JP6326167B1 - Learning device, learning method, learning program, moving image distribution device, activity device, activity program, and moving image generation device - Google Patents

Abstract

To learn the movement of an object. A learning device includes: a motion generation unit that generates motion data that associates an identifier of an action to be performed by an object with a target of a combination of movements of an operation unit for performing the action; According to the motion data 16, the motion of the movement part of the object O is tried, an evaluation value for the action is output, and an update unit 23 that associates the identifier of the action, the motion data, and the evaluation value and stores them in the learning data T is provided. . The motion generation unit 22 generates new motion data with reference to the learning data T associated with the identifier of the action to be performed by the object. [Selection] Figure 3

Description

The present invention relates to a learning device, a learning method, a learning program, a moving image distribution device, an activity device, an activity program, and a moving image generation device in which an object including an operation unit learns in a virtual region.

  In recent years, machine learning has become widespread, and a technique for automatically constructing a thinking routine of a robot or the like and a thinking routine of a non-player character (NPC) appearing in various games is known (see, for example, Patent Document 1). In Patent Document 1, a method of automatically constructing a thought routine for determining an action to be taken by an agent in a predetermined situation is disclosed.

Japanese Patent No. 5874292

  In the learning of an object called an artificial life form, it is known to be divided into two stages: a learning stage of decision making that determines an action to be taken and a learning stage that learns a series of actions for performing an action. . However, the technique described in Patent Document 1 does not mention anything about learning of a series of operations.

Accordingly, an object of the present invention is to provide a learning device, a learning method, a learning program, a moving image distribution device, an activity device, an activity program, and a moving image generation device in which an object learns an action.

  In order to solve the above-described problem, a first feature of the present invention relates to a learning apparatus in which an object including an operation unit learns. The learning device according to the first feature includes a motion generation unit that generates motion data in which an identifier of an action to be performed by an object is associated with a target of a combination of movements of the operation unit for performing the action, and motion data The motion generation unit includes an update unit that tries to move the motion unit of the object, outputs an evaluation value for the action, and stores the action identifier, the motion data, and the evaluation value in the learning data in association with each other. New motion data is generated with reference to the learning data associated with the identifier of the action to be performed by the object.

  The object is active in a virtual area, and the virtual area includes a plurality of partial areas, and each partial area is associated with an environmental value identifier of the partial area, and the learning data is an action identifier, motion data, and evaluation value. Further, the environmental value identifier of the partial area is stored in association with each other, and the motion generation unit generates new motion data by referring to the learning data associated with the environmental value identifier of the partial area where the object is located. Also good.

  The motion data includes a plurality of pairs in which the motion of the motion unit is associated with the time to try the motion, and the update unit attempts to move based on the plurality of pairs generated by the motion data. The evaluation value may be output.

  The object may further include a distribution unit that sequentially generates and distributes moving image data indicating a state in which the object tries to move the operation unit according to the motion data.

  The second feature of the present invention relates to a learning method in which an object having an operation unit learns. In the learning method according to the second aspect of the present invention, the computer generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of motions of the operation unit for performing the action. The computer tries to move the motion part of the object according to the motion data, outputs an evaluation value for the action, associates the action identifier, the motion data, and the evaluation value in the learning data, and the computer Includes a step of generating new motion data with reference to the learning data associated with the identifier of the action to be performed by the object.

  A third feature of the present invention relates to a learning program for learning an object having an operation unit. The learning program according to the third aspect of the present invention is a program for generating motion data that associates an identifier of an action that causes an object to be performed with an object and a target of a combination of motions of an operation unit that causes the action to be performed. According to the generation unit and the motion data, the movement of the motion unit of the object is attempted, the evaluation value for the action is output, and the action identifier, the motion data, and the evaluation value are associated with each other and stored in the learning data. The motion generation unit generates new motion data with reference to the learning data associated with the identifier of the action to be performed by the object.

  A fourth feature of the present invention relates to a moving image distribution apparatus connected to a learning apparatus that an object having an operation unit learns. In a fourth aspect of the present invention, the learning device includes a motion generation unit that generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of movements of the operation unit for performing the action; A motion generation unit including an update unit that attempts to move the motion unit of the object in accordance with the motion data, outputs an evaluation value for the action, and stores the action identifier, the motion data, and the evaluation value in association with the learning data; Refers to the learning data associated with the identifier of the action to be performed by the object, generates new motion data, and the video distribution device uses the learning data generated in the learning device to operate the object The video data indicating the state of trying the movement of the part is sequentially generated and distributed.

