Support system, support method and support program

Abstract

An input means 81 receives input of observation data observed in association with operation of a device, and input of a cost function in which the factors of the behavior intended by the operator of the device are explanatory variables. A learning means 82 generates the cost function through inverse reinforcement learning using the observation data. A distribution map generation means 83 extracts the weights of the explanatory variables of the generated cost function as feature quantities representing the intention of the operator, and generates a distribution map in which information relating to the cost function is positioned at corresponding positions according to the extracted feature quantities in a multidimensional space having dimensional axes representing the explanatory variables.

Description

[Technical field]

[0001] The present invention relates to an assistance system, an assistance method, and a storage medium for storing an assistance program, which assist in understanding the intention of a user who is operating a device. [Background technology]

[0002] To make a device automatically execute the actions intended by the user, it is necessary to conduct realistic tests and ensure the quality of the device that executes automatically. For example, to realize autonomous driving in the real world, it is necessary to ensure the quality of autonomous driving by conducting tests of patterns that conform to the user's intentions.

[0003] Patent Document 1 describes an intention feature extraction device that extracts features that represent the intention of a target person. The device described in Patent Document 1 extracts weights of explanatory variables learned based on the driving history of the target person as features that represent the target person's driving intention. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2021 / 130916 Summary of the Invention [Problem to be solved by the invention]

[0005] If one were to perform all anticipated tests, there would be many parameters to consider in order to realize automation, and the number of combinations of parameter settings would be enormous, so creating test scenarios manually would require a lot of time. Therefore, in order to create test scenarios more efficiently, it is preferable to be able to accurately grasp the intentions of the user who operates the equipment.

[0006] Therefore, an object of the present invention is to provide an assistance system, an assistance method, and an assistance program that can assist in understanding a user's intention estimated from observation data of a device. [Means for solving the problem]

[0007] The assistance system according to the present invention is characterized by comprising: an input means for accepting an input of observation data observed in association with the operation of a device and an input of a cost function having factors of the behavior intended by the operator of the device as explanatory variables; a learning means for generating a cost function by inverse reinforcement learning using the observation data; and a distribution map generating means for extracting weights of the explanatory variables of the generated cost function as features representing the operator's intention, and for generating a distribution map in which information about the cost function is arranged at corresponding positions in a multidimensional space having the explanatory variables as dimensional axes in accordance with the extracted features.

[0008] The method for assisting according to the present invention comprises the steps of: The computer The method receives input of observation data observed in association with the operation of a device, and input of a cost function having factors of the intended action of the operator of the device as explanatory variables; The computer A cost function is generated by inverse reinforcement learning using observed data. The computer The weights of the explanatory variables of the generated cost function are extracted as features representing the operator's intention, and a distribution diagram is generated in which information about the cost function is arranged at corresponding positions in a multidimensional space with the explanatory variables as dimensional axes, according to the extracted features.

[0009] The assistance program according to the present invention is characterized in that it causes a computer to execute an input process for accepting input of observation data observed in association with the operation of an equipment and an input of a cost function whose explanatory variables are factors of the behavior intended by the operator of the equipment, a learning process for generating a cost function by inverse reinforcement learning using the observation data, and a distribution map generation process for extracting weights of the explanatory variables of the generated cost function as features representing the operator's intention, and generating a distribution map in which information about the cost function is arranged at corresponding positions in a multidimensional space whose dimensional axes are the explanatory variables, in accordance with the extracted features. Effect of the Invention

[0010] According to the present invention, it is possible to assist in understanding a user's intention that is estimated from device observation data. [Brief description of the drawings]

