teaching points corrected by the correction model 41 to the controller 4. The controller 4 controls the robot 2 according to the corrected teaching points. The teaching device 5 having the correction model 41 also learns the correction model 41.

[0100] The teaching device 5 may be incorporated into the controller 4. In other words, the controller 4 may have the functions of the teaching device 5 described above.

[0101] The teaching device 5 in the teaching system 1 may be on the cloud. The input device 51 and the output device 52 are terminals for the teaching device 5 on the cloud and function as terminals related to operations by an operator. The controller 4 may not be directly connected to the teaching device 5 on the cloud, but may be connected to the teaching device 5 via a terminal.

[0102] The functions of the elements disclosed herein can be performed using circuits or processing circuits, including general-purpose processors, special-purpose processors, integrated circuits, ASICs (Application Specific Integrated Circuits), conventional circuits, and / or combinations thereof, configured or programmed to perform the disclosed functions. A processor is considered a processing circuit or circuit because it includes transistors and other circuitry. In this disclosure, a circuit, unit, or means is hardware that performs the recited functions or hardware that is programmed to perform the recited functions. The hardware may be hardware disclosed herein or other known hardware that is programmed or configured to perform the recited functions. Where the hardware is a processor, which is considered a type of circuit, the circuit, means, or unit is a combination of hardware and software, and the software is used to configure the hardware and / or processor.

[0103] (Aspect) The above embodiments are specific examples of the following aspects.

[0104] (Aspect 1) a computer (5) reads in a plurality of teaching points that are given to the robot (2) by offline teaching, the teaching points being used by the robot (2) controlled using the absolute position error correction model (41) to perform an operation on the workpiece (3); The computer (5) extracts, from the plurality of teaching points, teaching points for which correction by the correction model (41) is to be adjusted, based on the similarity of the motion of the robot (2) at each of the plurality of teaching points; The computer (5) outputs the extracted teaching points. Robot teaching method.

[0105] Based on the similarity of the robot (2)'s movements, teaching points for adjusting the correction by the correction model (41) are extracted, and the operator simply performs the work of teaching the robot (2) or learning the correction model (41) for the extracted teaching points. This robot teaching method can reduce the operator's workload. Even if the operator's workload is reduced, the absolute position error of the robot (2) is suppressed.

[0106] (Aspect 2) The correction model (41) is learned based on the teaching points, The computer (5) reads in a first teaching point for a first workpiece (31) that has already been learned in the correction model (41), and a plurality of second teaching points for a second workpiece (32) that is a different type from the first workpiece and that have not yet been learned in the correction model (41); The computer (5) extracts, from the plurality of second teaching points, the second teaching points that are not similar to the first teaching points based on the similarity between the motion of the robot (2) at the first teaching points and the motion of the robot (2) at the second teaching points. A robot teaching method according to aspect 1.

[0107] When a correction model 41 has been learned for the first workpiece 31, the robot 2 is taught or the correction model 41 is learned for the second taught points of the second workpiece 32 that are not similar to the first taught points of the first workpiece 31. This robot teaching method can reduce the amount of work for the operator.

[0108] (Aspect 3) The first teaching points are classified into one or more groups based on the similarity of the movements of the robot (2) at the first teaching points; The computer (5) extracts the second teaching point that does not belong to the group based on the similarity between the motion of the robot (2) at the first teaching point and the motion of the robot (2) at the second teaching point. A robot teaching method according to aspect 2.

[0109] Whether or not the second teaching point needs to be taught to the robot (2) or to learn the correction model (41) is determined based on whether or not the second teaching point belongs to a group, so the computer (5) can appropriately extract the second teaching point to be used for teaching the robot (2) or to learn the correction model (41).

[0110] (Aspect 4) The correction model (41) is learned individually for each group based on the teaching points included in the group. A robot teaching method according to aspect 3.

[0111] If a correction model (41) is trained individually for each group classified based on similarity, the training accuracy of the correction model (41) is improved. The trained correction model (41) can accurately suppress absolute position errors for the teaching points of the corresponding group.

[0112] (Aspect 5) The correction model (41) is learned based on the output teaching points. A robot teaching method according to any one of aspects 1 to 4.

[0113] Learning the correction model (41) makes it possible to reduce the workload of the operator when the number of types of workpieces (3) increases and new teaching points are given to the robot (2).

