Information processing device, information processing method, and information processing program

Abstract

Reduce the probability of failure in tasks performed by robots. [Solution] The information processing device classifies each of the multiple sets of motion data and state data when the robot performs a task as success data or failure data for the task. Based on the motion data classified as success data from the multiple sets of motion data and state data, the device uses a first motion data that satisfies the conditions for a high success rate of the task to be executed and predicted state data based on the first motion data to predict whether the robot's task based on the first motion data will succeed or fail. If it is predicted that the robot's task will fail, the device modifies the first motion data.

Description

[Technical Field]

[0001] This disclosure relates to an information processing device, an information processing method, and an information processing program. [Background technology]

[0002] Patent Document 1 discloses a technology that acquires sensor information measuring the state of a robot or the robot's working environment, uses a machine learning model to take the sensor information as input and output robot operation commands and abnormality detection results for the operation commands, generates operation commands for abnormal situations, and switches the commands transmitted to the robot based on the abnormality detection results. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2023-146535 [Overview of the project] [Problems that the invention aims to solve]

[0004] However, the technology described in Patent Document 1 has room for improvement in terms of reducing the probability of failure of tasks performed by robots.

[0005] This disclosure aims to provide an information processing device, an information processing method, and an information processing program that can reduce the probability of failure of tasks performed by robots. [Means for solving the problem]

[0006] The information processing device in the first embodiment includes a processor, which performs clustering on motion data, which is time-series data of motion commands to the robot when the robot performs a task, and state data, which includes time-series data representing the state of the robot, thereby classifying each of a plurality of sets of motion data and state data as success data or failure data for the task, and generates first motion data, which is time-series data of motion commands to the robot that satisfies conditions defined as conditions for increasing the success rate of the task to be executed, based on the motion data classified as success data among the plurality of sets of motion data and state data. Using the generated first motion data and predicted state data including time-series data representing the robot's state predicted based on the first motion data, the system predicts whether the robot's operation based on the first motion data will be successful or unsuccessful. If the robot's operation is predicted to fail, the system performs clustering based on similarity between the multiple sets of motion data and state data and the first motion data and predicted state data. The system then modifies the first motion data using motion data that is classified as successful data among the multiple sets of motion data and state data and that is of the same classification as the first motion data and predicted state data.

[0007] In the first embodiment of the information processing device, each of multiple sets of motion data and state data is classified as either success data or failure data for the operation. Based on the motion data classified as success data from the multiple sets of motion data and state data, a first motion data is generated. Using the first motion data and predicted state data, which is state data predicted based on the first motion data, it is predicted whether the robot's operation based on the first motion data will be successful or unsuccessful. If it is predicted that the robot's operation will fail, the first motion data is modified using motion data classified as success data from the multiple sets of motion data and state data, and which is of the same classification as the first motion data and predicted state data. This reduces the probability of failure of the operation performed by the robot.

[0008] In the second embodiment of the information processing device, the processor extracts operation data from among the plurality of sets of operation data and operation data classified as success data among the state data, which includes the final state or initial state of the operation to be executed, whose difference from the target state or initial state is less than a threshold, and generates the first operation data based on the extracted operation data.

[0009] In the information processing device of the second embodiment, first operation data is generated based on operation data in which the initial state or final state is close to the initial state or target state of the operation to be performed. Therefore, the probability of failure of the operation performed by the robot can be further reduced.

[0010] In the third embodiment, the information processing device, in the first or second embodiment, if it is predicted that the robot's operation will fail, the processor modifies the first operation data so that the first operation data gradually approaches the operation data of the same classification, until it is predicted that the robot's operation based on the modified first operation data will be successful.

[0011] In the third embodiment of the information processing device, the first motion data is modified to approximate the motion data classified as successful data. Therefore, the probability of failure of the task performed by the robot can be further reduced.

[0012] In the fourth embodiment, the information processing device, in the first or second embodiment, assigns a success or failure label to the pair of operation data and state data based on the difference between the state data and target data, which is time-series data of the target state, and uses the assigned label in the clustering.