[0011] [Figure 1] 1 is a block diagram showing a configuration example of a first embodiment of a support system according to the present invention; [Diagram 2] FIG. 2 is an explanatory diagram showing an example of observation data. [Diagram 3] FIG. 11 is an explanatory diagram showing an example of a learning process. [Figure 4] FIG. 13 is an explanatory diagram showing an example of a distribution map. [Diagram 5] FIG. 11 is an explanatory diagram showing an example of clustering of each cost function. [Figure 6] 4 is a flowchart showing an operation example of the support system of the first embodiment. [Figure 7] FIG. 11 is a block diagram showing a configuration example of a second embodiment of a support system according to the present invention. [Figure 8] FIG. 11 is an explanatory diagram showing an example of a process for changing a feature amount; [Figure 9] 13 is a flowchart showing an operation example of the support system according to the second embodiment. [Figure 10] 1 is a block diagram showing an overview of a support system according to the present invention; [Figure 11]FIG. 1 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0013] Embodiment 1. 1 is a block diagram showing a configuration example of a first embodiment of a support system according to the present invention. The support system 100 of this embodiment includes a storage unit 10, an input unit 20, a learning unit 30, a distribution map generating unit 40, a clustering unit 50, an identifying unit 60, a scenario generating unit 70, and an output unit 80.

[0014] The storage unit 10 stores various information used for processing by the assistance system 100 of this embodiment. The storage unit 10 may store, for example, data used for learning by the learning unit 30 described below, and a cost function generated as a result of the learning. Additionally, the storage unit 10 may store a scenario generated by the scenario generating unit 70. The storage unit 10 is realized, for example, by a magnetic disk or the like.

[0015] The input unit 20 accepts input of information used for learning by the learning unit 30 described later. Specifically, the input unit 20 accepts input of data observed in association with the operation of an appliance (hereinafter referred to as observed data). Here, the observed data includes not only data observed as a result of operating the appliance, but also data indicating the situation in which the appliance is being operated, data indicating the situation or event that caused the appliance to be operated, data indicating information set in the appliance to be operated, and the like. For example, if the appliance is a vehicle, the observed data is driving data observed in association with the driving of the vehicle.

[0016] The method of obtaining the driving data and the contents of the driving data are arbitrary. For example, various data obtained by a global positioning system (GPS) may be used as the driving data. In addition, when an image is taken while driving a vehicle (such as an image of the front or the back), various information that can be extracted from the image may be used as the driving data.

[0017] FIG. 2 is an explanatory diagram showing an example of observation data. In FIG. 2, driving data is illustrated as a specific example of observation data. For example, items of the vehicle obtained from the GPS include position information (latitude, longitude, altitude), speed (forward / backward, lateral), direction, and driving lane (ego lane). Items of other vehicles obtained by object recognition of images, etc. include object type (car, bus, truck, motorcycle, etc.), relative distance and relative speed from the vehicle, and obstacles. Furthermore, white line detection technology may be used instead of or in combination with object recognition.

[0018] In this way, by using not only items related to the vehicle itself but also items indicating the relationship with other vehicles, it becomes possible to more appropriately learn the user's intentions with respect to other vehicles.

[0019] In this embodiment, the case where the input unit 20 accepts the input of observation data is illustrated as an example, but the learning unit 30, which will be described later, may generate observation data from various types of input information.

[0020] The input unit 20 also accepts input of a cost function in which factors of the behavior intended by the operator of the equipment are used as explanatory variables. These explanatory variables can be derived from observed data, and in the case of operation data, for example, correspond to each item exemplified in FIG. 2. These explanatory variables are selected in advance by an engineer or the like, and a cost function using these explanatory variables is also determined in advance. An example of the cost function is, for example, a function expressed by a linear regression equation of each explanatory variable. The input unit 20 may also accept input of constraint conditions for the cost function together with the cost function.

[0021] The learning unit 30 learns a cost function based on the observation data. More specifically, the learning unit 30 learns a cost function in which factors of the action intended by the operator are explanatory variables by inverse reinforcement learning using the observation data. Note that the method by which the learning unit 30 performs inverse reinforcement learning is arbitrary. The learning unit 30 may generate the cost function by using, for example, the inverse reinforcement learning method described in Patent Document 1. For example, when the device is a vehicle, the learning unit 30 generates the cost function by inverse reinforcement learning using the driving data as described above.

[0022] The learning unit 30 may learn the cost function based on each piece of observation data, or may generate the cost function by inverse reinforcement learning using a group of observation data classified by similar attributes or situations. The learning unit 30 may generate the cost function using only one piece of observation data in which an attribute or situation is clearly indicated.