[0114] (Aspect 6) Furthermore, the computer (5) classifies the plurality of teaching points into one or more groups based on the similarity of the movements of the robot (2) at each of the plurality of teaching points for one workpiece (3); The computer (5) extracts a teaching point selected from the one group or a teaching point selected from each of the plurality of groups, The correction model (41) is initially trained based on the extracted teaching points. A robot teaching method according to any one of aspects 1 to 5.

[0115] When the correction model (41) is initially trained, if it is trained only for the teaching points extracted based on the similarity, the correction model (41) can suppress absolute position errors for all of the given teaching points. Since the correction model (41) is not trained for all of the given teaching points, efficient training of the correction model (41) is achieved. This teaching method can reduce the amount of operator work when the correction model (41) is initially trained.

[0116] (Aspect 7) The computer (5) determines the similarity using a statistical method. A robot teaching method according to any one of aspects 1 to 6.

[0117] The computer (5) can accurately determine the similarity using a statistical method, and therefore can accurately extract teaching points for teaching the robot (2) or learning the correction model (41).

[0118] (Aspect 8) The robot (2) includes a multi-joint robot having a plurality of joints (JT1-JT6), The motion of the robot (2) at the teaching point is represented by the axis values ​​of each of the plurality of joints (JT1-JT6). A robot teaching method according to aspect 7.

[0119] The axis values ​​of each of the multiple joints (JT1-JT6) can appropriately represent the movement of the robot (2).

[0120] (Aspect 9) The computer (5) represents the axis values ​​of the joints (JT1-JT6) as vectors in determining the similarity. A robot teaching method according to aspect 8.

[0121] The vector representation of the axis values ​​increases the accuracy of determining the similarity based on the axis values ​​of the joints (JT1-JT6), so the computer (5) can appropriately extract teaching points.

[0122] (Aspect 10) In determining the degree of similarity, the computer (5) normalizes the axis values ​​of the joints (JT1-JT6) by the range of motion of the joints (JT1-JT6). A robot teaching method according to aspect 8.

[0123] Normalization of the axis values ​​of the joints (JT1-JT6) improves the accuracy of determining the similarity based on the axis values ​​of the joints (JT1-JT6), so that the computer (5) can appropriately extract teaching points.

[0124] (Aspect 11) The computer (5) changes weights among the plurality of joints (JT1-JT6) in determining the similarity. A robot teaching method according to aspect 8.

[0125] By changing the weights, the influence of each of the multiple joints (JT1-JT6) on the absolute position error is taken into account, so the computer (5) can more accurately determine the similarity based on the axis values ​​of the joints (JT1-JT6).The computer (5) can then appropriately extract teaching points.

[0126] (Aspect 12) The motion of the robot (2) at the teaching point is represented by the position or the position and posture of the end effector of the robot (2). A robot teaching method according to aspect 7.

[0127] The position of the end effector, or the position and orientation, can also adequately describe the motion of the robot (2).

[0128] (Aspect 13) The computer (5) extracts teaching points for adjusting the correction by the correction model (41) based on the degree of similarity between the operation of the robot (2) at each of the plurality of teaching points and information related to the work on the workpiece (3). A robot teaching method according to any one of aspects 1 to 12.

[0129] Determining the similarity based on the information about the robot (2)'s movements and tasks makes it possible to determine the similarity more accurately in accordance with the tendency of the robot (2)'s absolute position error to occur.

[0130] (Aspect 14) The computer (5) reads a group of teaching points including a plurality of consecutive teaching points for the robot (2) to perform a continuous operation on the workpiece (3), The computer (5) extracts teaching point groups for adjusting the correction by the correction model (41) based on the similarity of the movements of the robot (2) in each of the plurality of teaching point groups; The computer (5) outputs the extracted teaching point group. A robot teaching method according to any one of aspects 1 to 13.

[0131] The technology disclosed herein is also applicable to a group of teaching points including a plurality of consecutive teaching points, and the teaching method reduces the workload of the operator.

[0132] (Aspect 15) The computer (5) determines the similarity based on the shape of a trajectory along which the end effector of the robot (2) moves in accordance with the teaching point cloud. A robot teaching method according to aspect 14.

[0133] The computer (5) can accurately determine the similarity of the robot (2)'s movement with respect to the teaching point group based on the shape of the trajectory.

[0134] (Aspect 16) a robot (2) that performs an operation on a workpiece (3), the robot (2) being controlled using an absolute position error correction model (41) in accordance with a plurality of teaching points that are given to the robot (2) by offline teaching, the teaching points being used by the robot (2) to perform an operation on the workpiece (3); a computer (5) that reads the plurality of teaching points, extracts teaching points for adjusting the correction by the correction model (41) from the plurality of teaching points based on the similarity of the movements of the robot (2) at each of the plurality of teaching points, and outputs the extracted teaching points. Teaching system (1).

