Recommended load determination device, ability parameter estimation model learning device, method, and program
The recommended load determining device uses an estimation model to analyze user data and determine optimal exercise loads, addressing the challenge of skill improvement by distinguishing between load-induced changes and actual skill development.
Patent Information
- Application Number
- JP2020210721
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-12-18
- Publication Date
- 2024-11-13
- Estimated Expiration
- 2040-12-18
AI Technical Summary
Existing technologies struggle to determine an appropriate load that improves a user's skills during exercise tasks, as they cannot distinguish between changes due to load adjustments or skill improvement, leading to ineffective skill enhancement.
A recommended load determining device that utilizes an estimation model to calculate user ability based on load and operation data, determining optimal loads for skill improvement by analyzing motion and performance data.
Enables the determination of loads that effectively enhance user skills by identifying suitable exercise challenges, improving performance through targeted load adjustments.
Smart Images

Figure 0007581840000001 
Figure 0007581840000002 
Figure 0007581840000003
Abstract
Description
[Technical field]
[0001] The present invention relates to a recommended load determining device, a capacity parameter estimation model learning device, a recommended load determining method, and a recommended load determining program. [Background technology]
[0002] 2. Description of the Related Art Conventionally, there have been proposed techniques for automatically determining the control amount of a machine based on biometric information of a person when the person uses the machine to exercise, perform work, undergo rehabilitation, or the like.
[0003] For example, a control device has been proposed that controls the drive unit of a production line according to the level of proficiency of a task (see Patent Document 1). The control device described in Patent Document 1 has a memory unit for storing biometric information corresponding to each level of proficiency of the task. The control device also acquires biometric information measured from a worker, and compares the feature amount of the biometric information acquired during the task with the feature amount of the biometric information for each level of proficiency stored in the memory unit. Based on the result of the comparison, the control device then determines which level of proficiency the acquired biometric information corresponds to, and determines the control amount of the drive unit based on the determined level of proficiency.
[0004] Also, for example, a rehabilitation support device that enables appropriate difficulty level setting has been proposed (see Patent Document 2). The device described in Patent Document 2 includes a reception unit, an acquisition unit, a calculation unit, and an output control unit. The reception unit receives input of a first index value of brain activity measured by fMRI, a first achievement rate of a rehabilitation task, and a first difficulty level of the rehabilitation task for a subject who is the subject of rehabilitation. The acquisition unit acquires a correlation between a second index value of brain activity measured by fMRI, a second achievement rate of a rehabilitation task, and a second difficulty level of the rehabilitation task for a plurality of persons who have undergone rehabilitation. The calculation unit calculates a third difficulty level of the rehabilitation performed by the subject based on the input and the correlation, and the output control unit outputs the third difficulty level.
[0005] Also, a motor ability evaluation device has been proposed that evaluates motor ability without restricting the subject, rarely becomes unable to evaluate, and does not require a special physiological information sensor (see Patent Document 3). The device described in Patent Document 3 increases the load intensity of the exercise in a stepwise manner from P1 to P2, and this increase in load causes the rotation speed of the rotating machine caused by the exercise to decrease by ΔR from R1. Furthermore, this device calculates the time 1 / 2tf at which the rotation speed returns by 1 / 2 of ΔR as the subject subsequently exercises to increase the rotation speed in an attempt to return it to the original rotation speed R1, and evaluates motor ability based on this time 1 / 2tf. [Prior art documents] [Patent documents]
[0006] [Patent Document 1] JP 2018-55611 A [Patent Document 2] JP 2019-107235 A [Patent Document 3] Japanese Patent Application Publication No. 11-235401 Summary of the Invention [Problem to be solved by the invention]
[0007] For example, when a load is applied to a user to perform a certain motor task, it is desirable to apply an appropriate load so that the user's skill for the motor task improves. However, the technology described in Patent Document 1 cannot distinguish whether a change in biological information is due to a change in the load applied from the robot to the person or due to the user's improvement in skill. In other words, it cannot be grasped whether the user's skill has improved. Therefore, the technology described in Patent Document 1 has a problem in that it cannot apply an appropriate load so that the user's skill for a certain motor task improves.
[0008] Moreover, the technology described in Patent Document 2 is for appropriately adjusting the difficulty of rehabilitation tasks to suit the patient. Moreover, the technology described in Patent Document 3 is for evaluating motor ability. Therefore, as described above, both technologies have the problem that they cannot be applied when applying an appropriate load that will improve the user's skill in a certain motor task.
[0009] The present invention has been made in consideration of the above points, and has an object to determine a load that is beneficial for improving a user's skills when applying a load to the user to perform an exercise task. [Means for solving the problem]
[0010] In order to achieve the above-mentioned objective, the disclosed recommended load determination device is a recommended load determination device that uses an estimation model that outputs parameters related to the user's ability calculated based on load data representing the load applied to a user and action data representing the action of one action unit of the user under the load, and is composed of an acquisition unit that acquires load data representing the load applied to a target user for which a recommended load is to be determined, and action data representing the action of one action unit of the target user under the load, and a determination unit that determines recommended load data representing a recommended load for the target user based on parameters related to the ability of the target user output by the estimation model by inputting the acquired load data and action data into the estimation model.
[0011] The disclosed recommended load determination device is a recommended load determination device that uses an estimation model that outputs parameters related to a user's ability, calculated based on load data representing the load applied to a user and action data representing the action of one action unit of the user under the load, in which an acquisition unit acquires load data representing the load applied to a target user for whom a recommended load is to be determined, and action data representing the action of one action unit of the target user under the load, and a determination unit determines recommended load data representing a recommended load for the target user based on parameters related to the ability of the target user output by the estimation model by inputting the acquired load data and action data into the estimation model.
[0012] This makes it possible to determine a load that is beneficial for improving the user's skills when the user is made to perform an exercise task while being given a load.
[0013] The motion data may be displacement data of a specific part of the user or an implement held by the user when performing the motion, muscle activity data of the user, or differential or integral data of the displacement data or the muscle activity data. The displacement data of the specific part of the user may be time-series data of skeletal information of the user.
[0014] In addition, the ability-related parameters output by the estimation model may be limit load data representing the load value that is the greatest burden on the user within a range of load values for one load item at which the likelihood of success for the action is greater than a predetermined standard, and the determination unit may determine the limit load data for the target user as the recommended load data.
[0015] In addition, the ability parameters output by the estimation model may be parameters representing the relationship between the load and the likelihood of success for the action, and the determination unit may determine the recommended load data based on the ability parameters for the target user.
[0016] In addition, the parameter representing the relationship between the load for the target user and the likelihood of success for the operation may be a parameter that specifies a graph representing the relationship between the load value for one load item and the likelihood of success, and the determination unit may determine as the recommended load data limit load data that represents the load value that is the greatest burden on the target user within a range of load values in the graph where the likelihood of success is greater than a predetermined standard.
[0017] In addition, the parameter representing the relationship between the load for the target user and the magnitude of the likelihood of success for the action may be a parameter that identifies a heat map that represents the relationship between the load values for multiple load items and the magnitude of the likelihood of success, and the determination unit may determine as the recommended load data a set of load values in the heat map where the likelihood of success is greater than a predetermined standard.
[0018] The determination unit may output data for visualizing and displaying the heat map.
[0019] The determination unit may be capable of using a comparison heat map, and in comparison with the comparison heat map, create a relative goodness map that identifies a relative goodness range, which is a range of pairs of load values where the target user's likelihood of success is greater, or a relative poorness map that identifies a relative poorness range, which is a range of pairs of load values where the target user's likelihood of success is smaller, and determine the set of load values within the relative goodness range or the relative poorness range as the recommended load data.
[0020] In addition, the comparison heat map may be prepared for each level of the user, and the determination unit may use the comparison heat map for a user at a level equal to or higher than the level of the target user and closest to the level of the target user.
[0021] In addition, the acquisition unit may further acquire a designation of strengths or weaknesses for the target user, and the determination unit may create the relative strengths map when the designation of strengths is acquired, and create the relative weaknesses map when the designation of weaknesses is acquired.
[0022] The determination unit may output data for visualizing and displaying at least one of the relative strengths map and the relative weakness map.
[0023] In addition, the estimation model may be prepared for each type of movement, and the acquisition unit may acquire a designation of one of the types from the target user, and the determination unit may determine the recommended load data using the estimation model corresponding to the designated type.
[0024] In addition, the acquisition unit may acquire multiple sets of the load data and the operation data, and the determination unit may input load data obtained by averaging the multiple sets of the load data and operation data obtained by averaging the multiple sets of the operation data into the estimation model.
[0025] In addition, the acquisition unit may acquire multiple sets of the load data and the operation data, and the determination unit may average parameters related to the capacity obtained by inputting each of the sets of the load data and the operation data into the estimation model, and determine the recommended load data based on the averaged parameters.
[0026] In addition, the acquisition unit may acquire multiple sets of the load data and the operation data, and the determination unit may calculate multiple recommended load data based on parameters related to each of the capabilities obtained by inputting each of the sets of the load data and the operation data into the estimation model, and further may average the multiple recommended load data to determine the final recommended load data.
[0027] In addition, the acquisition unit may acquire a set of the load data and the operation data and actual measured parameters related to the corresponding actual operation results, and the determination unit may determine the recommended load data so as to reduce the burden on the target user if the ability of the target user evaluated by the acquired actual measured parameters is lower than the ability of the target user evaluated by the parameters which are the output of the estimation model.
[0028] The determination unit may output the recommended load data to a load application device for controlling the load application device that applies the load to the target user, and may output data for displaying the latest recommended load data at a predetermined time interval.
[0029] In addition, the disclosed recommended load determination device may be configured to include a learning unit that generates the estimation model by learning the correspondence between the load data and the operation data and parameters related to the user's ability calculated based on the load data and the operation data.
[0030] The learning unit may generate the estimation model using the operation data compressed so that a difference between a number of dimensions of the load data and a number of dimensions of the operation data falls within a predetermined range.
[0031] The disclosed ability parameter estimation model learning device includes an acquisition unit that acquires load data representing a load applied to a user, action data indicating the action of one action unit of the user under the load, and performance data associated with the action, and a learning unit that learns the correspondence between the load data and the action data, and parameters related to the ability of the user calculated based on the load data and the performance data, thereby generating an estimation model that outputs parameters related to the ability of the target user in response to input of the load data and the action data for the target user for whom a recommended load is to be determined.