[0013] In the fourth embodiment, a success or failure label is assigned based on the difference between the state data and the target data. Therefore, success or failure can be determined without human intervention.

[0014] The information processing method of the fifth embodiment performs clustering on motion data, which is time-series data of motion commands to the robot when the robot performs a task, and state data, which includes time-series data representing the state of the robot, thereby classifying each of the multiple sets of motion data and state data as success data or failure data for the task, and based on the motion data classified as success data among the multiple sets of motion data and state data, generates first motion data, which is time-series data of motion commands to the robot that satisfy conditions defined as conditions for a high success rate of the task to be executed, and the generated first motion data and The computer performs a process to predict whether the robot's operation based on the first operation data will be successful or unsuccessful, using predicted state data which includes time-series data representing the robot's state predicted based on the first operation data. If it is predicted that the robot's operation will be unsuccessful, the computer performs clustering based on similarity between the multiple sets of operation data and state data and the first operation data and predicted state data, and modifies the first operation data using operation data that is classified as successful data among the multiple sets of operation data and state data and is of the same classification as the first operation data and predicted state data.

[0015] The information processing program according to the sixth aspect causes a computer to execute processing of: performing clustering on motion data, which is time-series data of motion commands for a robot when the robot executes a work, and state data including time-series data representing the state of the robot, thereby classifying each of a plurality of sets of said motion data and said state data as either success data or failure data of said work; based on the motion data classified as success data among said plurality of sets of said motion data and said state data, generating first motion data which is time-series data of motion commands for said robot that satisfies a condition defined as a condition for increasing the success rate of a work to be executed; using the generated first motion data and predicted state data including time-series data representing the predicted state of said robot based on said first motion data, predicting whether the work of said robot based on said first motion data will succeed or fail; when it is predicted that the work of said robot will fail, performing clustering based on similarity with respect to said plurality of sets of said motion data and said state data, and said first motion data and said predicted state data; and correcting said first motion data using motion data that is classified as success data among said plurality of sets of said motion data and said state data and that is in the same classification as said first motion data and said predicted state data.

Effects of the Invention

[0016] According to the present disclosure, the failure probability of work performed by a robot can be reduced.

Brief Description of Drawings

[0017] [Figure 1] It is a diagram showing a schematic configuration of a robot system. [Figure 2] It is a block diagram showing an example of the hardware configuration of an information processing apparatus. [Figure 3] It is a flowchart showing the flow of processing executed by the information processing apparatus in a preliminary phase. [Figure 4] It is a conceptual diagram showing the flow of data processing in the preliminary phase. [Figure 5] It is a graph for explaining motion data and state data. [Figure 6] It is a conceptual diagram showing an example of a clustering result. [Figure 7] It is a flowchart showing the flow of processing executed by the information processing apparatus in an execution phase. [Figure 8] It is a diagram for explaining state prediction processing. DETAILED DESCRIPTION OF EMBODIMENTS

[0018] Referring to Fig. 1, the configuration of the robot system 1 according to the present embodiment will be described. As shown in Fig. 1, the robot system 1 includes an information processing apparatus 10, a robot 12, and a sensor group 18. Examples of the information processing apparatus 10 include computers such as personal computers and server computers.

[0019] The robot 12 includes a robot arm 14 and a gripper 16. The robot arm 14 includes links and joints that connect the links, and rotate or linearly expand / contract by driving of a motor. In the robot arm 14, the motor is driven according to a command value output from the information processing apparatus 10, so that the rotation angle or expansion / contraction state of the joint is changed. Accordingly, the gripper 16 is controlled to be at a specified position and in a specified posture in a three-dimensional space.

[0020] The gripper 16 is a tool provided at the tip end of the robot arm 14 and capable of gripping an object O placed on a support surface P. The gripper 16 includes two fingers, and can grip the object O with the two fingers. Although the case where the gripper 16 includes two fingers is described in the present embodiment, the gripper 16 may include three or more fingers.

[0021] In this embodiment, the task performed by the robot 12 is described as stacking multiple objects O, initially placed at their respective positions on the support surface P, to a target position on the support surface P. It is assumed that the 3D data of the objects O and the height of the support surface P are known. The tasks performed by the robot 12 are not limited to this example. For example, the tasks performed by the robot 12 may not involve any objects O.