[0135] The teaching system (1) can reduce the workload of the operator. [Explanation of symbols]

[0136] 1 Teaching system 2. Robot 3 Work 31 First Work 32 Second Work 4 Controller 41 Correction Model 5 Teaching device JT1 joint JT2 joint JT3 joint JT4 joint JT5 joint JT6 Joint

Claims

1. A computer reads a plurality of teaching points for a robot controlled using a correction model of an absolute position error to perform an operation on a workpiece, the teaching points being given to the robot by offline teaching; the computer extracts, from the plurality of teaching points, teaching points for which correction by the correction model is to be adjusted, based on a similarity between the movements of the robot at each of the plurality of teaching points; The computer outputs the extracted teaching points. A method for teaching a robot.

2. The robot teaching method according to claim 1, The correction model is trained based on the teaching points; The computer reads in a first teaching point for a first workpiece, the first teaching point having been learned in the correction model, and a plurality of second teaching points for a second workpiece of a different type from the first workpiece, the second teaching points having not been learned in the correction model; The computer extracts, from among the plurality of second teaching points, the second teaching points that are not similar to the first teaching points based on a similarity between a motion of the robot at the first teaching point and a motion of the robot at the second teaching point. A method for teaching a robot.

3. The robot teaching method according to claim 2, The first teaching points are classified into one or more groups based on a similarity of the motion of the robot at the first teaching points; The computer extracts the second teaching point that does not belong to the group based on a similarity between an operation of the robot at the first teaching point and an operation of the robot at the second teaching point. A method for teaching a robot.

4. The robot teaching method according to claim 3, The correction model is trained for each group individually based on the teaching points included in the group. A method for teaching a robot.

5. The robot teaching method according to claim 1, The correction model is trained based on the output teaching points. A method for teaching a robot.

6. The robot teaching method according to claim 1, Furthermore, the computer classifies the plurality of teaching points into one or more groups based on a similarity of the operation of the robot at each of the plurality of teaching points for one workpiece, The computer extracts a teaching point selected from the one group or a teaching point selected from each of the plurality of groups; The correction model is initially trained based on the extracted teaching points. A method for teaching a robot.

7. The robot teaching method according to claim 1, The computer determines the degree of similarity using a statistical method. A method for teaching a robot.

8. The robot teaching method according to claim 7, The robot includes an articulated robot having a plurality of joints, The motion of the robot at the teaching point is represented by the axis values ​​of the joints. A method for teaching a robot.

9. The robot teaching method according to claim 8, The computer represents the axis values ​​of the joints as vectors in determining the similarity. A method for teaching a robot.

10. The robot teaching method according to claim 8, the computer normalizes the axis value of the joint by the movable range of the joint in determining the similarity; A method for teaching a robot.

11. The robot teaching method according to claim 8, The computer changes weights among a plurality of joints in determining the similarity. A method for teaching a robot.

12. The robot teaching method according to claim 7, The motion of the robot at the teaching point is represented by a position or a position and a posture of an end effector of the robot. A method for teaching a robot.

13. The robot teaching method according to claim 1, The computer extracts teaching points based on a similarity between an operation of the robot at each of the plurality of teaching points and information related to an operation on the workpiece. A method for teaching a robot.

14. The robot teaching method according to claim 1, The computer reads a group of teaching points including a plurality of consecutive teaching points for the robot to perform a continuous operation on the workpiece, and the computer extracts a group of teaching points for adjusting a correction by the correction model based on a similarity of the motion of the robot in each of the plurality of groups of teaching points; The computer outputs the extracted teaching point group. A method for teaching a robot.

15. The robot teaching method according to claim 14, the computer determines a similarity based on a shape of a trajectory along which an end effector of the robot moves in accordance with the teaching point group; A method for teaching a robot.

16. a robot that performs an operation on a workpiece, the robot being controlled using a correction model of an absolute position error in accordance with a plurality of teaching points for the robot to perform the operation on the workpiece, the teaching points being given to the robot by offline teaching; a computer that reads the plurality of teaching points, extracts teaching points for adjusting a correction by the correction model from among the plurality of teaching points based on a similarity between the movements of the robot at each of the plurality of teaching points, and outputs the extracted teaching points. Teaching system.

Images