[0032] The disclosed recommended load determination method is a method in which an estimation model is prepared that outputs parameters related to the ability of a user, the parameters being calculated based on load data representing the load applied to a user and action data representing the action of one action unit of the user under the load, an acquisition unit acquires load data representing the load applied to a target user for whom a recommended load is to be determined, and action data representing the action of one action unit of the target user under the load, and a determination unit inputs the acquired load data and action data into the estimation model, thereby determining recommended load data representing a recommended load for the target user based on the parameters related to the ability of the target user output by the estimation model.
[0033] The disclosed recommended load determination program is a program for causing a computer to function as a recommended load determination device that uses an estimation model that outputs parameters related to the user's ability, calculated based on load data representing the load applied to a user and action data representing the action of one unit of action of the user under that load, and causes the computer to function as an acquisition unit that acquires load data representing the load applied to a target user for whom a recommended load is to be determined, and action data representing the action of one unit of action of the target user under that load, and a determination unit that determines recommended load data representing a recommended load for the target user based on parameters related to the ability of the target user output by the estimation model by inputting the acquired load data and action data into the estimation model. Effect of the Invention
[0034] According to the disclosed recommended load determination device, ability parameter estimation model learning device, method, and program, when a user is given a load to perform an exercise task, a load that is beneficial for improving the user's skills can be determined. [Brief description of the drawings]
[0035] [Figure 1] 1 is a functional block diagram of a recommended load determining device according to a first embodiment. [Diagram 2] FIG. 4 is a diagram illustrating an example of a learning data DB in the first embodiment. [Diagram 3] 2 is a block diagram showing a hardware configuration of a recommended load determining device. [Figure 4] 5 is a flowchart showing an example of a learning mode process in the first embodiment. [Diagram 5] 13 is a flowchart showing an example of a practice mode process in the first embodiment. [Figure 6] FIG. 4 is a functional block diagram of a recommended load determining device according to second and third embodiments. [Figure 7] FIG. 1 is an external view of a table tennis robot. [Figure 8] FIG. 11 is a diagram illustrating an example of a learning data DB in the second embodiment. [Figure 9] 13 is a graph showing the relationship between ball speed and return rate. [Figure 10] 13 is a flowchart illustrating an example of a recommended load determination process in the second embodiment. [Figure 11] FIG. 13 is a diagram showing another example of a learning data DB. [Figure 12] FIG. 13 is a diagram showing another example of a learning data DB. [Figure 13] FIG. 13 is a diagram illustrating an example of a learning data DB in the third embodiment. [Figure 14] FIG. 13 is a diagram illustrating an example of an estimation model in the third embodiment. [Figure 15] FIG. 11 is a diagram for explaining determination of recommended load data based on a heat map. [Figure 16] 13 is a flowchart illustrating an example of a recommended load determination process in the third embodiment. [Figure 17] FIG. 13 is a functional block diagram of a recommended load determining device according to a fourth embodiment. [Figure 18] FIG. 13 is a diagram for explaining a comparative heat map. [Figure 19] FIG. 13 is a diagram for explaining a relative strength and weakness map. [Figure 20]FIG. 11 is a diagram for explaining the determination of recommended load data based on a relative strength / weakness map. [Figure 21] FIG. 11 is a diagram for explaining the determination of recommended load data based on a relative strength / weakness map. [Figure 22] 13 is a flowchart showing an example of a comparison heat map creation process in the fourth embodiment. [Figure 23] 13 is a flowchart showing an example of a decision process in the fourth embodiment. [Figure 24] FIG. 13 is a functional block diagram of a recommended load determining device according to a fifth embodiment. [Diagram 25] FIG. 13 is a diagram illustrating an example of an estimation model in the fifth embodiment. [Figure 26] FIG. 13 is a functional block diagram of a recommended load determining device according to a sixth embodiment. [Figure 27] FIG. 13 is a diagram illustrating an example of a performance DB. [Figure 28] FIG. 13 is a diagram for explaining a direction in which a load decreases. [Figure 29] FIG. 13 is a diagram for explaining a direction in which a load increases. [Diagram 30] 23 is a flowchart showing an example of a decision process in the sixth embodiment. [Diagram 31] FIG. 11 is a diagram for explaining a sudden change in load. [Diagram 32] FIG. 13 is a diagram for explaining averaging of inputs to an estimation model. [Diagram 33] FIG. 13 is a diagram for explaining averaging of outputs from an estimation model. [Diagram 34] FIG. 13 is a diagram for explaining dimensional compression of motion data. [Diagram 35] FIG. 11 is a diagram for explaining an example of dimensional compression of action data. [Diagram 36] FIG. 11 is a diagram for explaining an example of dimensional compression of action data. [Figure 37] FIG. 11 is a diagram for explaining an example of dimensional compression of action data. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0036] An example of an embodiment of the disclosed technology will be described below with reference to the drawings. In each of the following embodiments, a recommended load determination device will be described that determines a load to be applied to a user when the user performs an action according to a predetermined exercise task against a load applied by a machine. The exercise task is, for example, improving the return rate in a table tennis rally, improving the hit rate in baseball batting, improving walking in rehabilitation, etc. In this case, the load application device, which is a machine that applies a load to the user, can be a table tennis robot that returns or sends a table tennis ball, a batting machine, a treadmill, etc.
[0037] In the following embodiments, when the entity of an estimation model is data that specifies processing content, the expressions "input to the model" and "output from the model" refer to input to a functional means (model execution unit) that is a computer or the like in which the model is implemented, and output from the same means. When the entity of an estimation model is data that specifies processing content, for example, the entity of the estimation model is structure data and connection weight data that specify a neural network.
[0038] In addition, the same reference numerals are used in each drawing to refer to the same or equivalent components and parts. Also, the dimensions and ratios of the drawings are exaggerated for the convenience of explanation and may differ from the actual ratios.
[0039] First Embodiment Fig. 1 is a block diagram showing the functional configuration of a recommended load determination device 10 according to the first embodiment. As shown in Fig. 1, the recommended load determination device 10 includes an acquisition unit 12, a learning data DB (database) 14, a learning unit 16, an estimation model 18, and a determination unit 20. The recommended load determination device 10 operates in two modes: a learning mode for learning the estimation model 18, and a practice mode for determining a recommended load using the learned estimation model 18.
[0040] The acquisition unit 12 acquires load data representing a load applied to a sample user in the learning mode. The sample user is a user who performs an exercise task to collect learning data for learning the estimation model 18. For example, when the load application device is a table tennis robot or a pitching machine, the load data is the ball speed, rotation, direction, etc. of a hit or pitched ball sent from the load application device. Also, when the load application device is a treadmill, the load data is the speed, inclination angle, etc. of a running belt. For example, the acquisition unit 12 can acquire the load data by acquiring control information related to the load, such as the torque of a motor, from the load application device, or by measuring the load actually applied to the sample user.
[0041] The value that specifies the load to be applied to the user may include multiple items such as the above-mentioned "ball speed, rotation, direction, etc." The acquisition unit 12 may acquire load data for one item out of the multiple items, or may acquire load data for multiple items, depending on the item that is changed when applying the load, i.e., the item that is determined as the recommended load.
[0042] Moreover, in the learning mode, the acquisition unit 12 acquires motion data representing one motion unit of the sample user under load. For example, in the case of an exercise task such as batting in table tennis or baseball, one motion unit is a motion indicating one swing, and in the case of walking training, it is a walking motion of a predetermined number of steps (for example, one or two steps). The motion data is, for example, displacement data of a specific part of the sample user or an implement held by the sample user when performing a motion, muscle activity data of the sample user, or differential data or integral data thereof. The acquisition unit 12 can acquire the motion data by performing image analysis of an image captured by a camera of the sample user performing a motion, or by performing data processing of an output value from a myoelectric potential sensor attached to the sample user.
[0043] Moreover, in the learning mode, the acquisition unit 12 acquires performance data associated with the sample user's movements under load. The performance data is a return rate in the case of a table tennis motor task, a hit rate in the case of a baseball batting motor task, etc. The acquisition unit 12 can acquire such performance data, for example, by performing image analysis on an image captured by a camera or by data processing on output values from various sensors. In the learning mode, the acquisition unit 12 may also acquire performance data that has been manually input.
[0044] The acquisition unit 12 stores the load data, movement data, and performance data for each sample user acquired in the learning mode in a learning data DB 14 as shown in FIG. 2, for example. In the example of FIG. 2, "user No." is an identification number that can individually identify the sample user. Also, "attempt No." is a number indicating which attempt each piece of data is for, with an attempt involving one action unit being considered as one attempt for each sample user. The example of FIG. 2 shows an example in which load data, movement data, and performance data for 100 attempts are stored for each sample user. The "ability parameters" in FIG. 2 will be described later.
[0045] In addition, in the practice mode, the acquisition unit 12 acquires load data representing the load applied to the target user for which the recommended load is determined, and action data representing one action unit of the target user under that load. Details of the load data and action data are similar to those of the load data and action data acquired in the learning mode described above. Note that when describing the sample user and the target user without distinction, they will simply be referred to as "users."
[0046] The learning unit 16 is a functional unit that functions in the learning mode. The learning unit 16 calculates a parameter (hereinafter referred to as "ability parameter") related to the ability of each sample user for the exercise task based on the load data and performance data stored in the learning data DB 14. The ability parameter can be, for example, limit load data that represents the maximum load value for the sample user within the range of load values for one load item in which the success probability indicated by the performance data for the movement is greater than a predetermined standard. The ability parameter may also be a parameter that represents the relationship between the load and the magnitude of the success probability for the movement. This parameter can also be a parameter that specifies a graph that represents the relationship between the load value for one load item and the magnitude of the success probability. This parameter can also be a parameter that specifies a heat map that represents the relationship between the load value for multiple load items and the magnitude of the success probability. Here, the "success probability" is a parameter that represents the magnitude of the possibility of success, and corresponds to, for example, a predicted success rate and a predicted performance value. An example of the performance value is the magnitude of the error in the landing position in an embodiment for table tennis described later. In the case of the error in the landing position, the smaller the predicted error, the greater the success probability. In addition, the "burden" on the user refers to the degree of difficulty for the user, and the greater the burden, the smaller the likelihood of success. For example, in the case of an embodiment of table tennis, the greater the ball speed (load) of the ball thrown by the table tennis robot, the greater the burden and the smaller the likelihood of success.
[0047] The learning unit 16 stores the ability parameters calculated for each sample user in the "ability parameter" column of the learning data DB 14. Each pair of the load data, the operation data, and the ability parameters stored in the learning data DB 14 becomes each learning data for learning the estimation model 18. The learning unit 16 learns the correspondence between the load data, the operation data, and the ability parameters, thereby generating the estimation model 18 that estimates and outputs the ability parameters of a target user for which a recommended load is to be determined in response to input of load data and operation data for the target user. The estimation model 18 is composed of, for example, a neural network.