[0022] The sensor group 18 includes various sensors such as a sensor for measuring the joint angles of the robot arm 14, a 6-axis force sensor for the gripping part 16, motion capture, and a camera positioned so that the robot 12 and the object O are included in the shooting range. The measurement results from the sensor group 18 are transmitted to the information processing device 10.

[0023] Referring to Figure 2, the hardware configuration of the information processing device 10 according to this embodiment will be described. As shown in Figure 2, the information processing device 10 includes a CPU (Central Processing Unit) 21, memory 22, storage 23, external I / F (Interface) 24, communication I / F 25, and input device 26. Each of the CPU 21, memory 22, storage 23, external I / F 24, communication I / F 25, and input device 26 is connected to each other via a bus 29 so that they can communicate with one another. The CPU 21 is an example of a "processor".

[0024] The CPU 21 is a central processing unit that executes various programs and controls various parts. The CPU 21 is an example of a processor. The memory 22 temporarily stores programs or data as a working area. The storage 23 consists of a storage device such as an HDD (Hard Disk Drive), SSD (Solid State Drive), or flash memory, and stores various programs and data.

[0025] The storage 23 contains the information processing program 30. The CPU 21 reads the information processing program 30 from the storage 23 and executes the information processing program 30 using the memory 22 as a working area.

[0026] External I / F 24 is an interface for connecting external devices such as the robot 12 and the sensor group 18 to the information processing device 10. Communication I / F 25 is an interface for connecting the information processing device 10 to a network. Input device 26 is a device for user input, such as a keyboard, touch panel, and mouse.

[0027] Furthermore, storage 23 stores a database (hereinafter referred to as "Robot DB") 32 containing data related to the work of the robot 12. The Robot DB 32 contains data related to the work of the robot 12 that has been collected in advance. For example, the Robot DB 32 stores motion data, which is time-series data of motion commands given to the robot 12 when the robot 12 performs a task, and state data, which includes time-series data representing the state of the robot 12. This state data is, for example, observation data including the state of the robot 12 and the state of the object O measured by the sensor group 18. The state data includes, for example, the position of the gripping part 16, the angles of the joints of the robot arm 14, the position of the object O, the inclination of the object O, and the shape of the object O. The Robot DB 32 stores multiple sets of motion data and state data. These sets of motion data and state data include various types of work-related data, such as data when the robot 12 succeeds in a task and data when the robot 12 fails to complete a task. The work performed by robot 12 corresponding to the data contained in robot DB32 may be work based on motion data defined by the designer, work based on motion data generated by changing parameters according to predetermined rules, or work based on motion data generated by randomly adjusting parameters.

[0028] Next, with reference to Figures 3 to 6, the processing performed by the information processing device 10 in the preliminary phase will be explained. Figure 3 is a flowchart showing the flow of processing performed by the information processing device 10. The CPU 21 reads the information processing program 30 from the storage 23, loads it into the memory 22, and executes it, thereby executing the processing shown in Figure 3. Figure 4 shows an example of the data input / output flow.

[0029] As shown in Figure 4, as described above, the robot DB32 stores motion data a and state data o measured by the sensor group 18 when the robot 12 performs work based on motion data a, with these data associated with each other. As an example, as shown in Figure 5, motion data a is time-series data of the control quantities of the robot 12, as an example of motion commands to the robot 12 corresponding to a series of motion steps when the robot 12 performs work. State data o is time-series data of measured quantities representing the state of the robot 12 and the state of the object O, as measured by the sensor group 18, as an example of time-series data representing the state of the robot 12. The subscript "0:T" in motion data a and state data o indicates that it is time-series data from time 0 to time T. The superscript "k" in motion data a and state data o is a subscript to distinguish the pair of motion data a and state data o, and for example, a non-negative integer is assigned sequentially. The motion data a and state data o may also be data obtained when the robot 12 is virtually operated in a simulation environment that reproduces the real environment.