  A fifth feature of the present invention relates to an activity device that is connected to a learning device that learns an object having an operation unit. In the fifth aspect of the present invention, the learning device includes a motion generation unit that generates motion data in which an identifier of an action to be performed by an object is associated with a target of a combination of movements of the operation unit for performing the action, A motion generation unit including an update unit that attempts to move the motion unit of the object in accordance with the motion data, outputs an evaluation value for the action, and stores the action identifier, the motion data, and the evaluation value in association with the learning data; Refers to the learning data associated with the identifier of the action to be performed by the object, generates new motion data, and the activity device operates the object using the learning data generated by the learning device.

  A sixth feature of the present invention relates to an activity program connected to a learning device that learns an object having an operation unit. In a sixth aspect of the present invention, the learning device includes a motion generation unit that generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of movements of the operation unit for performing the action; A motion generation unit including an update unit that attempts to move the motion unit of the object in accordance with the motion data, outputs an evaluation value for the action, and stores the action identifier, the motion data, and the evaluation value in association with the learning data; Refers to the learning data associated with the identifier of the action to be performed by the object, generates new motion data, and the activity program uses the learning data generated by the learning device to the computer to identify the object. Make it work.

  A seventh feature of the present invention relates to a moving image generating apparatus that is connected to a learning apparatus that learns an object having an operation unit. In a seventh aspect of the present invention, the learning device includes a motion generation unit that generates motion data in which an identifier of an action to be performed by an object is associated with a target of a combination of movements of the operation unit for performing the action; A motion generation unit including an update unit that attempts to move the motion unit of the object in accordance with the motion data, outputs an evaluation value for the action, and stores the action identifier, the motion data, and the evaluation value in association with the learning data; Refers to the learning data associated with the identifier of the action to be performed by the object, generates new motion data, and the video generation device uses the learning data generated by the learning device to operate the object The moving image data indicating the state of trying the movement of the part is sequentially generated.

According to the present invention, it is possible to provide a learning device, a learning method, a learning program, a moving image distribution device, an activity device, an activity program, and a moving image generation device in which an object learns an action.

It is a figure explaining the system configuration of the learning system concerning an embodiment of the invention. It is a figure explaining an example of a mode that an object activates a virtual area. It is a figure explaining the hardware constitutions and functional block of the learning apparatus which concerns on embodiment of this invention. It is a figure explaining the virtual area and partial area which concern on embodiment of this invention. It is a figure explaining an example of a data structure and data of area attribute data concerning an embodiment of the invention. It is a figure explaining an example of the data structure and data of environmental value data which concern on embodiment of this invention. It is a figure explaining an example of the data structure and data of object attribute data concerning an embodiment of the invention. It is a figure explaining an example of data structure and data of operation part attribute data concerning an embodiment of the invention. It is a figure explaining an example of data structure and data of evaluation value index data concerning an embodiment of the invention. It is a figure explaining an example of data structure and data of learning data concerning an embodiment of the invention. It is a flowchart explaining the learning process by the learning part which concerns on embodiment of this invention. It is a figure explaining an example of motion data concerning an embodiment of the invention. It is a figure explaining an example of a motion of the operation part by motion data concerning an embodiment of the invention. It is a figure explaining the system configuration | structure of the learning system which concerns on the modification of this invention. It is a figure explaining an example of the flow of data in the learning system concerning the modification of the present invention.

  Next, embodiments of the present invention will be described with reference to the drawings. In the following description of the drawings, the same or similar parts are denoted by the same or similar reference numerals.

(Learning system)
The learning system 9 according to the embodiment of the present invention shown in FIG. 1 simulates the activity of an object O called an artificial life form on a computer and provides the simulation result as a moving image. Particularly in the embodiment of the present invention, the object O includes an operation unit. The object O learns the motion of the motion unit that can obtain a better effect under the restrictions of the attribute of the object O and the environment value of the area where the object O is arranged.

  The object O according to the embodiment of the present invention is described on the assumption that the object O is active in the virtual region V, but is not limited thereto. The object O is a tangible object such as a robot, and may include a component corresponding to the operation unit and a computer that controls the component.

  As shown in FIG. 1, the learning system 9 includes a learning device 1, a moving image distribution device 2, and a terminal 3. The learning device 1, the moving image distribution device 2, and the terminal 3 are connected to each other via a communication network 8 so as to communicate with each other.