[0048] The determination unit 20 is a functional unit that functions in the practice mode. The determination unit 20 inputs the load data and the motion data of the target user acquired by the acquisition unit 12 to the estimation model 18. Then, the determination unit 20 determines recommended load data representing a recommended load for the target user based on the ability parameters of the target user output by the estimation model 18. As described above, when the limit load data is output from the estimation model 18 as the ability parameters, the determination unit 20 determines the limit load data as the recommended load data. Also, as described above, when a parameter representing the relationship between the load and the likelihood of success for the motion is output as the ability parameter, the determination unit 20 determines the recommended load data based on the ability parameters.
[0049] For example, the determination unit 20 can determine, as the recommended load data, limit load data that represents the load value that is the greatest burden on the target user within a range of load values where the likelihood of success is greater than a predetermined standard in a graph specified by the ability parameters. Also, the determination unit 20 can determine, as the recommended load data, a set of load values where the likelihood of success is greater than a predetermined standard in a heat map specified by the ability parameters. The determination unit 20 outputs the determined recommended load data. The output recommended load data may be used as control information for controlling a load application device, or may be displayed on a display device to be presented to the target user and the target user's instructor, etc.
[0050] Fig. 3 is a block diagram showing the hardware configuration of the recommended load determination device 10 according to the first embodiment. As shown in Fig. 3, the recommended load determination device 10 has a CPU (Central Processing Unit) 32, a memory 34, a storage device 36, an input device 38, an output device 40, a storage medium reading device 42, and a communication I / F (Interface) 44. Each component is connected to each other via a bus 46 so as to be able to communicate with each other.
[0051] The storage device 36 stores a recommended load determination program for executing each of the learning mode processing and practice mode processing described below. The CPU 32 is a central processing unit, and executes various programs and controls each component. That is, the CPU 32 reads the programs from the storage device 36 and executes the programs using the memory 34 as a working area. The CPU 32 controls each of the components and performs various calculation processes according to the programs stored in the storage device 36.
[0052] The memory 34 is made up of a RAM (Random Access Memory) and temporarily stores programs and data as a working area. The storage device 36 is made up of a ROM (Read Only Memory), HDD (Hard Disk Drive), SSD (Solid State Drive), etc., and stores various programs including an operating system, and various data.
[0053] The input device 38 is a device for performing various inputs, such as a keyboard or a mouse. The output device 40 is a device for outputting various information, such as a display or a printer. A touch panel display may be adopted as the output device 40 to function as the input device 38. The storage medium reader 42 reads data stored in various storage media, such as a CD (Compact Disc)-ROM, a DVD (Digital Versatile Disc)-ROM, a Blu-ray disc, and a USB (Universal Serial Bus) memory, and writes data to the storage media.
[0054] The communication I / F 44 is an interface for communicating with other devices, and uses standards such as Ethernet (registered trademark), FDDI, and Wi-Fi (registered trademark).
[0055] Next, the operation of the recommended load determining device 10 according to the first embodiment will be described.
[0056] Fig. 4 is a flowchart showing the flow of the learning mode process executed by the CPU 32 of the recommended load determination device 10. When the learning mode is selected, the CPU 32 reads out a recommended load determination program from the storage device 36, loads it into the memory 34, and executes it, whereby the CPU 32 functions as each functional component of the recommended load determination device 10, and the learning mode process shown in Fig. 4 is executed.
[0057] In step S10, the acquisition unit 12 acquires the load data, the action data, and the performance data for each trial for each of the multiple sample users, and stores them in the learning data DB 14.
[0058] Next, in step S12, the learning unit 16 calculates ability parameters from the load data and performance data for each sample user based on the data stored in the learning data DB , and stores the calculated ability parameters in the learning data DB .
[0059] Next, in step S14, the learning unit 16 learns the correspondence between the load data, operation data, and ability parameters, thereby generating an estimation model 18 that outputs the ability parameters of the target user in response to input of load data and operation data about the target user. The learning unit 16 stores the generated estimation model 18 in a predetermined storage area of the recommended load determination device 10, and the learning mode process ends.
[0060] Fig. 5 is a flowchart showing the flow of practice mode processing executed by the CPU 32 of the recommended load determination device 10. When the practice mode is selected, the CPU 32 reads a recommended load determination program from the storage device 36, deploys it in the memory 34, and executes it, whereby the CPU 32 functions as each functional component of the recommended load determination device 10, and the practice mode processing shown in Fig. 5 is executed. Note that the practice mode processing is an example of the disclosed recommended load determination method.
[0061] In step S20, the acquisition unit 12 acquires load data and operation data for the target user during the current trial. Next, in step S22, the determination unit 20 inputs the load data and operation data for the target user acquired by the acquisition unit 12 into the estimation model 18, and estimates the ability parameters of the target user.
[0062] Next, in step S24, the determination unit 20 determines and outputs recommended load data representing a recommended load for the target user's next attempt based on the ability parameters of the target user estimated and output by the estimation model 18, and the practice mode processing is then terminated.
[0063] By repeatedly executing the practice mode process for each trial, the recommended load is repeatedly determined, so that the target user can continuously perform the exercise task.
[0064] As described above, the recommended load determination device according to the first embodiment acquires load data representing a load applied to a target user and motion data representing a motion of one motion unit of the target user under that load. The recommended load determination device also estimates the ability parameters of the target user by inputting the acquired load data and motion data into an estimation model that estimates and outputs the ability parameters of the user in response to the input of the load data and motion data. The recommended load determination device then determines recommended load data representing a recommended load for the target user based on the estimated ability parameters of the target user. This makes it possible to determine a load that is beneficial to improving the user's skills when applying a load to the user to perform an exercise task.
[0065] The first embodiment may be modified as follows. The recommended load determination device 10 according to the first embodiment has a learning mode and a practice mode, but the recommended load determination device 10 may have only the practice mode and not the learning mode. In this case, the recommended load determination device 10 does not have the learning data DB 14 and the learning unit 16 required for learning. The acquisition unit 12 does not acquire load data, operation data, and performance data of the sample user.
[0066] The recommended load determiner 10 according to the first embodiment may not include the learning unit 16, but may use a learning function provided externally, such as a server on a network, via communication. Also, the recommended load determiner 10 may not include the learning data DB 14, but may transmit data to be recorded in an external learning data DB.
[0067] The recommended load determination device 10 according to the first embodiment may acquire and use the trained estimation model 18 from an external device. Also, the recommended load determination device 10 may use an external trained estimation model via communication.
[0068] The above modifications of the first embodiment can be similarly applied to the second and subsequent embodiments.
[0069] <Second embodiment> Next, a second embodiment will be described. In the second embodiment, the exercise task is to improve the return rate in a table tennis rally, the load given to the user is to send or return the ball from the table tennis robot, and the user's motion is a swing to return the ball hit by the table tennis robot.
[0070] 6 is a block diagram showing the functional configuration of a recommended load determination device 210 according to the second embodiment. As shown in FIG. 6, the recommended load determination device 210 is connected to each of the table tennis robot 60, the sensor 70, and the sensor 72.
[0071] 7 shows an external view of the table tennis robot 60. The table tennis robot 60 is a robot that hits back a table tennis ball BL, which is hit by a user (not shown) and bounces off a table tennis table TB, with a racket RK as a hitting tool. Note that this embodiment focuses on the load applied to the user, that is, sending or returning the ball BL with a specific load, and therefore a detailed description of the function of recognizing a return shot from the user will be omitted.
[0072] As shown in FIG. 7, the table tennis robot 60 includes a movable part 62 for supporting and moving the racket RK. In the example of FIG. 7, the movable part 62 includes three robot arms 64A, 64B, and 64C, and a support part 66 for the racket RK. That is, the movable part 62 is a so-called parallel link robot, and is a robot that controls the behavior of the racket RK supported by the support part 66 at the tip of the robot arms 64A, 64B, and 64C by controlling the robot arms 64A, 64B, and 64C in parallel. The support part 66 includes a drive mechanism for changing the attitude of the racket RK. Note that the movable part 62 is not limited to a parallel link robot, and may be another type of robot. The table tennis robot 60 controls the drive mechanism by a control part (not shown) so as to return the ball BL at a hitting speed specified as control information. The "hitting speed" is an example of load data.
[0073] The table tennis robot 60 is also provided with cameras 74A, 74B, and 74C. The cameras 74A and 74B are installed at positions where they can capture the user from different angles, and capture the user at a predetermined frame rate. The cameras 74A and 74B function as so-called stereo cameras. The sensor 70 is composed of the cameras 74A and 74B and an image analysis unit (not shown). The sensor 70 outputs skeletal information of the user by performing image analysis on the images captured by the cameras 74A and 74B for each frame. The skeletal information is time-series data of the three-dimensional positions (X, Y, Z) of each joint (joint 1 to joint n) of the user. The joints included in the skeletal information may be limited to joints necessary for identifying a swing motion, for example, the shoulder, elbow, and wrist of the dominant arm. In addition to the user's joints, the skeletal information may also include the three-dimensional position of the center of the face of the racket RK held by the user. The "skeletal information" is an example of motion data.
[0074] The camera 74C is installed in a position where it can capture an image of the surface of the table TB. The sensor 72 is composed of the camera 74C and an image analysis unit (not shown). The sensor 72 detects the landing position of the user's return shot on the table TB by analyzing the image captured by the camera 74C, and calculates and outputs the error between a preset position (e.g., the center of the table TB) and the landing position (hereinafter referred to as "return shot error"). The "return shot error" is an example of performance data.
[0075] As shown in FIG. 6, the recommended load determiner 210 functionally includes an acquisition unit 212, a learning data DB 214, a learning unit 216, an estimation model 218, and a determination unit 220.
[0076] In the learning mode, the acquisition unit 212 acquires from the table tennis robot 60 the speed of the ball sent (returned) from the table tennis robot 60 as load data indicating the load applied to the sample user. For example, the acquisition unit 212 can acquire the ball speed from control information set for controlling the drive mechanism of the table tennis robot 60 or the analysis results of images captured by the cameras 74A and 74B. The acquisition unit 212 may also measure the speed of the ball sent from the table tennis robot 60 by a sensor (not shown) and acquire the ball speed indicated by the measurement value output from the sensor.