[0030] In step S10 of Figure 3, the CPU 21 assigns a success or failure label to each pair of operation data a and state data o from the robot DB 32. Specifically, for example, the CPU 21 assigns a success or failure label to the state data o and the target data s, which is time-series data of the target state, according to equation (1) below. goal The difference δ and the threshold θ thr By comparison with, the label of success or failure r k Assign target data s goal This is a time-series data containing the target state corresponding to the state data o at each point in time. The threshold θthr This is predetermined through experiments, etc. Here, 0 corresponds to the failure label, and 1 corresponds to the success label. As shown in Figure 4, the assigned label r k The operation data a and state data o are associated with a set and stored in the robot DB32. The user may make a success or failure judgment based on each of the operation data a and state data o sets in the robot DB32. In this case, the user may input the judgment result indicating success or failure via the input device 26. Furthermore, in this case, the CPU 21 may assign a success or failure label according to the input judgment result.

[0031]

number

[0032] In step S12, the CPU 21 performs clustering on multiple sets of operation data a and state data o to classify whether each set of operation data a and state data o is success data indicating that the task was successful or failure data indicating that the task was unsuccessful. A specific example of this clustering process will be explained with reference to Figure 6.

[0033] The CPU 21 receives labels r assigned to each of the multiple sets of operation data a and state data o. kUsing this method, each pair of action data a and state data o is classified into either success data or failure data. Next, the CPU 21 classifies similar pairs of action data a and state data o into the same set by clustering based on the similarity between the pairs of action data a and state data o. Hereinafter, this set will be referred to as the "similar set". Pairs of action data a and state data o classified into the same similar set mean that their similarity is above a certain degree. For this similarity, the CPU 21 uses methods such as dynamic time stretching, Euclidean distance, and Manhattan distance. Furthermore, based on the similarity, the CPU 21 classifies similar pairs of action data a and state data o into the same set using unsupervised learning methods such as k-means, SVM (Support Vector Machine), and Gaussian mixture models. As a result, as shown in Figure 6 as an example, each pair of action data a and state data o is classified into one of the similar sets, and within that similar set, it is classified into either a set of success data (hereinafter referred to as the "success set") or a set of failure data (hereinafter referred to as the "failure set"). In the example in Figure 6, the black circles represent a pair of action data a and state data o, and the distance between the black circles indicates the similarity between the pairs of action data a and state data o. As shown in Figure 4, the classification result c is obtained by the above clustering. k This data is saved in the robot DB32. When step S12 is completed, the process ends.

[0034] Next, referring to Figures 7 and 8, the processing performed by the information processing device 10 in the execution phase will be described. In the execution phase, the initial state and target state of the task to be performed by the robot 12 are set in advance. The initial state and target state include the state of the robot 12 and the state of the object O.

[0035] In step S20 shown in Figure 7, the CPU 21 generates first motion data, which is time-series data of motion commands for robot 12 that satisfy conditions defined as conditions for a high success rate of the task to be executed, based on the motion data classified as success data in step S12 from among multiple sets of motion data a and state data o stored in robot DB 32.

[0036] Specifically, the CPU 21 extracts motion data a corresponding to state data o including a final state whose difference from the target state of the work to be executed is less than a threshold value, from among motion data a classified as success data among a plurality of sets of motion data a and state data o. The final state in state data o refers to state data o representing the last state of state data o which is time-series data k T . It should be noted that the CPU 21 may extract motion data a corresponding to state data o including an initial state whose difference from the initial state of the work to be executed is less than a threshold value, from among motion data a classified as success data among a plurality of sets of motion data a and state data o. The initial state in state data o refers to state data o representing the first state of state data o which is time-series data k 0.

[0037] Then, the CPU 21 generates the extracted motion data a as first motion data that satisfies a condition defined as a condition for increasing the success rate of the work to be executed. It should be noted that, when there exists a plurality of pieces of motion data a corresponding to state data o including a final state where the difference is less than the threshold, the CPU 21 may extract the motion data a with the smallest difference. Further, in this case, for example, the CPU 21 may generate the first motion data by performing processing such as waveform regression on the plurality of extracted pieces of motion data a.