[0023] For example, in the case of driving data, the learning unit 30 may learn a cost function for each group of driving data classified according to surrounding driving conditions (e.g., cut-in, cut-out, etc.). By using classified observation data or observation data with attributes or conditions clearly indicated, it becomes possible to easily grasp the operator's intention represented by the cost function.

[0024] Fig. 3 is an explanatory diagram showing an example of the learning process. In the example shown in Fig. 3, three observation data groups (driving data A, driving data B, and driving data C) are generated from driving data 101 acquired from actual driving, and cost functions 102, 103, and 104 are generated for each observation data group. In the example shown in Fig. 3, each cost function includes fuel efficiency, vehicle distance, field of view, road shape, and acceleration as explanatory variables, and the explanatory variable surrounded by a double frame has the largest weight.

[0025] As shown in Fig. 3, the driver's intention changes for each driving data. For example, cost function 102 generated based on driving data A indicates that the driver's intention is to prioritize fuel efficiency, and cost function 103 and cost function 104 indicate that the driver's intention is to prioritize visibility. This makes it possible to understand the driver's intention represented by each cost function.

[0026] The distribution map generating unit 40 generates a distribution map in which information related to the generated cost function is arranged in a multidimensional space. Specifically, the distribution map generating unit 40 extracts the weights of explanatory variables of the generated cost function as features representing the operator's intention. Then, the distribution map generating unit 40 generates a distribution map in which information related to the cost function is arranged at positions corresponding to the features in a multidimensional space with the explanatory variables as dimensional axes according to the extracted features.

[0027] The information on the cost function may be any information that allows the contents of the cost function to be understood. The distribution map generating unit 40 may generate a distribution map in which the cost function itself is arranged. In addition, for example, when the cost function is generated using data classified by attributes or situations, the distribution map generating unit 40 may create a distribution map in which information indicating the attributes or situations is arranged as information on the cost function.

[0028] The information on the cost functions is not limited to the above-mentioned content, and for example, a distribution diagram may be created in which an identification number for identifying a cost function, a illustrated cost function as exemplified in FIG. 3, and the identification numbers of the cost functions are arranged.

[0029] Fig. 4 is an explanatory diagram showing an example of a distribution diagram. The distribution diagram shown in Fig. 4 is a distribution diagram in which the illustrated cost functions shown in Fig. 3 are arranged as information related to the cost functions. Specifically, the example shown in Fig. 4 shows that the cost functions showing the driver's intention in a cut-out situation are arranged together, the cost functions showing the driver's intention in a cut-in situation are arranged together except for a few, and the cost functions showing the driver's intention in a sudden stop situation are arranged in a dispersed manner.

[0030] 4 shows the outer frame of the information on the cost function in a manner that allows each situation to be identified. In addition, label information indicating the content of each situation may be placed close to the cost function.

[0031] The clustering unit 50 clusters each cost function using information on the arranged cost function. Specifically, the clustering unit 50 clusters each cost function based on the feature amount of the arranged cost function. There is no particular limitation on the method by which the clustering unit 50 groups the cost functions, and any clustering method (non-hierarchical clustering, for example, k-means, etc.) may be used.

[0032] The clustering unit 50 may reflect the clustering result in the distribution map. The clustering unit 50 may specify a group of clustered cost functions by, for example, surrounding the periphery of information related to the clustered cost functions or reflecting a straight line or a plane that separates an area of ​​the multidimensional space in the distribution map.

[0033] In addition, the distribution map generation unit 40 labels the situation in which the observation data that was the basis for generating each cost function was acquired, making it easier for the user to understand the intended content of the boundary surfaces reflected by the clustering unit 50.

[0034] Fig. 5 is an explanatory diagram showing an example of clustering each cost function in the distribution diagram shown in Fig. 4. Specifically, Fig. 5 shows an example of a distribution diagram in which each clustered cost function is divided by a straight line. The example shown in Fig. 5 shows that, for example, a cost function 105 indicating the driver's intention in a cutout situation shows different characteristics from a cost function indicating the driver's intention in a similar situation.

[0035] The identification unit 60 identifies a predetermined characteristic cost function based on the clustering result. Specifically, the identification unit 60 identifies a typical cost function or a rare cost function for each situation from the clustering result.