  As illustrated in FIG. 2, the learning device 1 performs a simulation in which the object O is active in the virtual region V. In the example shown in FIG. 2, the object O is not limited to the dog-shaped shape that has limbs that support the torso and walks on four legs. The object O may have a humanoid shape for biped walking, or may have a shape having limbs radially as shown in FIG.

  The moving image distribution device 2 distributes the moving image of the movement of the object O learned by the learning device 1 to the terminal 3.

  The terminal 3 reproduces the moving image distributed from the moving image distribution device 2. Further, the terminal 3 may input an instruction to start or end learning to the learning device 1 or specify an action to be performed by the object O.

  In the embodiment of the present invention, a case where the learning device 1, the moving image distribution device 2, and the terminal 3 are different devices will be described, but a predetermined function may be mounted on the same device. For example, the learning device 1 may distribute the moving image data to the terminal 3, or the learning device 1 may reproduce the moving image.

(Learning device)
With reference to FIG. 3, the learning apparatus 1 which concerns on embodiment of this invention is demonstrated.

  The learning device 1 according to the embodiment of the present invention simulates the activity of an object also called an artificial life form.

  In the object O, the operation unit is a joint that connects components constituting the object O. In addition, when a part constituting the object is an operable part such as a weapon or a tool, this part is also an operation part. A series of motions in which the joint is bent or rotated in a predetermined direction is tried to learn so that a better effect can be obtained in a predetermined action. In the embodiment of the present invention, the case where the object O includes a plurality of operation units will be described, but one operation unit may be included.

In the embodiment of the present invention, the “action” is an operation performed on the entire object O, for example, forward, backward, rotation, jump, and the like. When the action part of the object O is a tool, “action” includes use of the tool. For example, when the operation portion is a gun, "action" includes a forward fire.

  In the embodiment of the present invention, “motion” is a combination of a series of movements of the respective operation units of the object O for performing an action. The effect of the action is evaluated in units of “motion”. A combination of motions of the motion parts of the object O performed within a predetermined time is generated as a motion. The object O is made to try to operate according to the generated motion under the restrictions of the attribute of the object O and the environment value of the area where the object O is arranged.

  Motion is evaluated for each action. For example, the motion generated to perform the forward action has a higher evaluation as the distance moved forward is larger as a result of the motion. The learning device 1 generates a new motion based on the motion evaluation performed for each action, and searches for and learns a motion with higher evaluation.

  As illustrated in FIG. 3, the learning device 1 is a general computer including a storage device 10, a processing device 20, an input device 30, an output device 40, and a communication control device 50. Each function shown in FIG. 3 is realized by a general computer executing a learning program for executing predetermined processing.

  The storage device 10 stores a learning program and also stores region attribute data 11, environment value data 12, object attribute data 13, operation unit attribute data 14, evaluation index data 15, motion data 16, learning data T, and moving image data M. Remember. The area attribute data 11, the environment value data 12, the object attribute data 13, the action part attribute data 14, and the evaluation index data 15 are reference data used when the object O is active in the virtual area V in the embodiment of the present invention. It is. The motion data 16, the learning data T, and the moving image data M are data obtained when the object O is active in the virtual region V, and are updated as appropriate.

  In the embodiment of the present invention, the virtual area V includes a plurality of partial areas D, and each partial area D is associated with the environment value identifier of the partial area D. As shown in FIG. 4A, when the object O is active on a plane, the planar virtual region V is divided into a plurality of partial regions D. As shown in FIG. 4B, when the object O is active in space, the space-shaped virtual region V is divided into a plurality of partial regions D. Each partial area D has such a size that it can move to another partial area D by learning 2 to 5 motion data, for example. One side of the partial region D preferably has a size of about 1.5 to 5 times the size of the object O, for example.

  The area attribute data 11 and the environment value data 12 associate the environment values of the partial areas D of the virtual area V with each other. The area attribute data 11 and the environment value data 12 correspond to activity conditions in the activity area of the object O.

  The area attribute data 11 is data that associates environmental values of the partial area D in which the object O is active. As shown in FIG. 5, the region attribute data 11 is data in which a partial region identifier is associated with an environment value identifier.