[0077] In addition, in the learning mode, the acquisition unit 212 acquires, as the motion data of the sample user, skeletal information of the sample user output by the sensor 70. Furthermore, in the learning mode, the acquisition unit 212 acquires, as the performance data associated with the motions of the sample user, the return ball error output by the sensor 72.
[0078] The acquisition unit 212 stores the acquired hit ball speed, skeletal information, and return shot error in the learning data DB 214. Fig. 8 shows an example of the learning data DB 214. The items stored in the learning data DB 214 are the same as those in the learning data DB 14 in the first embodiment.
[0079] In addition, in the practice mode, the acquisition unit 212 acquires the ball speed, which is load data representing the load applied to the target user, and the skeletal information, which is motion data representing the return swing motion of the target user to the ball. The details of the method of acquiring the ball speed and the skeletal information are the same as those in the learning mode. Note that the acquisition unit 212 may acquire the ball speed indicated by the recommended load determined by the determination unit 220, which will be described later, as the ball speed for the target user.
[0080] The learning unit 216 calculates the maximum possible return ball speed, which is an example of limit load data, as an ability parameter for each sample user. Specifically, the learning unit 216 calculates the return ball rate by the sample user for each hitting ball speed from the learning data for each sample user stored in the learning data DB 214. More specifically, the learning unit 216 determines a return ball error to be a predetermined value (e.g., 75 cm) or less as a success, and calculates the ratio of the number of attempts in which the return ball is successful to the total number of attempts at a hitting ball speed v±α [m / sec] as the return ball rate at that hitting ball speed v. For example, for the sample user User No. 1, assume that the number of attempts in which the hitting ball speed was in the range of 2.5 to 3.5 m / sec was 10, and the number of attempts in which the return ball was successful was 5. In this case, the learning unit 216 calculates the return ball rate for the sample user User No. 1 at a hitting ball speed of 3 m / sec to be 50%. The learning unit 216 calculates the return ball rate for each hitting ball speed in the same manner. Then, the learning unit 216 creates a graph showing the relationship between the hitting ball speed and the return ball percentage as shown in Fig. 9, and from this graph, determines the maximum hitting ball speed at which the return ball percentage is equal to or less than a predetermined value (e.g., 85%) as the maximum possible return ball speed. The learning unit 216 stores the maximum possible return ball speed determined for each sample user in the learning data DB 214 as shown in Fig. 8.
[0081] The learning unit 216 learns the correspondence between the ball speed and skeletal information of the sample user and the ability parameters, thereby generating an estimation model 218 that estimates and outputs the maximum possible return ball speed, which is an ability parameter of the target user, in response to the input of the ball speed and skeletal information of the target user.
[0082] The determining unit 220 inputs the ball speed and skeletal information about the target user acquired by the acquiring unit 212 to the estimation model 218. Then, the determining unit 220 determines the maximum possible return ball speed output by the estimation model 218 as recommended load data for the target user, and outputs it to the table tennis robot 60. As a result, the table tennis robot 60 sets the recommended load data as control information, and is controlled to send (return) a ball at the ball speed indicated by the recommended load data at the next attempt.
[0083] The hardware configuration of the recommended load determination device 210 is similar to that of the recommended load determination device 10 according to the first embodiment shown in Fig. 3, and therefore a description thereof will be omitted. Note that the storage device 36 stores a recommended load determination program for executing a recommended load determination process, which will be described later.
[0084] Next, the operation of the recommended load determining device 210 according to the second embodiment will be described.
[0085] Fig. 10 is a flowchart showing the flow of a recommended load determination process executed by the CPU 32 of the recommended load determination device 210. The CPU 32 reads out a recommended load determination program from the storage device 36, deploys it in the memory 34, and executes it, whereby the CPU 32 functions as each functional component of the recommended load determination device 210, and the recommended load determination process shown in Fig. 10 is executed. Note that the recommended load determination process is an example of the disclosed recommended load determination method.
[0086] In step S200, the acquisition unit 212 determines whether the selected mode is the learning mode or not. If the learning mode is selected, the process proceeds to step S202, and if the practice mode is selected, the process proceeds to step S214.
[0087] In step S202, the acquisition unit 212 instructs the table tennis robot 60 to send (return) a ball. The control information may be set in the table tennis robot 60 so that the ball speed of the ball sent in this step is changed stepwise from the initial value for each step for the same sample user.
[0088] Next, in step S204, the acquisition unit 212 acquires the ball speed of the ball sent in response to the instruction in step S202 above, and skeletal information indicating the sample user's swing when returning the ball, and a return shot error, and stores these in the learning data DB 214.
[0089] Next, in step S206, the acquisition unit 212 determines whether the number of trials for the sample user is equal to or greater than a predetermined number (e.g., 100 times). This predetermined number may be determined in advance as a number of trials required as the number of learning data for each sample user. If the number of trials is equal to or greater than the predetermined number, the process proceeds to step S208, and if it is less than the predetermined number, the process returns to step S202.
[0090] In step S208, the learning unit 216 calculates the return rate for each hitting ball speed from the learning data of the corresponding sample user stored in the learning data DB 214. Then, the learning unit 216 determines the hitting ball speed at which the return rate is equal to or lower than a predetermined value (e.g., 85%) from a graph showing the relationship between the hitting ball speed and the return rate as the maximum possible return ball speed, and stores this in the learning data DB 214.
[0091] Next, in step S210, the learning unit 216 determines whether or not the learning data DB 214 stores enough learning data for the number of sample users for which the estimation model 218 can be generated. If the estimation model 218 can be generated, the process proceeds to step S212, and if sufficient learning data for generating the estimation model 218 is not stored, the recommended load determination process ends.
[0092] In step S212, the learning unit 216 generates an estimation model 218 by learning the correspondence between the ball speed and skeletal information of the sample user and the ability parameters, and the recommended load determination process ends.
[0093] On the other hand, if the practice mode is selected and the process proceeds to step S214, the determination unit 220 determines whether or not the estimation model 218 has been generated. If the estimation model 218 has been generated, the process proceeds to step S216, and if not, the process proceeds to step S202.
[0094] In step S216, the acquisition unit 212 instructs the table tennis robot 60 to set the ball speed to an initial value and start sending the ball. Next, in step S218, the acquisition unit 212 acquires the ball speed of the ball sent from the table tennis robot 60 in response to the instruction in step S216, and skeletal information indicating the swing of the target user when returning the ball.
[0095] Next, in step S220, the determining unit 220 inputs the hit ball speed and skeletal information about the target user acquired by the acquiring unit 212 into the estimation model 218, and estimates the maximum possible return ball speed of the target user.
[0096] Next, in step S222, the determination unit 220 determines the maximum possible return ball speed estimated in step S220 as recommended load data for the target user, and instructs the table tennis robot 60 to return the ball at the maximum possible return ball speed indicated by the recommended load data.
[0097] Next, in step S224, the determination unit 220 determines whether or not to end the motor task. For example, if there is no return shot from the target user or if a command indicating the end of the motor task is input, it is determined that the motor task is to be ended, and the recommended load determination process ends. On the other hand, if the motor task is not to be ended, the process returns to step S218, and the process of acquiring the ball speed of the ball sent (returned) from the table tennis robot 60 in response to the instruction in step S222 above, and skeletal information indicating the swing of the target user when returning the ball is repeated.
[0098] As described above, in the recommended load determination device of the second embodiment, in the exercise task of improving return rate using a table tennis robot, the ball is hit as fast as possible within the range that the target user can return, thereby providing a load that is beneficial to improving the skills of the target user.
[0099] In the second embodiment, an example of determining a recommended load for an exercise task of improving a return rate using a table tennis robot has been described, but the present invention is not limited to this. Exercise tasks of improving a hitting rate using a batting machine and exercise tasks of improving walking in rehabilitation using a treadmill, as exemplified in the first embodiment, can also be realized as specific embodiments similar to the second embodiment.
[0100] FIG. 11 shows an example of the learning data DB214A in the case of an exercise task for improving hitting percentage using a batting machine. In this case, the load data is the pitching speed of the ball sent from the batting machine. Also, the motion data is skeletal information showing the swing of the user when batting the pitched ball. Also, the performance data is the batting result (hit or grounder). Then, similar to the graph shown in FIG. 9, from the relationship between the pitching speed and the hitting percentage, the pitching speed at which the hitting percentage is a predetermined value (for example, 85%) or less can be obtained as the maximum hitable ball speed, and this can be used as the ability parameter. From the learning data shown in FIG. 11, an estimation model can be generated that estimates the fastest ball speed possible as recommended load data within the range at which the target user can hit.
[0101] FIG. 12 shows an example of the learning data DB214B in the case of an exercise task for improving walking in rehabilitation using a treadmill. In this case, the speed of the running belt of the treadmill is acquired as the load data. Furthermore, skeletal information of the user walking on the running belt is acquired as the motion data. Furthermore, for example, the variation of the center of gravity during one motion is acquired as the performance data based on the skeletal information. Note that the performance data is not limited to the variation of the center of gravity, and it is sufficient if it is an index capable of evaluating a motion in which walking can be performed without falling. Then, similarly to the graph shown in FIG. 9, from the relationship between the speed of the running belt and the variation of the center of gravity, the speed of the running belt at which the variation of the center of gravity is equal to or greater than a predetermined value can be obtained as the maximum walkable speed, and this can be used as the ability parameter. Note that the predetermined value may be determined in advance as a value at which the variation of the center of gravity beyond that value may cause a fall. From the learning data shown in FIG. 12, an estimation model can be generated that estimates the fastest possible running belt speed as recommended load data within a range in which the target user can walk without falling.
[0102] In the second embodiment, the case where the determined recommended load data is output to the table tennis robot, that is, the case where the data is output to the load application device to control the load application device that applies a load to the target user, has been described, but the present invention is not limited to this. For example, data for displaying the latest recommended load data on the display device at a predetermined time interval may be output to the display device. For example, the determined recommended load data may be presented to an instructor such as a trainer or physical therapist who provides instruction to the target user, and control information related to the load may be set in the load application device under the instructor's judgment. This allows the safety of the target user to be taken into consideration, particularly when the exercise task is related to rehabilitation.
[0103] In the second embodiment, the maximum possible return ball speed, which is an example of limit load data, is estimated as the ability parameter, but the present invention is not limited to this. For example, a parameter specifying a graph such as that shown in Fig. 9 may be estimated as the ability parameter, and recommended load data may be determined based on this graph. For example, recommended load data may be determined randomly from loads at which the return ball rate falls within a predetermined range (e.g., 80% to 85%). Alternatively, a statistical value such as the average, median, or mode of loads at which the return ball rate falls within a predetermined range may be determined as the recommended load data.