[0038] In step S22, the CPU 21 predicts state data including time-series data representing states of the robot 12 and the object O when it is assumed that the robot 12 performs work based on the first motion data generated in step S20 (hereinafter referred to as "predicted state data"). As an example shown in FIG. 8, the CPU 21 obtains a state x at a certain time point t t , predicted state data is generated using a state prediction unit 40 that predicts a state change based on motion from said state x tThis corresponds to image data measured by the sensor group 18 and state data including the joint angles of the robot 12. The state prediction unit 40 may be a prediction model constructed using deep learning such as RNN (Recurrent Neural Network) based on the robot DB 32, or it may be constructed using state transition probabilities used in reinforcement learning. The CPU 21 may predict the robot's joint angles using a dynamics model if the characteristics of the robot arm 14 (e.g., inertia, mass, and dimensions) are known.

[0039] In step S24, the CPU 21 uses the predicted state data generated in step S22 to determine whether the robot 12's operation based on the first motion data generated in step S20 will be successful or unsuccessful. For example, the CPU 21 predicts success or failure by performing binary classification, classifying the first motion data and the predicted state data into success or failure. The binary classification model may be a model using an RNN or CNN (Convolutional Neural Network) that classifies time series data as input. Alternatively, the binary classification model may use a classification algorithm such as SVM or logistic regression, after reducing the dimensionality of the time series data features using an autoencoder. The user may also determine whether the robot 12's operation based on the first motion data will be successful or unsuccessful. In this case, the user may input a judgment result indicating success or failure via the input device 26. Furthermore, in this case, the CPU 21 may determine success or failure according to the input judgment result. If the CPU 21 determines in step S24 that the robot 12's operation will fail, the process proceeds to step S26.

[0040] In step S26, the CPU 21 performs clustering based on similarity between the first motion data and predicted state data and the multiple sets of motion data a and state data o stored in the robot DB 32, similar to step S12. In step S28, the CPU 21 identifies motion data a that, based on the clustering in step S26, is classified as successful data among the multiple sets of motion data a and state data o, and is of the same classification as the first motion data and predicted state data. Specifically, the CPU 21 identifies motion data a that is classified into the same similarity set as the first motion data and predicted state data, and is classified into the success set.

[0041] In step S30, the CPU 21 modifies the first operation data using the operation data identified in step S28. Specifically, the CPU 21 modifies the first operation data by integrating the first operation data and the operation data identified in step S28 at a predetermined ratio according to the following equation (2). (2) a new This shows the first operational data after the correction, a old This represents the first operation data before correction. Also, in equation (2), a * b represents the motion data obtained by regressing the waveforms of the multiple motion data identified in step S28. b represents the ratio and is a value greater than 0 and less than 1 (i.e., 0 <b<1)である。

[0042]

number

[0043] In step S32, the CPU 21, similar to step S24, determines whether the robot 12's operation based on the first operation data modified in step S30 will succeed or fail. If the CPU 21 determines in step S32 that the robot 12's operation will fail, the process returns to step S30. In this case, in step S30, the CPU 21 gradually brings b in equation (2) closer to 1 each time the number of iterations of step S30 increases. That is, if the CPU 21 determines that the robot 12's operation will fail, it modifies the first operation data so that the first operation data gradually approaches the operation data identified in step S28, repeating this process until it is predicted that the robot 12's operation based on the modified first operation data will succeed. If the CPU 21 determines in step S32 that the robot 12's operation will succeed, the process proceeds to step S34.

[0044] In step S34, the CPU 21 controls the robot 12 based on the first operation data modified in step S30, causing the robot 12 to perform the task. In step S36, the CPU 21 determines whether the task performed in step S34 was successful or unsuccessful. This determination may be made by the CPU 21, for example, as in step S10, or by the user. If the CPU 21 determines in step S36 that the robot 12's task was unsuccessful, the process returns to step S30; if it determines that the robot 12's task was successful, the process terminates.