  The number of environmental values used in the embodiment of the present invention is preferably set to be smaller than the number of partial regions D used in each virtual region V. In other words, the predetermined environment value is associated with a plurality of partial areas D, and the learning data in the predetermined environment value is considered as learning data for the plurality of partial areas D associated with the environment value. Thus, by dividing the virtual region V into a plurality of partial regions D and accumulating learning data for each partial region D, it is possible to reduce the amount of calculation in learning and efficiently learn.

  The environmental value data 12 is data that associates an environmental value identifier associated with the partial region D with a combination of environmental value parameters. As shown in FIG. 6, the environmental value data 12 is associated with an environmental value identifier, inclination, friction, gravity, adsorption force, water depth, temperature, and the like. For example, the temperature affects the energy consumption of the object O. The environment value data 12 associates the combination of environment values of each partial area D. In the environment value data 12, there may be an item that is not set depending on the attribute of the partial region D.

  The object attribute data 13 and the action part attribute data 14 associate the object O with the action part attribute. The object attribute data 13 and the action part attribute data 14 correspond to the activity conditions of the object O itself when the object O is active.

  The object attribute data 13 is data that associates the attributes of the entire object O with each other. As shown in FIG. 7, the object attribute data 13 is data that associates an object identifier that identifies an object O that is active in the virtual region V, a type of the object O, an identifier of an action that can be selected by the object O, and an individual value. It is. Individual values include basal metabolism, behavioral metabolism, physical fitness, and the like. Even when the types of the objects O are common, different individual values are associated with each object identifier.

  The action part attribute data 14 is data in which attributes of action parts included in each object O are associated. As shown in FIG. 8, the motion part attribute data 14 is data in which an object identifier, a motion part identifier, and an individual value are associated with each other. Individual values include shape, muscle strength, size, weight, center of gravity, texture, and the like. Even when the type of the object O and the motion part identifier are common, different individual values are associated with each object identifier. Note that the texture information of each operation unit is taken into account when drawing the movement of the object O.

  The evaluation index data 15 is data in which an evaluation function for each action performed by the object O is associated with an evaluation value. As shown in FIG. 9, the evaluation index data 15 is data in which an action identifier, an action name, an evaluation function, and an evaluation index are associated with each other. Specifically, the “forward” action is evaluated based on the “forward movement amount”, and an evaluation value of “evaluation is better as the movement amount is larger” is given. Further, the action of “forward shooting” is evaluated based on “the deviation of the fired bullet”, and the evaluation value “the smaller the deviation is, the better the evaluation” is given. Here, “the deviation of the fired bullet” is “the deviation of the solid angle between the firing direction of the fired bullet and the forward direction of the object O”. The evaluation value of the evaluation index data 15 may be evaluated by a numerical value, or may be evaluated at a level obtained by dividing the numerical value.

  The motion data 16 is a series of motion data to be performed by each motion unit of the object O in motion learning. The motion data 16 is data in which an identifier of an action to be performed by the object O is associated with a target of a combination of movements of the operation unit for performing the action. The motion data 16 includes a plurality of pairs in which the motion of the motion unit is associated with the time for attempting the motion. The motion data 16 will be described later with reference to FIG.

  The learning data T is data obtained as a result of operating the object O based on the motion data. The learning data T is data obtained as a result of learning for each action identifier and environment identifier. Specifically, the learning data T is data in which an action identifier, an action name, an environment value identifier, a motion, and an evaluation value are associated with each other.

  The moving image data M is moving image data that the object O learns.

  The processing device 20 includes a learning unit 21 and a moving image processing unit 26.

  The learning unit 21 includes a motion generation unit 22 and an update unit 23 in order to cause the object O to learn.

  The motion generation unit 22 generates motion data in which an identifier of an action to be performed by the object O is associated with a target of a combination of movements of the operation unit for performing the action. The motion generation unit 22 generates new motion data with reference to the learning data associated with the identifier of the action to be performed by the object O.

  In many cases, the motion generation unit 22 generates a motion that can be expected to have a good evaluation, based on the motion that has been evaluated well, from the learning data T that indicates the result of learning in the past for each operation unit of the object O. In addition, the motion generation unit 22 randomly generates a motion at a predetermined ratio or less regardless of the evaluation of the learning data T. By selecting not only motions that can be expected to be good, but also randomly selecting motions, new evaluations that are difficult to predict can be obtained. As a result, even better evaluations can be obtained, which contributes to the evolution of the object O.