[0036] The identification unit 60 may, for example, identify the center of each cluster as a typical cost function. Specifically, for example, when a cluster includes a predetermined ratio or more of cost functions indicating a similar situation, the identification unit 60 may identify a cost function corresponding to the center of the cluster as a cost function indicating a typical intention of the user in that situation. On the other hand, when the ratio of cost functions indicating a certain situation among the cost functions included in the cluster is less than a predetermined ratio, the identification unit 60 may identify the cost function as a cost function indicating a rare intention of the user in that situation.

[0037] In addition, the identification unit 60 may identify a cost function that does not belong to any cluster among cost functions included in a cluster with a classification number less than a predetermined threshold value and cost functions generated based on a manually created scenario (Functional Scenario) as a cost function indicating a rare intention of the user. Furthermore, the identification unit 60 may identify a cost function close to a boundary surface as illustrated in FIG. 5 as a cost function indicating a rare intention of the user. That is, the identification unit 60 may identify a cost function that is located within a predetermined distance from a cluster boundary as a cost function indicating a rare intention of the user.

[0038] Furthermore, it is assumed that the cost function is generated by inverse reinforcement learning using a group of observation data classified by attribute or situation. In this case, the identification unit 60 may identify a cost function that is not included in the maximum proportion of clusters among the cost functions created from a group of observation data classified by the same attribute or situation as a cost function indicating a rare intention of the user. In addition, at this time, the identification unit 60 may identify a cost function that is included in a cluster in which the number of classified clusters is equal to or less than a predetermined number as a cost function indicating a rare intention of the user.

[0039] Note that the cost functions identified by the identification unit 60 are not limited to rare cost functions and typical cost functions. For example, the identification unit 60 may identify a cost function that meets a predetermined condition.

[0040] The scenario generating unit 70 generates a scenario for the equipment using the identified cost function. Note that the equipment scenario here means, for example, the operation of the equipment estimated using the identified cost function, and is time-series data of the operation of the equipment that operates in response to the user's operation. The scenario generating unit 70 may, for example, apply the identified cost function to a simulator to generate the equipment scenario.

[0041] In this way, since a scenario is created using the identified cost function, the support system of this embodiment can be called a scenario creation support system.

[0042] The output unit 80 outputs the distribution map generated by the distribution map generating unit 40 and the scenario generated by the scenario generating unit 70. The output unit 80 may, for example, display the distribution map on a display device (not shown) or store the generated scenario in the storage unit 10.

[0043] The input unit 20, learning unit 30, distribution map generation unit 40, clustering unit 50, identification unit 60, scenario generation unit 70, and output unit 80 are realized by a computer processor (e.g., a CPU (Central Processing Unit), GPU (Graphics Processing Unit)) that operates according to a program (assistance program).

[0044] For example, a program may be stored in the storage unit 10, and the processor may read the program and operate in accordance with the program as the input unit 20, the learning unit 30, the distribution map generating unit 40, the clustering unit 50, the identification unit 60, the scenario generating unit 70, and the output unit 80. Furthermore, each function of the input unit 20, the learning unit 30, the distribution map generating unit 40, the clustering unit 50, the identification unit 60, the scenario generating unit 70, and the output unit 80 may be provided in a SaaS (Software as a Service) format.

[0045] Furthermore, the input unit 20, the learning unit 30, the distribution map generating unit 40, the clustering unit 50, the identifying unit 60, the scenario generating unit 70, and the output unit 80 may each be realized by dedicated hardware. Furthermore, a part or all of the components of each device may be realized by a general-purpose or dedicated circuit, processor, or the like, or a combination of these. These may be configured by a single chip, or may be configured by multiple chips connected via a bus. A part or all of the components of each device may be realized by a combination of the above-mentioned circuits, etc., and a program.

[0046] Furthermore, when some or all of the components of the input unit 20, the learning unit 30, the distribution map generating unit 40, the clustering unit 50, the identifying unit 60, the scenario generating unit 70, and the output unit 80 are realized by a plurality of information processing devices, circuits, etc., the plurality of information processing devices, circuits, etc. may be centrally arranged or distributed. For example, the information processing devices, circuits, etc. may be realized as a client-server system, a cloud computing system, etc., in a form in which each is connected via a communication network.