[0045] On the other hand, if the CPU 21 determines in step S24 that the robot 12's work is successful, the process proceeds to step S38. In step S38, the CPU 21 controls the robot 12 based on the first motion data generated in step S20, causing the robot 12 to perform the work. In step S40, the CPU 21 determines, similar to step S36, whether the work performed in step S38 was successful or unsuccessful. If the CPU 21 determines in step S40 that the robot 12's work was unsuccessful, the process proceeds to step S26; if it determines that the robot 12's work was successful, the process terminates.

[0046] As described above, this embodiment makes it possible to reduce the probability of failure in the tasks performed by the robot 12. Furthermore, by collecting motion data for various tasks, it is possible to handle not only specific tasks but also various tasks such as when the target changes. In addition, by using not only data on successful tasks but also data on failed tasks, it is possible to generate motion data that takes failures into account.

[0047] The processing described in the above embodiment can also be implemented using dedicated hardware circuits. In this case, it may be executed on a single piece of hardware or on multiple pieces of hardware.

[0048] In the above embodiment, the term "processor" refers to a broad type of processor, including general-purpose processors (e.g., CPU: Central Processing Unit, etc.) and dedicated processors (e.g., GPU: Graphics Processing Unit, ASIC: Application Specific Integrated Circuit, FPGA: Field Programmable Gate Array, programmable logic device, etc.).

[0049] Furthermore, the processor operations in the above embodiments may not be performed by a single processor, but may be performed by multiple processors located in physically separate locations working together. Alternatively, some or all of the operations performed by specific multiple processors in each of the above embodiments may be integrated and performed by a single processor. In addition, the order of the processor operations is not limited to the order described in each of the above embodiments, and may be changed as appropriate.

[0050] Furthermore, the program for operating the information processing device 10 may be provided on a computer-readable recording medium such as a USB (Universal Serial Bus) memory, flexible disk, CD-ROM (Compact Disc Read Only Memory), or DVD (Digital Versatile Disc), or it may be provided online via a network such as the Internet. In this case, the program recorded on the computer-readable recording medium is usually transferred to memory or storage and stored therein. This program may also be provided, for example, as a standalone application software, or it may be incorporated into the software of each device as a function of the information processing device 10. This program can also be provided as a program product. A program product includes all forms of products for providing a program. For example, a program product includes a program provided via a network such as the Internet, and non-temporary computer-readable recording media such as CD-ROMs and DVDs on which the program is stored.

[0051] The following additional information is disclosed regarding the embodiments described above. (Note 1) Equipped with a processor, The aforementioned processor, By performing clustering on motion data, which is time-series data of motion commands given to the robot when the robot performs a task, and state data, which includes time-series data representing the state of the robot, it is possible to classify whether each of multiple sets of motion data and state data is success data or failure data for the task. Based on the aforementioned multiple sets of motion data and the motion data classified as success data among the state data, a first motion data is generated, which is time-series data of motion commands to the robot that satisfy conditions defined as conditions for increasing the success rate of the task to be executed. Using the generated first motion data and predicted state data including time-series data representing the state of the robot predicted based on the first motion data, the robot predicts whether the operation based on the first motion data will be successful or unsuccessful. If it is predicted that the robot's operation will fail, clustering based on similarity is performed on the multiple sets of motion data and state data and the first motion data and predicted state data, and the first motion data is modified using motion data that is classified as successful data from the multiple sets of motion data and state data and is of the same classification as the first motion data and predicted state data. Information processing device.

[0052] (Note 2) The aforementioned processor, From among the multiple sets of operation data and the operation data classified as success data among the state data, the operation data corresponding to the state data that includes the final state or initial state in which the difference from the target state or initial state of the operation to be executed is less than a threshold is extracted. The first operation data is generated based on the extracted operation data. The information processing device described in Appendix 1.

[0053] (Note 3) The aforementioned processor, If it is predicted that the robot's operation will fail, the process of modifying the first operation data so that it gradually approaches the operation data of the same classification will be repeated until it is predicted that the robot's operation based on the modified first operation data will be successful. The information processing device described in Appendix 1 or Appendix 2.