  The motion generation unit 22 generates new motion data with reference to the learning data T associated with the environment value identifier of the partial region D where the object O is located. In the embodiment of the present invention, the learning data T is linked to the environmental value identifier of the partial area D, so the object O refers to the past learning result in the environmental value identifier of the partial area where the object O is located. To learn more. By managing the environmental value where the object O is located as a combination of parameters instead of individual parameters, it is possible to reduce the amount of calculation in learning and obtain a desired learning result in a shorter time.

  The update unit 23 attempts to move the motion unit of the object O according to the motion data 16, outputs an evaluation value for the action, and stores the action identifier, the motion data, and the evaluation value in the learning data T in association with each other. The update unit 23 outputs an evaluation value after trying to move based on a plurality of pairs generated by the motion data 16.

  The updating unit 23 sets the object O so as to realize the target of the motion unit motion combination based on the pair generated by generating the motion data 16 and associating the motion of each motion unit with the time to try the motion. Make it work. At this time, the update unit 23 operates the object O using the environment value of the partial region D where the object O is located, the attribute of the object O, and the attribute of the operation unit as constraints.

  Here, depending on the environmental value of the partial area in which the object O is located, the attribute of the object O, and the attribute of the action part, the object O may not be able to achieve its action and may have to give up halfway. The object O tries a new operation specified at the next time even if the operation cannot be achieved by trying the operation for the time specified by the motion data 16.

  The update unit 23 evaluates the action performed on the object O after trying the operation for all pairs generated as the motion data 16. For example, when the “forward” action is performed, the “forward movement amount” obtained as a result of the movement of the object O based on the motion data 16 is calculated. Further, the updating unit 23 converts the calculated “forward movement amount” into an evaluation value based on the evaluation index data 15 and stores it in the learning data T.

  The learning unit 21 repeats the generation of the motion data 16 and the update of the learning data T until the learning is completed by satisfying a predetermined condition. The learning unit 21 ends the process when an end instruction is input from the input device 30 or the communication control device 50, for example. The learning device 1 may end the process when the predetermined number of learning times or the learning time is reached, or may end the process when a predetermined evaluation value is obtained.

  The moving image processing unit 26 includes a moving image generation unit 27 and a moving image distribution unit 28.

  The moving image generation unit 27 sequentially generates moving image data indicating a state in which the object O attempts to move the operation unit according to the motion data 16. The moving image generating unit 27 attempts to move each motion unit at each time based on a pair in which the motion of the motion unit of the object O is associated with the time to try the motion, and the environment of the partial region where the object O is located. Depending on the value, the attribute of the object O, and the attribute of the motion part, moving image data M is generated that draws a state where the motion can be achieved or cannot be achieved and stops halfway.

  The moving image distribution unit 28 distributes the moving image data M generated by the moving image generation unit 27 to the terminal 3. The moving image distribution unit 28 may transmit the moving image data M generated by the moving image generation unit 27 to the moving image distribution device 2 and distribute it to the moving image distribution device 2.

  With reference to FIG. 11, the learning process by the learning part 21 which concerns on embodiment of this invention is demonstrated.

  First, in step S <b> 1, the learning unit 21 specifies an action to be performed by the object O and specifies the position of the object O. The action may be input via the input device 30, may be input from the terminal 3, or may be set in advance as a script. In step S2, the learning unit 21 extracts past learning data corresponding to the action value specified in step S1 and the environment value identifier of the position of the object O from the learning data T.

  In step S3, the learning unit 21 generates the motion data 16 of each operation unit with reference to the past learning data extracted in step S2. Here, the learning unit 21 generates the motion data 16 based on the motion data for which a good evaluation value is obtained from the past learning data T. In addition, the learning unit 21 generates the motion data 16 regardless of the evaluation value at a predetermined ratio such as 50 times or less out of 100 times when the motion data is generated 100 times.

  In step S4, the learning unit 21 tries the movement of each operation unit according to the motion data generated in step S3. In step S5, the learning unit 21 evaluates an action performed by the object O using an evaluation function associated with the action, and updates the learning data T in step S6.

  If learning is continued in step S7, the process returns to step S1. When the learning is finished, the process is finished.

  Learning of the operation unit will be described with reference to FIGS. 12 and 13.

  As shown in FIG. 12A, the object O has a spherical part P1 and limbs provided around the part P1. One of the limbs includes two cylindrical parts P2 and P3. The parts P1 and P2 are connected by a joint N1, and the parts P2 and P3 are connected by a joint N2.