[0047] Next, the operation of the support system of this embodiment will be described. FIG. 6 is a flowchart showing an example of the operation of the support system 100 of the first embodiment. The input unit 20 accepts input of observation data and a cost function (step S11). The learning unit 30 generates a cost function by inverse reinforcement learning using the observation data (step S12). The distribution map generating unit 40 extracts the weights of explanatory variables of the generated cost function as features (step S13). Then, the distribution map generating unit 40 generates a distribution map in which information about the cost function is arranged at the corresponding position in the multidimensional space according to the extracted features (step S14).

[0048] As described above, in this embodiment, the input unit 20 accepts input of observation data and a cost function, and the learning unit 30 generates a cost function by inverse reinforcement learning using the observation data. Then, the distribution map generating unit 40 extracts the weights of explanatory variables of the generated cost function as features, and generates a distribution map in which information about the cost function is arranged at corresponding positions in the multidimensional space according to the extracted features. The information about the cost function arranged on the distribution map reflects the user's intention, and therefore can assist in understanding the user's intention estimated from the observation data of the device.

[0049] Embodiment 2. Next, a second embodiment of the support system according to the present invention will be described. Fig. 7 is a block diagram showing a configuration example of the second embodiment of the support system according to the present invention. The support system 200 of this embodiment includes a storage unit 10, an input unit 20, a learning unit 30, a distribution map generating unit 40, a clustering unit 50, an identifying unit 60, a feature amount changing unit 110, a scenario generating unit 70, and an output unit 80.

[0050] That is, the support system 200 of this embodiment differs from the support system 100 of the first embodiment in that it further includes a feature amount change unit 110. The other configurations are the same as those of the first embodiment.

[0051] The feature amount modification unit 110 modifies the feature amount of the cost function identified by the identification unit 60. For example, when the cost function is expressed by a linear regression equation of explanatory variables, the feature amount modification unit 110 modifies the weight of each explanatory variable. The feature amount modification unit 110 may, for example, modify the weight (feature amount) of explanatory variables other than the explanatory variable of interest (emphasis). For example, it is desired to modify the feature amount by placing emphasis on fuel efficiency among the explanatory variables of the cost function illustrated in FIG. 3. In this case, the feature amount modification unit 110 may normalize the weight and modify the feature amounts other than fuel efficiency.

[0052] Thereafter, the process in which the scenario generating unit 70 generates a scenario for an equipment using the cost function with the changed feature amount is the same as in the first embodiment. That is, the scenario generating unit 70 generates a scenario for an equipment using the cost function with the changed feature amount. In this way, it becomes possible to create a desired scenario from a cost function that indicates the intention of a focused operator.

[0053] FIG. 8 is an explanatory diagram showing an example of a process for changing a feature quantity. For example, a case where a fuel efficiency-oriented scenario is generated is assumed. First, the identification unit 60 identifies a cost function that indicates the intention of an operator who values ​​fuel efficiency. At this time, as illustrated in FIG. 5, a cost function 105 for a rare case may be identified. Next, the feature quantity change unit 110 sets the weight of fuel efficiency to 1 by normalization, and appropriately changes the weights of the other explanatory variables. The feature quantity change unit 110 may change the weights in any manner, for example, at a fixed rate, at fixed intervals, up to the vicinity of a boundary value, and the like.

[0054] Then, the scenario generation unit 70 generates a scenario using the changed cost function. For example, by changing the feature amount of the cost function for a rare case, it becomes possible to create many scenarios assuming rare cases.

[0055] The input unit 20, learning unit 30, distribution map generation unit 40, clustering unit 50, identification unit 60, feature modification unit 110, scenario generation unit 70, and output unit 80 are realized by a computer processor that operates in accordance with a program (assistance program).

[0056] Next, the operation of the support system of this embodiment will be described. Fig. 9 is a flowchart showing an example of the operation of the support system 200 of the second embodiment. Note that the process of learning the cost function based on the observation data, extracting the feature amount, and generating the distribution diagram is similar to the process from step S11 to step S14 illustrated in Fig. 6.