[0054] (Note 4) The aforementioned processor, Based on the difference between the state data and the target data, which is time-series data of the target state, a success or failure label is assigned to the pair of operation data and state data. In the aforementioned clustering, the assigned labels are used An information processing device as described in any one of the appendices 1 through 3.

[0055] 1. Robot System 10 Information Processing Devices 12 Robots 21 CPU 30 Information Processing Programs

Claims

1. Equipped with a processor, The aforementioned processor, By performing clustering on motion data, which is time-series data of motion commands given to the robot when the robot performs a task, and state data, which includes time-series data representing the state of the robot, it is possible to classify whether each of multiple sets of motion data and state data is success data or failure data for the task. Based on the aforementioned multiple sets of motion data and the motion data classified as success data among the state data, a first motion data is generated, which is time-series data of motion commands to the robot that satisfy conditions defined as conditions for increasing the success rate of the task to be executed. Using the generated first motion data and predicted state data including time-series data representing the state of the robot predicted based on the first motion data, the robot predicts whether the operation based on the first motion data will be successful or unsuccessful. If it is predicted that the robot's operation will fail, clustering based on similarity is performed on the multiple sets of motion data and state data and the first motion data and predicted state data, and the first motion data is modified using motion data that is classified as successful data from the multiple sets of motion data and state data and is of the same classification as the first motion data and predicted state data. Information processing device.

2. The aforementioned processor, From among the multiple sets of operation data and the operation data classified as success data among the state data, the operation data corresponding to the state data that includes the final state or initial state in which the difference from the target state or initial state of the operation to be executed is less than a threshold is extracted. The first operation data is generated based on the extracted operation data. The information processing apparatus according to claim 1.

3. The aforementioned processor, If it is predicted that the robot's operation will fail, the process of modifying the first operation data so that it gradually approaches the operation data of the same classification will be repeated until it is predicted that the robot's operation based on the modified first operation data will be successful. The information processing apparatus according to claim 1 or claim 2.

4. The aforementioned processor, Based on the difference between the state data and the target data, which is time-series data of the target state, a success or failure label is assigned to the pair of operation data and state data. In the aforementioned clustering, the assigned labels are used The information processing apparatus according to claim 1 or claim 2.

5. By performing clustering on motion data, which is time-series data of motion commands given to the robot when the robot performs a task, and state data, which includes time-series data representing the state of the robot, it is possible to classify whether each of multiple sets of motion data and state data is success data or failure data for the task. Based on the aforementioned multiple sets of motion data and the motion data classified as success data among the state data, a first motion data is generated, which is time-series data of motion commands to the robot that satisfy conditions defined as conditions for increasing the success rate of the task to be executed. Using the generated first motion data and predicted state data including time-series data representing the state of the robot predicted based on the first motion data, the robot predicts whether the operation based on the first motion data will be successful or unsuccessful. If it is predicted that the robot's operation will fail, clustering based on similarity is performed on the multiple sets of motion data and state data and the first motion data and predicted state data, and the first motion data is modified using motion data that is classified as successful data from the multiple sets of motion data and state data and is of the same classification as the first motion data and predicted state data. An information processing method in which a computer performs the processing.

6. By performing clustering on motion data, which is time-series data of motion commands given to the robot when the robot performs a task, and state data, which includes time-series data representing the state of the robot, it is possible to classify whether each of multiple sets of motion data and state data is success data or failure data for the task. Based on the aforementioned multiple sets of motion data and the motion data classified as success data among the state data, a first motion data is generated, which is time-series data of motion commands to the robot that satisfy conditions defined as conditions for increasing the success rate of the task to be executed. Using the generated first motion data and predicted state data including time-series data representing the state of the robot predicted based on the first motion data, the robot predicts whether the operation based on the first motion data will be successful or unsuccessful. If it is predicted that the robot's operation will fail, clustering based on similarity is performed on the multiple sets of motion data and state data and the first motion data and predicted state data, and the first motion data is modified using motion data that is classified as successful data from the multiple sets of motion data and state data and is of the same classification as the first motion data and predicted state data. An information processing program that instructs a computer to perform a task.

Images