[0104] <Third embodiment> Next, a third embodiment will be described. In the third embodiment, as in the second embodiment, the exercise task is to improve the return rate in a table tennis rally, the load given to the user is a return shot from a table tennis robot, and the user's motion is a swing to return the ball hit by the table tennis robot.
[0105] As shown in FIG. 6, the recommended load determiner 310 according to the third embodiment functionally includes an acquisition unit 312, a learning data DB 314, a learning unit 316, an estimation model 318, and a determination unit 320.
[0106] The acquisition unit 312, like the acquisition unit 212 in the second embodiment, acquires load data, movement data, and performance data in the learning mode, and acquires load data and movement data in the practice mode. The movement data is skeletal information of the user like the second embodiment, and the performance data is return shot error like the second embodiment.
[0107] In the third embodiment, the load is made multidimensional, and the acquisition unit 312 acquires load data of multiple items. For example, the acquisition unit 312 acquires, as load data of multiple items, a ball rotation indicating the rotation speed of the ball, and a relative position, in addition to the ball speed as in the second embodiment. The relative position is the relative position between the landing position of the ball BL returned from the table tennis robot 60 on the table tennis table TB and the position of the user. The acquisition unit 312 can acquire these load data from control information set in the table tennis robot 60. Moreover, the load data may be acquired by detecting the landing position of the ball BL, the position of the user, the ball rotation, etc., using a sensor (not shown).
[0108] The acquisition unit 312 stores the acquired ball speed, ball spin, relative position, skeleton information, and return shot error in the learning data DB 314. An example of the learning data DB 314 is shown in Fig. 13. The items in the learning data DB 314 are the same as those in the learning data DB 214 in the second embodiment, except that there are multiple items of load data.
[0109] The learning unit 316 calculates a heat map showing the relationship between the load value of each of the load data items as the ability parameters of each sample user and the return shot percentage when the load indicated by the load data is applied. Specifically, the learning unit 316 allocates the load data value to multiple stages. For example, a ball speed of 2.5 to 3.5 m / sec is set to 3 m / sec. The learning unit 316 calculates the return shot percentage for each set of allocated values of the ball speed, ball spin, and relative position in the same manner as in the second embodiment. Then, the learning unit 316 obtains a heat map by storing the calculated return shot percentage in a cell corresponding to the load data value of each set in a heat map having the ball speed, ball spin, and relative position as each axis, and stores the heat map in the "ability parameter" column of the learning data DB 314 as shown in FIG. 13. In the example of FIG. 13, the return shot percentage in the heat map is represented by shading.
[0110] The learning unit 316 learns the correspondence between multiple items of load data (ball speed, ball spin, relative position) and skeletal information for a sample user and ability parameters. In this way, the learning unit 316 generates an estimation model 318 that estimates and outputs a heat map, which is the ability parameters of a target user, for input of multiple items of load data and skeletal information for the target user. The estimation model 318 can be realized by, for example, a neural network as shown in FIG. 14. In this case, the input to each neuron in the input layer corresponds to each element of multidimensional data combining multiple items of load data and skeletal information, and the output from each neuron in the output layer corresponds to the value of each cell of the heat map.
[0111] The determination unit 320 estimates a heat map for the target user by inputting multiple items of load data (ball speed, ball spin, relative position) and skeletal information for the target user acquired by the acquisition unit 312 into the estimation model 318. Then, the determination unit 320 determines recommended load data for the next attempt based on the estimated heat map. In addition, the determination unit 320 may output data for visualizing and displaying the estimated heat map.
[0112] A method for determining the recommended load data will be described with reference to the heat map shown in FIG. 15. For ease of explanation, FIG. 15 uses a two-dimensional heat map of two items of load data (ball speed and relative position). The same applies to the following FIGS. 18 to 21, 28, and 29. The determination unit 320 identifies cells (shaded cells in FIG. 15) in the heat map where the return ball rate is a predetermined value (for example, 85%) or more. The determination unit 320 determines the recommended load data from a combination of load data values included in the range indicated by the identified cells.
[0113] For example, the determination unit 320 can select the cell with the smallest return rate from among the identified cells, and determine the recommended load data from a combination of load data values included in the range indicated by that cell. For example, the determination unit 320 can select the cell indicated by a circle in FIG. 15, and determine the recommended load data of a ball velocity of 4 m / sec and a relative position of -0.3 m. Note that since the load data value corresponding to each cell is a value with a range, it can be preset to select the median, maximum value, or minimum value within that range. It may also be preset to determine a value randomly within the range from the minimum value to the maximum value.
[0114] In addition, since the purpose is to improve the skills of the target user, the recommended load data may be determined so as to impose a more severe load. Specifically, the determination unit 320 can determine the recommended load data from the values in the range indicated by the cells (cells indicated by circles or triangles in FIG. 15) adjacent to the cells whose return rates are less than the predetermined value among the cells whose return rates are equal to or greater than the predetermined value in the heat map. More precisely, the determination unit 320 can determine the recommended load data from the values on the border (thick line in FIG. 15) between the cells whose return rates are equal to or greater than the predetermined value and the cells whose return rates are less than the predetermined value in the heat map. The determination unit 320 outputs the determined load data to the table tennis robot 60. As a result, the recommended load data is set as control information in the table tennis robot 60, and the table tennis robot 60 is controlled to send (return) the ball at the ball speed, ball spin, and relative position indicated by the recommended load data at the next trial.
[0115] The hardware configuration of the recommended load determination device 310 is similar to that of the recommended load determination device 10 according to the first embodiment shown in Fig. 3, and therefore a description thereof will be omitted. Note that the storage device 36 stores a recommended load determination program for executing a recommended load determination process, which will be described later.
[0116] Next, the operation of the recommended load determining device 310 according to the third embodiment will be described.
[0117] Fig. 16 is a flowchart showing the flow of recommended load determination processing executed by the CPU 32 of the recommended load determination device 310. The CPU 32 reads out a recommended load determination program from the storage device 36, deploys it in the memory 34, and executes it, whereby the CPU 32 functions as each functional component of the recommended load determination device 310, and the recommended load determination processing shown in Fig. 16 is executed. Note that in the recommended load determination processing of the third embodiment, processing similar to that in the recommended load determination processing of the second embodiment (Fig. 10) is given the same step numbers, and detailed description thereof will be omitted.
[0118] If a positive determination is made in step S200 and the process proceeds to the learning mode process from step S202 onwards, the process of step S304 is executed instead of step S204 in the recommended load determination process of the second embodiment. In step S304, the acquisition unit 312 acquires multiple items of load data (ball speed, ball spin, relative position), skeletal information, and return shot error, and stores them in the learning data DB 314.
[0119] Moreover, instead of step S208 of the recommended load determination process of the second embodiment, the process of step S308 is executed. In step S308, the learning unit 316 calculates a heat map indicating the relationship between the load value of each of the multiple load data items and the return ball rate when the load indicated by the load data is applied, as the ability parameter of each sample user.
[0120] On the other hand, if a negative decision is made in step S200, the process proceeds to the practice mode process in step S214 and thereafter. If it is determined in step S214 that the estimation model 318 has already been generated, the process proceeds to step S316.
[0121] In step S316, the acquisition unit 312 instructs the table tennis robot 60 to set the ball speed, ball spin, and relative position to their initial values and start sending out the ball. Next, in step S318, the acquisition unit 312 acquires the ball speed, ball spin, and relative position of the ball sent out from the table tennis robot 60 in response to the instruction in step S316, as well as skeletal information indicating the swing of the target user when returning the ball.
[0122] Next, in step S320, the determination unit 320 inputs multiple items of load data (ball speed, ball spin, relative position) and skeletal information for the target user acquired by the acquisition unit 312 into the estimation model 318, and estimates a heat map for the target user. Next, in step S321, the determination unit 320 determines recommended load data for the next attempt based on the estimated heat map.
[0123] Next, in step S322, the determination unit 320 instructs the table tennis robot 60 to return the ball with the ball speed, ball spin, and relative position indicated by the recommended load data determined in step S321 above.
[0124] As described above, in the recommended load determination device according to the third embodiment, in an exercise task for improving the return rate using a table tennis robot, a load combining multiple items can be applied to the target user within the range in which the target user can return the ball. Also, by selecting a load that places a greater burden on the user within the range in which the target user can return the ball, a load that is beneficial to improving the skill of the target user can be applied.
[0125] <Fourth embodiment> Next, a fourth embodiment will be described. In the fourth embodiment, as in the third embodiment, the exercise task is to improve the return rate in a table tennis rally, the load given to the user is a return shot from a table tennis robot, and the user's motion is a swing to return the ball hit by the table tennis robot. In the third embodiment, as described with reference to FIG. 15, the range of load data in which the return rate is equal to or greater than a predetermined value is specified, and the value of the recommended load data is determined from this range. The recommended load data for the next trial is a set of values of load data for multiple items, but the specified range includes a set of multiple load values as candidates for the recommended load data. Therefore, in the fourth embodiment, a method of narrowing down the recommended load data to be determined from the specified range so that more useful recommended load data can be determined.
[0126] As shown in FIG. 17, the recommended load determination device 410 according to the fourth embodiment functionally includes an acquisition unit 412, a learning data DB 314, a learning unit 416, a comparison heat map DB 422, an estimation model 318, and a determination unit 420.
[0127] The acquisition unit 412, like the acquisition unit 312 according to the third embodiment, acquires multiple items of load data, motion data, and performance data in the learning mode, and acquires multiple items of load data and motion data in the practice mode. For example, the multiple items of load data are ball speed, ball spin, and relative position, the motion data is the user's skeletal information, and the performance data is the return shot error. Furthermore, the acquisition unit 412 acquires a strength / weakness designation indicating whether the target user wants to practice his / her strengths or weaknesses in the practice mode. This designation is input to the recommended load determination device 410 via the input device 38.
[0128] The learning unit 416, like the learning unit 316 in the third embodiment, calculates a heat map as ability parameters for each sample user and generates an estimation model 318. Furthermore, the learning unit 416 creates a comparison heat map for identifying a relative good range indicating a range of loads in which the target user is relatively good compared to other users, and a relative bad range indicating a range of loads in which the target user is relatively bad. Being good means that the target user has a higher chance of success than other users, and being bad means that the target user has a lower chance of success than other users.