[0057] The clustering unit 50 clusters the arranged cost functions based on their feature quantities (step S21). The identification unit 60 identifies a predetermined characteristic cost function based on the clustering result (step S22). The feature quantity modification unit 110 modifies the feature quantities of the identified cost function (step S23). The scenario generation unit 70 generates a scenario for the device using the cost function with the feature quantity modified (step S24). The output unit 80 outputs the generated scenario (step S25).

[0058] As described above, in this embodiment, in addition to the configuration of the first embodiment, the feature modification unit 110 modifies the feature of the cost function identified by the identification unit 60, and the scenario generation unit 70 generates a scenario for the equipment using the cost function with the modified feature. Therefore, in addition to the effects of the first embodiment, it becomes possible to create many scenarios assuming various cases based on the feature of interest.

[0059] Next, an outline of the present invention will be described. Fig. 10 is a block diagram showing an outline of the assistance system according to the present invention. The assistance system 1 according to the present invention includes an input means 81 (e.g., input unit 20) that receives an input of observation data observed accompanying the operation of a device (e.g., driving a vehicle) and an input of a cost function having factors of an action intended by an operator (e.g., a driver) of the device as explanatory variables, a learning means 82 (e.g., learning unit 30) that generates a cost function by inverse reinforcement learning using the observation data, and a distribution map generating means 83 (e.g., distribution map generating unit 40) that extracts weights of explanatory variables of the generated cost function as feature quantities representing the intention of the operator, and generates a distribution map in which information on the cost function is arranged at corresponding positions in a multidimensional space with explanatory variables as dimensional axes according to the extracted feature quantities.

[0060] Such a configuration can assist in understanding the user's intentions, which can be inferred from device observation data.

[0061] The support system 1 may also include a clustering unit (e.g., the clustering unit 50) that clusters the arranged cost functions based on the feature quantities of the cost functions, and a specification unit (e.g., the specification unit 60) that specifies a predetermined characteristic cost function based on the result of the clustering. With such a configuration, it is possible to support understanding of the distribution tendency of the cost functions.

[0062] Specifically, the identifying unit may identify a cost function that is included in a cluster in which the number of classified functions is less than a predetermined threshold, or a cost function that does not belong to any cluster. With such a configuration, a cost function that indicates a rare intent can be identified.

[0063] Alternatively, the identifying unit may identify a cost function that is located within a predetermined distance from a cluster boundary. With such a configuration, rare cost functions that are far from the cluster center can be identified even if they are present in the same cluster.

[0064] The learning means 82 may generate a cost function by inverse reinforcement learning using a group of observation data classified by attribute or situation. The identifying means may identify, among the cost functions created from a group of observation data classified by the same attribute or situation, a cost function that is not included in the maximum proportion of clusters, or a cost function that is included in a cluster where the number of clusters included is a predetermined number or less. With such a configuration, a cost function that indicates a rare intention for each attribute or situation can be identified.

[0065] Alternatively, the identifying means may identify a cost function corresponding to the center of a cluster. Such a configuration makes it possible to identify a typical cost function.

[0066] Here, the cost function may be defined as a linear regression equation of the explanatory variables.

[0067] The support system 1 may also include a scenario generating means (e.g., a scenario generating unit 70) that generates a scenario for the device using the identified cost function. With such a configuration, it becomes possible to create many scenarios assuming various cases.

[0068] Furthermore, the support system 1 may include a feature amount changing means (e.g., the feature amount changing unit 110) for changing the feature amount of the cost function. Then, the scenario generating means may generate a scenario for the device using the cost function with the changed feature amount. With such a configuration, it becomes possible to create many scenarios assuming various cases based on the feature amount of interest.

[0069] 11 is a schematic block diagram showing the configuration of a computer according to at least one embodiment. A computer 1000 includes a processor 1001, a main memory device 1002, an auxiliary memory device 1003, and an interface 1004.

[0070] The above-mentioned assistance system 1 is implemented in a computer 1000. The operations of the above-mentioned processing units are stored in the form of a program (assistance program) in an auxiliary storage device 1003. The processor 1001 reads the program from the auxiliary storage device 1003, loads it in the main storage device 1002, and executes the above-mentioned processing in accordance with the program.