  Since the object O shown in FIG. 12A is composed of a total of eight joints, the motion data 16 shown in FIG. 12B has eight data sets for each joint. In the motion data 16 shown in FIG. 12 (b), the data in the first row indicates the motion of the 0th joint, the data in the second row indicates the motion of the first joint, and a total of 8 rows. Consists of data.

  With reference to FIG.12 (c), the operation | movement matched with each joint is demonstrated. The 0th head description [1, [3, 2, 1]] will be described. [1, [3, 2, 1]] means that the 0th joint tries to be in the state of [3, 2, 1] after 1 second. The state of [3, 2, 1] means that the Euler angle is [3, 2, 1], specifically, [0.3 * (70-30) +30, 0.2 *. (70-30) +30, 0.1 * (70-30) +30]. Here, “30” and “70” are constants set in advance as the movable range of the 0th joint.

FIG. 13A shows a state at time t = 0 and the object O is not trying to move. In FIG. 13A , the joint N2 (0th joint) connects the parts P2 and P3 vertically. On the other hand, FIG. 13B shows the state of the Euler angles [3, 2, 1] set as the target after 1 second in the motion data 16. In FIG. 13B, the joint N2 connects the parts P2 and P3 in a state bent by a predetermined angle.

FIG. 13C shows the result of the object O moving toward the Euler angle [3, 2, 1] state for one second with respect to the 0th joint. Since the speed at which the joint is bent varies depending on the muscle strength and weight of the object O, it does not always reach the Euler angle [3, 2, 1] within one second, and FIG. Indicates that the state has not been reached. In FIG. 13C, the joint N2 connects the parts P2 and P3 at a predetermined angle, but is smaller than the target state angle shown in FIG. 13B.

  In this way, in the motion data 16, an operation target for each time at each joint is set. The learning unit 21 moves each joint according to the motion data 16 and operates toward the next set target according to the motion data 16 even when the target is not reached. The learning unit 21 tries motion targets for all joints and times set in the motion data 16 and evaluates the results.

  Although the operation of bending the joint has been described with reference to FIGS. 12 and 13, the present invention is not limited thereto. For example, when the operation unit has a spring shape, a distance that expands or contracts within a predetermined time may be set as a target in the motion data 16. When the operating unit has a rotating wheel shape, a rotation angle that rotates within a predetermined time may be set as a target in the motion data 16. When the operation unit is a gun, a firing operation when a predetermined condition is satisfied may be set as a target.

  As described above, the learning device 1 according to the embodiment of the present invention learns the object O for each action. Therefore, the learning device 1 operates various actions on the object O and generates the moving image data M, so that the user can observe how the object O operates various actions. Further, by dividing the virtual region V in which the object O is active into the partial regions D, it is not necessary to learn according to various environmental values for each position of the object O, and the amount of calculation required for learning can be reduced.

(Modification)
With reference to FIG. 14 and FIG. 15, the learning system 9a which concerns on the modification of this invention is demonstrated. A learning system 9a shown in FIG. 14 includes a learning device 1a, an activity device 5, a moving image generating device 6, and a moving image distribution device 7. The learning device 1a includes the learning unit 21 and does not include the moving image processing unit 26 in the learning device 1 illustrated in FIG.

  Each of the activity device 5, the moving image generating device 6, and the moving image distribution device 7 is a general computer, and is formed so as to execute a desired process. The activity device 5 and the moving image generation device 6 cooperate with the learning device 1a to visualize the state in which the object O operates according to the motion data generated from the learning data generated in the learning device 1a.

  When the activity device 5 is active in the virtual region V, the activity device 5 tries to move the object O in the virtual region V according to the motion data generated from the learning data generated in the learning device 1a. Draw and display. When the object O is a tangible object, the activity device 5 performs control so that parts corresponding to each operation unit included in the object O move according to the motion data generated from the learning data generated in the learning device 1a.

  When the object O is active in the virtual region V, the moving image generating device 6 attempts to move the object O in the virtual region V according to the motion data generated from the learning data generated in the learning device 1a. Is generated. The moving image generating device 6 corresponds to the moving image generating unit 27 of the learning device 1 shown in FIG. The moving image data M generated by the moving image generating device 6 is transmitted to the moving image distribution device 7, and the moving image data M is distributed from the moving image distribution device 7 to each terminal (not shown), and the moving image data M is reproduced at each terminal. Is done.