[0129] Specifically, the learning unit 416 divides each sample user whose learning data is stored in the learning data DB 314 into levels according to their abilities for the motor task. This division into levels may be performed manually, or may be performed using ability parameters calculated by the learning unit 416 in the process of generating the estimation model 318. For example, the learning unit 416 determines the level of each sample user based on a level stage defined according to the ratio of the number of cells whose return rate is a predetermined value (e.g., 85%) or more to the total number of cells in the heat map. The number of levels may be set appropriately according to the number of sample users, etc. Then, as shown in FIG. 18, the learning unit 416 calculates a typical heat map for each level from the heat maps calculated for each sample user at the same level. For example, the learning unit 416 calculates a statistical value such as the average value, median value, mode value, etc. of the values of each cell of the heat maps of sample users at the same level as the value of each cell of the typical heat map. Then, the learning unit 416 stores the typical heat map calculated for each level in the comparison heat map DB 422 as a comparison heat map.
[0130] The determination unit 420, like the determination unit 320 of the third embodiment, estimates a heat map of the target user using the estimation model 318. Then, the determination unit 420 compares the heat map of the target user with the comparative heat map to create a relative strength and weakness map for identifying the relative strength and weakness ranges of the target user. The determination unit 420 determines, as recommended load data, a set of load values within the relative strength and weakness range in the created relative strength and weakness map.
[0131] Specifically, the determining unit 420 determines the level of the target user based on the heat map of the target user, identifies a typical heat map of the same level as the target user or one rank higher as a comparison heat map, and acquires it from the comparison heat map DB 422. The one rank higher level is used in consideration of skill improvement. The determining unit 420 then prepares an empty relative strength and weakness map having cells corresponding to each cell of the heat map. The determining unit 420 creates the relative strength and weakness map by storing, in each cell of the empty relative strength and weakness map, information corresponding to the result of comparing the value of each cell (return shot rate) between the heat map of the target user and the comparison heat map.
[0132] More specifically, as shown in FIG. 19, the determination unit 420 specifies cells where the return ball rate of the heat map of the target user is equal to or greater than a predetermined value (e.g., 85%) and the return ball rate of the comparison heat map is less than the predetermined value as a relative good range. Also, the determination unit 420 specifies cells where the return ball rate of the heat map of the target user is less than a predetermined value and the return ball rate of the comparison heat map is equal to or greater than the predetermined value as a relative poor range. Note that, in the example of FIG. 19, cells where the return ball rates of the heat map of the target user and the comparison heat map are both equal to or greater than a predetermined value are specified as a range where there is no problem. Also, cells where the return ball rates of the heat map of the target user and the comparison heat map are both less than a predetermined value are specified as a range where both are NG.
[0133] When the strength / weakness designation acquired by the acquisition unit 412 specifies "strength", the determination unit 420 determines the recommended load data based on the relative strength range of the relative strength / weakness map. When the strength / weakness designation acquired by the acquisition unit 412 specifies "weak", the determination unit 420 determines the recommended load data based on the relative weakness range of the relative strength / weakness map. A specific method of determining the recommended load data based on the relative strength range or the relative weakness range will be described below.
[0134] Considering skill improvement, it is desirable to practice in a way that broadens the relative range of strengths. Therefore, when determining recommended load data based on the relative range of strengths, the determination unit 420 can determine the recommended load data from the boundary (thick line in FIG. 20) between the cell indicating the relative range of strengths and the cell with a return rate lower than a predetermined value, as shown in FIG. 20. This can improve the return rate of the target user for loads included in the range in the direction indicated by the arrow in FIG. 20.
[0135] In addition, it is considered that practicing with a load that is just enough to make the return shot successful is more beneficial for improving skills than continuing to practice with a load range in which the target user's return rate is low. Therefore, when determining recommended load data based on the relative poor range, the determination unit 420 can determine the recommended load data from the range of cells (cells indicated by circles in FIG. 21) adjacent to the relative poor cell and having a return rate of a predetermined value or more, as shown in FIG. 21. Alternatively, the determination unit 420 can determine the recommended load data from the boundary (thick line in FIG. 21) between the relative poor cell and the cell having a return rate of a predetermined value or more. This can improve the return rate of the target user for loads included in the range in the direction indicated by the arrow in FIG. 21, that is, the relative poor range.
[0136] The hardware configuration of the recommended load determination device 410 is similar to that of the recommended load determination device 10 according to the first embodiment shown in Fig. 3, and therefore a description thereof will be omitted. Note that the storage device 36 stores a recommended load determination program for executing a recommended load determination process described later.
[0137] Next, the operation of the recommended load determining device 410 according to the fourth embodiment will be described.
[0138] In the fourth embodiment, if a positive judgment is made in step S200 in the recommended load determination process (FIG. 16) in the third embodiment and the process proceeds to the learning mode process from step S202 onwards, the comparison heat map creation process shown in FIG. 22 is executed after step S212.
[0139] In the comparative heat map creation process shown in FIG. 22, in step S400, the learning unit 416 determines the level of each sample user based on a level stage defined according to the ratio of the number of cells with a return rate of a predetermined value (e.g., 85%) or more to the total number of cells in the heat map.
[0140] Next, in step S402, the learning unit 416 calculates a typical heat map for each level from the heat maps calculated for each sample user at the same level, and stores the heat map as a comparison heat map in the comparison heat map DB 422. Then, the comparison heat map creation process ends, and the recommended load determination process also ends.
[0141] On the other hand, suppose that the result of the determination in step S200 is negative, the process proceeds to the practice mode process in step S214 and thereafter, and it is determined in step S214 that the estimation model 318 has already been generated. In this case, the acquisition unit 412 acquires the strength / weakness designation of the target user input to the recommended load determination device 410 via the input device 38.
[0142] Then, the processes of steps S316 to S320 are executed in the same manner as the recommended load determination process of the third embodiment, and in step S321, the determination process shown in FIG. 23 is executed.
[0143] 23, in step S410, the determination unit 420 determines the level of the target user based on the heat map of the target user estimated in step S320. Next, in step S412, the determination unit 420 specifies a typical heat map of the same level as the level of the target user determined in step S410 or one rank higher as a comparison heat map, and acquires it from the comparison heat map DB 422.
[0144] Next, in step S414, the determination unit 420 compares the value of each cell (return rate) between the heat map of the target user and the comparative heat map to create a relative strength and weakness map. Next, in step S416, recommended load data is determined based on the relative strength and weakness map created in step S414 and the strength / weakness designation acquired by the acquisition unit 412. Then, the process returns to the recommended load determination process (FIG. 16).
[0145] As described above, the recommended load determination device according to the fourth embodiment creates a relative strength and weakness map in which the relative strength and weakness ranges of the target user compared to other users are specified. Then, the recommended load data is determined based on this relative strength and weakness map, so that, compared to the third embodiment, it is possible to determine, as the recommended load data, a set that is more beneficial for skill improvement from among the candidate sets of load values that can become the recommended load data.
[0146] In the fourth embodiment, a case where a relative strength / weakness map in which both a relative strength range and a relative weakness range are specified is described, but the present invention is not limited to this. A relative strength map in which only a relative strength range is specified may be created, or a relative weakness map in which only a relative weakness range is specified may be created. In this case, in the practice mode, the target user may be made to practice the strengths of the target user based on the relative strength map, or the target user may be made to practice the weaknesses of the target user based on the relative weakness map. In addition, a designation of strength / weakness may be obtained, and in the case of the designation of "strength", a relative strength map may be created, and in the case of the designation of "weak", a relative weakness map may be created.
[0147] In the fourth embodiment, the determining unit may output data for visualizing and displaying the created relative strength and weakness map, the relative strength map, or the relative weakness map. This allows the target user to objectively grasp his / her own relative strength range or relative weakness range.
[0148] <Fifth embodiment> Next, a fifth embodiment will be described. In the fifth embodiment, as in the second to fourth embodiments, the motor task is to improve the return rate in a table tennis rally, the load given to the user is a return ball from a table tennis robot, and the user's motion is a swing to return the ball hit by the table tennis robot. When aiming to improve the skill of a motor task, there are cases where it is desired to improve the skill for the motor task under more detailed constraints. For example, in the example of table tennis, there are cases where it is desired to improve the return rate in a specific hitting style such as a cut, smash, or fore drive, rather than simply aiming to improve the return rate. Therefore, in the fifth embodiment, a method for determining useful recommended load data in response to more detailed requests for a motor task will be described.
[0149] As shown in FIG. 24, a recommended load determination device 510 according to the fifth embodiment functionally includes an acquisition unit 512, a learning data DB 514, a learning unit 516, an estimation model 518, and a determination unit 520.
[0150] The acquisition unit 512 acquires load data, motion data, and performance data in the learning mode, similar to the acquisition unit 212 according to the second embodiment, and acquires load data and motion data in the practice mode. For example, the load data is the ball speed, the motion data is the user's skeletal information, and the performance data is the return error. The acquisition unit 512 may acquire, for example, the ball speed, the ball rotation, and the relative position as load data for multiple items, similar to the acquisition unit 312 according to the third embodiment or the acquisition unit 412 according to the fourth embodiment. In the learning mode, the acquisition unit acquires motion data and performance data in the case where the sample user returns the ball after specifying a stroke style such as a cut, a smash, or a forehand drive. Therefore, the learning data DB 514 stores data such as that shown in FIG. 8 or FIG. 13 for each stroke style. The "stroke style" is an example of the "type of motion" of the disclosed technology.
[0151] Furthermore, the acquisition unit 512 acquires a practice menu designation indicating a hitting style that the target user wants to practice in the practice mode. This designation is input to the recommended load determination device 510 via the input device .
[0152] The learning unit 516 generates an estimation model 518 using the learning data for each hitting style. The specific method of generating the estimation model 518 is the same as in the second to fourth embodiments. Therefore, as shown in FIG. 25, the estimation model 518 is a collection of estimation models for each hitting style, such as for cuts, smashes, and forehand drives. In the example of FIG. 25, the estimation model 518 also includes an estimation model for random that is generated without limiting the hitting style.
[0153] The determination unit 520 estimates the ability parameters of the target user using an estimation model for a hitting style corresponding to the specification of the practice menu acquired by the acquisition unit 512 from among the estimation models 518, and determines recommended load data based on the ability parameters. The specific method of determining the recommended load data based on the ability parameters is similar to the second to fourth embodiments.
[0154] The hardware configuration of the recommended load determination device 510 is similar to that of the recommended load determination device 10 according to the first embodiment shown in Fig. 3, and therefore a detailed description thereof will be omitted. Also, the operation of the recommended load determination device 510 according to the fifth embodiment differs from the recommended load determination process of the second to fourth embodiments only in that a practice menu designation is acquired, and that an estimation model for a hitting style corresponding to the designated practice menu is used in step S220 of Fig. 10 or step S320 of Fig. 16, and therefore a detailed description thereof will be omitted.