[0071] In at least one embodiment, the auxiliary storage device 1003 is an example of a non-transitory tangible medium. Other examples of non-transitory tangible media include a magnetic disk, a magneto-optical disk, a CD-ROM (Compact Disc Read-only memory), a DVD-ROM (Read-only memory), and a semiconductor memory connected via the interface 1004. In addition, when this program is distributed to the computer 1000 via a communication line, the computer 1000 that receives the program may load the program into the main storage device 1002 and execute the above-mentioned process.

[0072] The program may be for realizing part of the above-mentioned functions. Furthermore, the program may be a so-called differential file (differential program) that realizes the above-mentioned functions in combination with another program already stored in the auxiliary storage device 1003.

[0073] A part or all of the above-described embodiments can be described as, but is not limited to, the following supplementary notes.

[0074] (Appendix 1) An input means for receiving an input of observation data observed in association with the operation of a device and an input of a cost function having factors of an action intended by an operator of the device as explanatory variables; A learning means for generating the cost function by inverse reinforcement learning using the observation data; a distribution map generating means for extracting weights of explanatory variables of the generated cost function as a feature amount representing the intention of the operator, and generating a distribution map in which information about the cost function is arranged at a corresponding position in a multidimensional space having the explanatory variables as dimensional axes in accordance with the extracted feature amount. A support system comprising:

[0075] (Appendix 2) A clustering means for clustering the arranged cost functions based on the feature quantities of the cost functions; and a specifying means for specifying a predetermined characteristic cost function based on the result of the clustering. Assistance system according to claim 1.

[0076] (Appendix 3) The identification means identifies cost functions that are included in clusters in which the number of classifications is less than a predetermined threshold, or cost functions that do not belong to any cluster. Assistance system according to claim 2.

[0077] (Appendix 4) The identifying means identifies a cost function that is located within a predetermined distance from a cluster boundary. Assistance system according to claim 2 or claim 3.

[0078] (Appendix 5) The learning means generates a cost function by inverse reinforcement learning using a group of observed data classified by attributes or situations, The identifying means identifies, among the cost functions created from the observation data group classified by the same attribute or situation, a cost function that is not included in the maximum proportion of clusters, or a cost function that is included in a cluster where the number of clusters included in the classified clusters is equal to or less than a predetermined number. Assistance system according to any one of claims 2 to 4.

[0079] (Appendix 6) The identification means identifies a cost function corresponding to the center of a cluster. Assistance system according to claim 2.

[0080] (Appendix 7) The cost function is defined as a linear regression equation of the explanatory variables. Assistance system according to any one of claims 1 to 6.

[0081] (Appendix 8) A scenario generation means for generating a scenario for an equipment using the specified cost function is provided. 8. An assistance system according to any one of claims 1 to 7.

[0082] (Appendix 9) A feature amount changing means for changing the feature amount of the cost function is provided, The scenario generating means generates a scenario for the device by using the cost function in which the feature amount has been changed. Assistance system according to claim 8.

[0083] (Appendix 10) Accepting input of observation data observed in association with the operation of a device and input of a cost function having factors of the intended action of the operator of the device as explanatory variables; generating the cost function by inverse reinforcement learning using the observation data; The weights of explanatory variables of the generated cost function are extracted as features representing the intention of the operator, and a distribution diagram is generated in which information about the cost function is arranged at corresponding positions in a multidimensional space having the explanatory variables as dimensional axes in accordance with the extracted features. A support method comprising:

[0084] (Appendix 11) Clustering the cost functions based on the features of the arranged cost functions; Identifying a predetermined characteristic cost function based on the result of the clustering. Support methods described in Appendix 10.

[0085] (Appendix 12) To the computer, an input process for receiving an input of observation data observed in association with the operation of the device and an input of a cost function having factors of an action intended by the operator of the device as explanatory variables; A learning process for generating the cost function by inverse reinforcement learning using the observation data; and A distribution map generation process for extracting weights of explanatory variables of the generated cost function as features representing the intention of the operator, and generating a distribution map in which information about the cost function is arranged at corresponding positions in a multidimensional space having the explanatory variables as dimensional axes in accordance with the extracted features. A storage medium for storing an assistance program for executing the above.