Here, the following method can be considered as a method in which the activity device 5 or the moving image generating device 6 performs processing using the learning data T of the learning device 1a. The motion data may be generated by the learning device 1b, or may be generated by the activity device 5 or the moving image generation device 6. In addition, the activity device 5 or the moving image generation device 6 may acquire the result of trying the movement of each operation unit according to the motion data in the learning device 1a, and reproduce the movement of the operation unit according to the result.
(1) The active device 5 or the moving image generating device 6 acquires new motion data generated based on the learning data T from the learning device 1a as shown in FIG. 6 is a method of trying the movement of the motion part of the object O according to the acquired motion data. Here, the new motion data generated by the learning device 1a is obtained by randomly mutating the learning data T regardless of the evaluation.
(2) The activity device 5 or the moving image generating device 6 acquires the learning data T from the learning device 1a, and the active device 5 or the moving image generating device 6 extracts and extracts the motion data from the acquired learning data T. A method of attempting to move the motion part of the object O according to the motion data. Here, the motion data acquired from the learning device 1a is generated according to the motion data for which good evaluation is obtained in the learning data T, and is motion data that can be expected to have good evaluation.

  The activity device 5 and the moving image generating device 6 can be applied to various scenes. For example, by generating and distributing moving image data that activates the object O in the moving image generating device 6, it can be provided as a game for observing how the object is active. It may be a game in which one user learns one object O, or a game in which a plurality of users learn one object O.

  In addition, by generating moving image data that activates a plurality of objects in the virtual region V in the moving image generation device 6, moving image data in which each object moves with individuality can be generated. Such moving image data can be applied to scenes such as movies and videos.

  In this way, by separating the device that generates the learning data T and the device in which the object O is active with reference to the learning data T, the processing load on the activity device 5 and the moving image generation device 6 is reduced. In addition, it is possible to reuse the learning data T created in advance according to the scene.

(Other embodiments)
As described above, the embodiments of the present invention and the modifications thereof have been described. However, it should not be understood that the descriptions and drawings constituting a part of this disclosure limit the present invention. From this disclosure, various alternative embodiments, examples, and operational techniques will be apparent to those skilled in the art.

  For example, the learning device described in the embodiment of the present invention may be configured on one piece of hardware as shown in FIG. 3, or may be configured on a plurality of pieces of hardware according to the functions and the number of processes. May be. Moreover, you may implement | achieve on the existing information processing system.

  It goes without saying that the present invention includes various embodiments not described herein. Therefore, the technical scope of the present invention is defined only by the invention specifying matters according to the scope of claims reasonable from the above description.

DESCRIPTION OF SYMBOLS 1 Learning apparatus 2, 7 Movie delivery apparatus 3 Terminal 5 Activity apparatus 6 Movie generation apparatus 8 Communication network 9 Learning system 10 Storage apparatus 11 Area | region attribute data 12 Environment value data 13 Object attribute data 14 Operation | movement part attribute data 15 Evaluation index data 16 Motion Data 20 processing device 21 learning unit 22 motion generation unit 23 update unit 26 video processing unit 27 video generation unit 28 video distribution unit D partial area M video data O object T learning data V virtual area

Claims (9)