[0155] As described above, in the recommended load determination device according to the fifth embodiment, for example, an estimation model is prepared for each hitting style, and the target user's ability parameters are estimated using the estimation model for the hitting style specified by the target user, and recommended load data is determined. This makes it possible to determine useful recommended load data in response to more detailed requests for exercise tasks.
[0156] Sixth embodiment Next, a sixth embodiment will be described. In the sixth embodiment, as in the second to fifth embodiments, the motor task is to improve the return rate in a table tennis rally, the load given to the user is a return ball from a table tennis robot, and the user's motion is a swing to return the ball hit by the table tennis robot. In each of the above embodiments, load data and motion data are input to an estimation model to estimate the ability parameters of a target user. Therefore, for example, in the case of table tennis, an excessively large load may be given to a target user who is good at swinging but has a poor return rate. For example, a fast ball may be sent to a target user who has diligently practiced swinging and has become good at swinging, but whose eyes cannot keep up with a fast ball. Conversely, the load may be insufficient for a target user who has a unique swing that cannot be said to be good but has a good return rate. In the sixth embodiment, a method of correcting the load in such a case and determining recommended load data that is useful for skill improvement will be described.
[0157] As shown in FIG. 26, the recommended load determination device 610 according to the sixth embodiment functionally includes an acquisition unit 612, a performance DB 624, a learning data DB 314, a learning unit 316, an estimation model 318, and a determination unit 620.
[0158] The acquisition unit 612 acquires multiple items of load data, motion data, and performance data in the learning mode, similarly to the acquisition unit 312 according to the third embodiment. In the practice mode, the acquisition unit 612 acquires performance data as an actual parameter of the motion result, in addition to multiple items of load data and motion data. For example, the multiple items of load data are the ball speed, ball spin, and relative position, the motion data is the user's skeletal information, and the performance data is the return shot error. The acquisition unit 612 stores the performance data acquired in the practice mode for each trial in, for example, a performance DB 624 as shown in FIG. 27.
[0159] The determination unit 620 determines the recommended load data so as to reduce the burden on the target user when the ability of the target user evaluated by the performance data is lower than the ability of the target user evaluated by the ability parameters that are the output of the estimation model 318. On the other hand, when the ability of the target user evaluated by the performance data is higher than the ability of the target user evaluated by the ability parameters that are the output of the estimation model 318, the determination unit 620 determines the recommended load data so as to increase the burden on the target user.
[0160] Specifically, the determination unit 620, like the determination unit 320 of the third embodiment, estimates a heat map as the ability parameter of the target user using the estimation model 318, and identifies a cell for determining recommended load data from the heat map. The determination unit 620 also calculates an actual return shot rate from return shot errors in a predetermined number of past trials stored in the performance DB 624. When the actual return shot rate is equal to or lower than a lower threshold (e.g., 85%), the estimated return shot rate stored in each cell of the estimated heat map is higher than the actual return shot rate corresponding to the target user's actual ability, and there is a possibility that a cell in a range of a load that is heavy for the target user has been identified. Therefore, the determination unit 620 corrects the cell identified from the heat map in a direction that reduces the load (decreasing direction). On the other hand, when the actual return shot rate is equal to or higher than an upper threshold (e.g., 95%), the estimated return shot rate stored in each cell of the estimated heat map is lower than the actual return shot rate corresponding to the target user's actual ability, and there is a possibility that a cell in a range of a load that is light for the target user has been identified. Therefore, the determining unit 620 corrects the cells identified from the heat map in a direction to increase the load (increase).
[0161] For example, as shown in Fig. 28, when the load included in the cell with a thick frame in Fig. 28 is used as a reference, the direction of load decrease is a direction toward a cell where the estimated return shot rate improves, as shown by the arrow in Fig. 28. Also, when the load included in the cell with a thick frame in Fig. 29 is used as a reference, the direction of load increase is a direction toward a cell where the estimated return shot rate decreases, as shown by the arrow in Fig. 29. For example, the determination unit 620 can identify a cell that is closest to the cell identified from the heat map in the increasing or decreasing direction, and determine a set of load values included in the range of the identified cell as the recommended load data.
[0162] The hardware configuration of the recommended load determination device 610 is similar to that of the recommended load determination device 10 according to the first embodiment shown in Fig. 3, and therefore a description thereof will be omitted. Note that the storage device 36 stores a recommended load determination program for executing a recommended load determination process, which will be described later.
[0163] Next, the operation of the recommended load determining device 610 according to the sixth embodiment will be described.
[0164] In the recommended load determination process in the sixth embodiment, similarly to the recommended load determination process in the third embodiment (FIG. 16), if a negative decision is made in step S200, the process proceeds to practice mode processing in step S214 and thereafter. Then, if it is determined in step S214 that the estimation model 318 has been generated, the determination process shown in FIG. 30 is executed instead of steps S316 to S321.
[0165] 30, in step S600, the acquisition unit 612 acquires multiple items of load data (ball speed, ball spin, relative position) and skeletal information, and stores them in the learning data DB 314. The acquisition unit 612 also acquires a return shot error, and stores it in the performance DB 624. Next, in step S602, the determination unit 620 inputs multiple items of load data and skeletal information to the estimation model 318 to estimate a heat map of the target user, and identifies cells from the heat map for determining recommended load data.
[0166] Next, in step S604, the determining unit 620 determines whether or not a predetermined number of pieces of performance data (return shot errors) are stored in the performance DB 624. If stored, the process proceeds to step S606, and if not stored, the process proceeds to step S614.
[0167] In step S606, the determination unit 620 judges whether the return shot rate calculated from the return shot errors for a predetermined number of past times stored in the performance DB 624 is equal to or lower than a lower threshold (e.g., 85%). If the return shot rate is equal to or lower than the lower threshold, the process proceeds to step S608, and if it exceeds the lower threshold, the process proceeds to step S610. In step S608, the determination unit 620 corrects a specified cell from the heat map in a decreasing direction, and determines a set of load values included in the range of the specified cell as the recommended load data.
[0168] Meanwhile, in step S610, the determination unit 620 judges whether the return ball rate is equal to or greater than an upper threshold (e.g., 95%). If the return ball rate is equal to or greater than the upper threshold, the process proceeds to step S612, and if it is less than the upper threshold, the process proceeds to step S614. In step S612, the determination unit 620 corrects a cell to be identified from the heat map in the increasing direction, and determines a set of load values included in the range of the identified cell as the recommended load data.
[0169] In step S614, the determination unit 620 determines the set of load values included in the range of the cell identified in step S602 as recommended load data, and then the process returns to the recommended load determination process (FIG. 16).
[0170] As described above, in the recommended load determination device according to the sixth embodiment, the recommended load data to be determined is corrected in a decreasing or increasing direction according to the capability indicated by the performance of the target user. This makes it possible to determine recommended load data that is beneficial for skill improvement according to the performance of the target user.
[0171] In the sixth embodiment, the load is increased for a target user who is not good at swinging but has a good return rate, but the present invention is not limited to this. For example, if the exercise task is to be able to swing well, the load may not be increased. In this case, the target user is given a low load, but since the target user will practice with a low load until the quality of the swing improves, there is no problem if the load is not increased.
[0172] In addition, in each of the above embodiments, a case has been described in which the recommended load data for the next attempt is determined based on the load data applied to the current attempt and the motion data for the load data. In this case, as shown in FIG. 31, if a swing is usually bad but a good swing is made just once, an excessive load may be applied in the next attempt. The example in FIG. 31 shows an example in which the first, second, fourth, and fifth attempts are bad swings and a good swing happens to be made in the third attempt. In this case, a low ability parameter is estimated for a bad swing, so that appropriate recommended load data according to the target user is determined. However, a high ability parameter is estimated for a good swing, so that recommended load data that places a heavy burden on the target user is determined.
[0173] In order to suppress such sudden fluctuations in the load data, k+1 pieces of load data and movement data in the past k+1 trials (nth trial, n-1th trial, ..., nkth trial) are averaged as shown in Fig. 32. Then, the averaged load data n_Ave and movement data n_Ave may be input to the estimation model to estimate the ability parameters of the nth trial.
[0174] 33, the k+1 capacity parameters estimated in the past k+1 trials (n trial, n-1 trial, . . . , nk trial) may be averaged, and the recommended load data may be determined from the average capacity parameter n_Ave. Also, the recommended load data determined based on each of the k+1 capacity parameters may be averaged to obtain the final recommended load data.
[0175] In the above-mentioned embodiments, as shown in FIG. 14, the load data and the movement data are inputted to an estimation model realized by a neural network or the like to estimate the ability parameters. As described above, when skeletal information is used as the movement data, the difference between the number of dimensions of the movement data and the number of dimensions of the load data becomes too large, and the difference in the number of dimensions may affect the learning of the estimation model, leading to a possibility that the learning may proceed in a biased manner. Therefore, as shown in FIG. 34, the movement data may be dimensionally compressed so that the number of dimensions of the movement data is approximately the same as the number of dimensions of the load data, and then used as learning data. In this case, the movement data is also dimensionally compressed when estimating the ability parameters, as in the case of learning. Note that the same number of dimensions refers to the number of dimensions where the difference between the number of dimensions of the load data and the number of dimensions of the movement data is within a predetermined range. In the example of FIG. 34, the number of dimensions of the movement data is n×3×f, assuming that the number of joints in the skeletal information, which is the movement data, is n, the coordinate value representing one joint is 3, and the number of frames of the image of one movement unit is f. In order to reduce the number of dimensions to s, which is approximately the same as the number of dimensions i of the load data, s types of feature amounts are extracted from the motion data to obtain the motion data after dimension reduction.
[0176] An example of a method for dimensional compression of motion data will be described. For example, a matrix as shown in FIG. 35 is created from motion data of a plurality of sample users. The coordinate values of actual skeletal information are entered in the shaded area in FIG. 35. Then, a principal component analysis is performed on this matrix, for example, the cumulative contribution rate of the principal components is obtained, and the top s principal components are identified. Then, as shown in FIG. 36, values obtained by mapping motion data showing each swing to each of the identified principal components P1, P2, . . . , Ps are calculated. The values mapped to each principal component are entered in the shaded area in FIG. 36. This allows the n×3×f dimensional motion data to be dimensionally compressed to s dimensions.