[0086] (Appendix 13) To the computer, A clustering process for clustering the arranged cost functions based on the feature quantities of the cost functions; and Executing a specification process for specifying a predetermined characteristic cost function based on the result of the clustering. 13. The storage medium of claim 12, storing an assistance program for performing the method.

[0087] (Appendix 14) To the computer, an input process for receiving an input of observation data observed in association with the operation of the device and an input of a cost function having factors of an action intended by the operator of the device as explanatory variables; A learning process for generating the cost function by inverse reinforcement learning using the observation data; and A distribution map generation process for extracting weights of explanatory variables of the generated cost function as features representing the intention of the operator, and generating a distribution map in which information about the cost function is arranged at corresponding positions in a multidimensional space having the explanatory variables as dimensional axes in accordance with the extracted features. A support program to help implement the above.

[0088] (Appendix 15) To the computer, A clustering process for clustering the arranged cost functions based on the feature quantities of the cost functions; and Executing a specification process for specifying a predetermined characteristic cost function based on the result of the clustering. Support program described in Appendix 14.

[0089] Although the present invention has been described above with reference to the embodiment, the present invention is not limited to the above embodiment. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention. [Explanation of symbols]

[0090] 10 Storage section 20 Input section 30 Learning Department 40 Distribution map generation section 50 Clustering Department 60 Specific section 70 Scenario Generation Section 80 Output section 100,200 Support System 110 Feature quantity change unit

Claims

1. an input means for receiving an input of observation data observed in association with an operation of a device, and an input of a cost function having factors of an action intended by an operator of the device as explanatory variables; A learning means for generating the cost function by inverse reinforcement learning using the observation data; a distribution map generating means for extracting weights of explanatory variables of the generated cost function as a feature amount representing the intention of the operator, and generating a distribution map in which information about the cost function is arranged at a corresponding position in a multidimensional space having the explanatory variables as dimensional axes in accordance with the extracted feature amount. A support system comprising:

2. A clustering means for clustering the arranged cost functions based on the feature quantities of the cost functions; and a specifying means for specifying a predetermined characteristic cost function based on the result of the clustering. The support system according to claim 1.

3. The identifying means identifies a cost function included in a cluster in which the number of classified functions is less than a predetermined threshold, or a cost function that does not belong to any cluster. The support system according to claim 2.

4. The specifying means specifies a cost function that is located within a predetermined distance from a cluster boundary. The support system according to claim 2 or 3.

5. The learning means generates a cost function by inverse reinforcement learning using a group of observed data classified by attributes or situations; The identifying means identifies, among the cost functions created from the group of observation data classified by the same attribute or situation, a cost function that is not included in the maximum proportion of clusters, or a cost function that is included in a cluster where the number of clusters included in the classified clusters is equal to or less than a predetermined number.

5. The assistance system according to claim 2.

6. The identifying means identifies a cost function corresponding to the center of the cluster. The support system according to claim 2.

7. The cost function is defined as a linear regression equation of the explanatory variables.

7. An assistance system according to any one of claims 1 to 6.

8. A scenario generating means is provided for generating a scenario for the device using the specified cost function.

8. An assistance system according to any one of claims 1 to 7.

9. A computer receives an input of observation data observed in association with the operation of a device, and an input of a cost function having factors of an action intended by an operator of the device as explanatory variables; the computer generates the cost function by inverse reinforcement learning using the observation data; The computer extracts weights of explanatory variables of the generated cost function as features representing the intention of the operator, and generates a distribution diagram in which information about the cost function is arranged at corresponding positions in a multidimensional space having the explanatory variables as dimensional axes in accordance with the extracted features. A support method comprising:

10. On the computer, an input process for receiving an input of observation data observed in association with the operation of the device and an input of a cost function having factors of an action intended by the operator of the device as explanatory variables; A learning process for generating the cost function by inverse reinforcement learning using the observation data; and A distribution map generation process for extracting weights of explanatory variables of the generated cost function as features representing the intention of the operator, and generating a distribution map in which information about the cost function is arranged at corresponding positions in a multidimensional space with the explanatory variables as dimensional axes in accordance with the extracted features. A support program to help implement the above.

Images