A learning device for learning an object having a moving part,
The object is active in a virtual domain;
The virtual area includes a plurality of partial areas, and each partial area is associated with an environmental value identifier of the partial area,
A motion generation unit that generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of movements of the operation unit to perform the action;
According to the motion data, the movement of the motion part of the object is attempted, and an evaluation value for the action is output. An identifier of the action, an environmental value identifier of a partial area where the object is located, the motion data, and the evaluation value And an update unit that stores them in learning data in association with each other,
The motion generation unit generates new motion data by referring to learning data associated with an identifier of an action to be performed by the object and an environment value identifier of a partial area where the object is located. Learning device. The motion data includes a plurality of pairs in which the motion of the motion unit and the time to try the motion are associated with the motion unit,
The learning apparatus according to claim 1, wherein the update unit outputs the evaluation value after attempting to move based on a plurality of pairs generated by the motion data. The learning apparatus according to claim 1, further comprising: a distribution unit that sequentially generates and distributes moving image data indicating a state in which the object attempts to move the operation unit in accordance with the motion data. A learning method in which an object having a moving part learns,
The object is active in a virtual domain;
The virtual area includes a plurality of partial areas, and each partial area is associated with an environmental value identifier of the partial area,
A step of generating motion data in which a computer associates an identifier of an action to be performed by the object with a target of a combination of movements of the operation unit for performing the action;
The computer attempts to move the motion part of the object according to the motion data, and outputs an evaluation value for the action. The action identifier, the environmental value identifier of the partial area where the object is located, and the motion data And storing the evaluation value in the learning data in association with each other;
The computer includes the step of generating new motion data by referring to learning data associated with an identifier of an action to be performed by the object and an environment value identifier of a partial area where the object is located. How to learn. A learning program for learning an object having a moving part,
The object is active in a virtual domain;
The virtual area includes a plurality of partial areas, and each partial area is associated with an environmental value identifier of the partial area,
Computer
A motion generation unit that generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of movements of the operation unit to perform the action;
According to the motion data, the movement of the motion part of the object is attempted, and an evaluation value for the action is output. An identifier of the action, an environmental value identifier of a partial area where the object is located, the motion data, and the evaluation value And function as an update unit that stores the learning data in association with
The motion generation unit generates new motion data by referring to learning data associated with an identifier of an action to be performed by the object and an environment value identifier of a partial area where the object is located. Learning program. A video distribution device connected to a learning device that learns an object having an operation unit,
The object is active in a virtual domain;
The virtual area includes a plurality of partial areas, and each partial area is associated with an environmental value identifier of the partial area,
The learning device
A motion generation unit that generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of movements of the operation unit to perform the action;
According to the motion data, the movement of the motion part of the object is attempted, and an evaluation value for the action is output. An identifier of the action, an environmental value identifier of a partial area where the object is located, the motion data, and the evaluation value And an update unit that stores them in learning data in association with each other,
The motion generation unit generates new motion data with reference to learning data associated with an identifier of an action to be performed on the object and an environment value identifier of a partial area where the object is located,
The video distribution device
A moving image distribution device, wherein learning data generated by the learning device is used to sequentially generate and distribute moving image data indicating a state in which the object attempts to move the motion unit. An activity device connected to a learning device for learning an object having a motion part,
The object is active in a virtual domain;
The virtual area includes a plurality of partial areas, and each partial area is associated with an environmental value identifier of the partial area,
The learning device
A motion generation unit that generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of movements of the operation unit to perform the action;
According to the motion data, the movement of the motion part of the object is attempted, and an evaluation value for the action is output. An identifier of the action, an environmental value identifier of a partial area where the object is located, the motion data, and the evaluation value And an update unit that stores them in learning data in association with each other,
The motion generation unit generates new motion data with reference to learning data associated with an identifier of an action to be performed on the object and an environment value identifier of a partial area where the object is located,
The active device is
An activity device that operates an object using learning data generated by the learning device. An activity program connected to a learning device for learning an object having a moving part,
The object is active in a virtual domain;
The virtual area includes a plurality of partial areas, and each partial area is associated with an environmental value identifier of the partial area,
The learning device
A motion generation unit that generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of movements of the operation unit to perform the action;
According to the motion data, the movement of the motion part of the object is attempted, and an evaluation value for the action is output. An identifier of the action, an environmental value identifier of a partial area where the object is located, the motion data, and the evaluation value And an update unit that stores them in learning data in association with each other,
The motion generation unit generates new motion data with reference to learning data associated with an identifier of an action to be performed on the object and an environment value identifier of a partial area where the object is located,
The activity program is
On the computer,
An activity program for operating an object using learning data generated by the learning device. A moving image generating device connected to a learning device that learns an object having an operation unit,
The object is active in a virtual domain;
The virtual area includes a plurality of partial areas, and each partial area is associated with an environmental value identifier of the partial area,
The learning device
A motion generation unit that generates motion data in which an identifier of an action to be performed by the object is associated with a target of a combination of movements of the operation unit to perform the action;
According to the motion data, the movement of the motion part of the object is attempted, and an evaluation value for the action is output. An identifier of the action, an environmental value identifier of a partial area where the object is located, the motion data, and the evaluation value And an update unit that stores them in learning data in association with each other,
The motion generation unit generates new motion data with reference to learning data associated with an identifier of an action to be performed on the object and an environment value identifier of a partial area where the object is located,
The moving image generating device includes:
The moving image generating device, wherein the moving image data indicating the state in which the object tries to move the motion unit is sequentially generated using the learning data generated by the learning device.

Images