[0177] As another method of dimensionality reduction, the motion data may be converted into a feature quantity representing the skill of the swing based on knowledge information on the corresponding motion or an analysis of the motion data. For example, if it is known that a person who is good at swinging has a constant elbow height, the variance of the elbow height is extracted as a feature quantity from the motion data. If it is known that a person who is good at swinging has a smooth swing trajectory, the trajectory shown by the motion data is analyzed, for example, by taking a differential, and an index showing the smoothness of the trajectory is extracted as a feature quantity. Then, the motion data showing each swing is converted into a feature quantity as shown in FIG. 37. The shaded parts in FIG. 37 are filled with the feature quantities extracted from the motion data for each feature.
[0178] In the above third to sixth embodiments, an example of determining a recommended load for an exercise task of improving a return shot rate using a table tennis robot has been described, but the present invention is not limited to this. As described at the end of the second embodiment, an exercise task of improving a hitting percentage using a batting machine, as exemplified in the first embodiment, and an exercise task of improving walking in rehabilitation using a treadmill can also be realized as specific embodiments similar to the third to sixth embodiments.
[0179] In addition, in each of the above embodiments, the functional units that function in the learning mode and the functional units that function in the practice mode are implemented in the same computer, but this is not limiting. The ability parameter estimation model learning device including the acquisition unit and the learning unit and the recommended load determination device including the acquisition unit and the determination unit may be implemented in separate computers.
[0180] In addition, the recommended load determination process executed by the CPU by reading the software (program) in each of the above embodiments may be executed by various processors other than the CPU. In this case, examples of the processor include a PLD (Programmable Logic Device) such as an FPGA (Field-Programmable Gate Array) whose circuit configuration can be changed after manufacture, and a dedicated electric circuit such as an ASIC (Application Specific Integrated Circuit) which is a processor having a circuit configuration designed exclusively for executing a specific process. In addition, the recommended load determination process may be executed by one of these various processors, or may be executed by a combination of two or more processors of the same or different types (for example, a plurality of FPGAs, and a combination of a CPU and an FPGA). In addition, the hardware structure of these various processors is, more specifically, an electric circuit that combines circuit elements such as semiconductor elements.
[0181] In addition, in each of the above embodiments, the recommended load determination program is described as being stored (installed) in advance in a storage device, but the present invention is not limited to this. The program may be provided in a form stored in a storage medium such as a CD-ROM, a DVD-ROM, a Blu-ray disc, or a USB memory. The program may also be downloaded from an external device via a network. [Explanation of symbols]
[0182] 10, 210, 310, 410, 510, 610 Recommended load determination device 12, 212, 312, 412, 512, 612 Acquisition part 14, 214, 214A, 214B, 314, 514 Learning data DB 16, 216, 316, 416, 516 Learning Department 18, 218, 318, 518 Estimation Model 20, 220, 320, 420, 520, 620 Decision section 422 Comparison Heatmap DB 624 Achievement DB 32 CPU 34 Memory 36 Storage device 38 Input Devices 40 Output Device 42 Storage media reader 44 Communication I / F 46 Bus 60 Table Tennis Robot 70, 72 Sensors
Claims
1. A recommended load determination device that utilizes an estimation model that outputs a parameter related to a user's ability, the parameter being calculated based on load data representing a load applied to a user and action data representing an action of the user under the load, the parameter comprising: an acquisition unit that acquires load data representing a load applied to a target user for which a recommended load is to be determined, and action data representing an action of the target user under the load; a determination unit that determines recommended load data representing a recommended load for the target user based on parameters related to the capability of the target user outputted from the estimation model by inputting the acquired load data and the operation data into the estimation model, The ability-related parameter output by the estimation model is a parameter that represents a relationship between the load and the likelihood of success for the movement. Recommended load determination device.
2. The recommended load determination device of claim 1, wherein the motion data is displacement data of a specific part of the user or an equipment held by the user when performing the motion, muscle activity data of the user, or differential data or integral data of the displacement data or the muscle activity data.
3. 3. The recommended load determining device according to claim 2, wherein the displacement data of the specific part of the user is time-series data of skeletal information of the user.
4. the ability-related parameter output by the estimation model is limit load data representing a load value that is the largest burden on the user within a range of load values for one load item, where the likelihood of success in the operation is greater than a predetermined standard; The determining unit determines the limit load data for the target user as the recommended load data. The recommended load determining device according to any one of claims 1 to 3.
5. the parameter representing the relationship between the load for the target user and the likelihood of success for the action is a parameter specifying a graph representing the relationship between the load value for one load item and the likelihood of success, The determination unit determines, as the recommended load data, limit load data that represents a load value that is the greatest burden on the target user within a range of load values in which the likelihood of success in the graph is greater than a predetermined standard. The recommended load determining device according to any one of claims 1 to 4.
6. the parameter representing the relationship between the load for the target user and the magnitude of the likelihood of success for the action is a parameter specifying a heat map representing the relationship between the load value for a plurality of load items and the magnitude of the likelihood of success; The determination unit determines, as the recommended load data, a set of load values in the heat map in which the likelihood of success is greater than a predetermined standard. The recommended load determining device according to any one of claims 1 to 4.
7. The recommended load determining device according to claim 6 , wherein the determining unit outputs data for visualizing and displaying the heat map.
8. The determination unit is capable of using a comparative heat map, and creates a relative goodness map that identifies a relative goodness range, which is a range of the load value pairs in which the target user has a higher likelihood of success, or a relative poorness map that identifies a relative poorness range, which is a range of the load value pairs in which the target user has a lower likelihood of success, in comparison with the comparative heat map, and determines the load value pairs within the relative goodness range or the relative poorness range as the recommended load data. The recommended load determining device according to claim 6 or 7.
9. The comparative heat map is prepared for each level of the user, The determination unit uses the comparative heat map for the user at a level equal to or higher than the level of the target user and closest to the level of the target user. The apparatus for determining a recommended load according to claim 8.
10. The acquisition unit further acquires a designation of strengths or weaknesses for the target user, The determination unit creates the relative strength map when the strength designation is acquired, and creates the relative weakness map when the weakness designation is acquired. The recommended load determining device according to claim 8 or 9.
11. The recommended load determination device according to any one of claims 8 to 10, wherein the determination unit outputs data for visualizing and displaying at least one of the relative strengths map and the relative weakness map.
12. the estimation model is prepared for each type of the motion, The acquisition unit acquires a designation of any one of the types from the target user, The determination unit determines the recommended load data by using the estimation model corresponding to the specified type. The recommended load determining device according to any one of claims 1 to 11.
13. The acquisition unit acquires a plurality of sets of the load data and the operation data, The determination unit inputs load data obtained by averaging a plurality of the load data and operation data obtained by averaging a plurality of the operation data into the estimation model. The recommended load determining device according to any one of claims 1 to 12.
14. The acquisition unit acquires a plurality of sets of the load data and the operation data, The determination unit averages parameters related to the capabilities obtained by inputting each of the sets of the load data and the operation data into the estimation model, and determines the recommended load data based on the averaged parameters. The recommended load determining device according to any one of claims 1 to 12.
15. The acquisition unit acquires a plurality of sets of the load data and the operation data, The determination unit calculates a plurality of recommended load data based on parameters related to the capabilities obtained by inputting each of the sets of the load data and the operation data into the estimation model, and further determines a final recommended load data by averaging the plurality of recommended load data. The recommended load determining device according to any one of claims 1 to 12.
16. The acquisition unit acquires a set of the load data and the operation data and corresponding actual measurement parameters related to actual operation results; When the ability of the target user evaluated by the acquired actual measurement parameters is lower than the ability of the target user evaluated by the parameters that are the output of the estimation model, the determination unit determines the recommended load data so as to reduce a burden on the target user. The recommended load determining device according to any one of claims 1 to 15.
17. A recommended load determination device as described in any one of claims 1 to 16, wherein the determination unit outputs the recommended load data to the load-assignment device in order to control the load-assignment device that assigns the load to the target user.
18. 17. The recommended load determination device according to claim 1, wherein the determination unit outputs data for displaying the latest recommended load data at a predetermined time interval.
19. A recommended load determination device as described in any one of claims 1 to 18, including a learning unit that generates the estimation model by learning the correspondence between the load data and the operation data and parameters related to the user's ability calculated based on the load data and the operation data.
20. The recommended load determination device according to claim 19 , wherein the learning unit generates the estimation model using the operation data compressed so that a difference between the number of dimensions of the load data and the number of dimensions of the operation data is within a predetermined range.
21. an acquisition unit that acquires load data representing a load applied to a user, action data representing an action of the user under the load, and performance data associated with the action; a learning unit that learns correspondence between the load data and the operation data and parameters related to the user's ability calculated based on the load data and the performance data, and generates an estimation model that outputs parameters related to the ability of the target user in response to input of the load data and the operation data for a target user for which a recommended load is to be determined; The ability-related parameter output by the estimation model is a parameter that represents a relationship between the load and the likelihood of success for the movement. Ability parameter estimation model learning device.
22. preparing an estimation model that outputs parameters related to the ability of the user, the parameters being calculated based on load data representing a load applied to the user and action data representing an action of the user under the load; The acquisition unit acquires load data representing a load applied to a target user for which a recommended load is to be determined, and action data representing an action of the target user under the load, A determination unit determines recommended load data representing a recommended load for the target user based on parameters related to the capability of the target user outputted from the estimation model by inputting the acquired load data and the operation data into the estimation model; The ability-related parameter output by the estimation model is a parameter that represents a relationship between the load and the likelihood of success for the movement. Recommended load determination method.
23. A recommended load determination program for causing a computer to function as a recommended load determiner that utilizes an estimation model that outputs parameters related to the ability of a user, the parameters being calculated based on load data representing a load applied to a user and action data representing an action of the user under the load, the recommended load determination program comprising: The computer, an acquisition unit that acquires load data representing a load applied to a target user for which a recommended load is to be determined, and action data representing an action of the target user under the load; and a determination unit that determines recommended load data representing a recommended load for the target user based on parameters related to the ability of the target user outputted from the estimation model by inputting the acquired load data and the operation data into the estimation model; The ability-related parameter output by the estimation model is a parameter that represents a relationship between the load and the likelihood of success for the movement. Recommended Load Determination Program.
Citation Information
Patent Citations
Exercising capacity ebaluator, exercising apparatus and exercising capacity ebaluation
JP1999235401A
Pattern learning method, device and program
JP2010205043A
Method of evaluating fatigue by near-infrared spectroscopy
JP2014104265A
Control device, system, and control method and program
JP2018055611A
Exercise apparatus, control device, and program
JP2018175670A
