Information processing device, shape difference model generation method, estimation method, and estimation program

Abstract

To provide a technology that accurately estimates a change in the surface shape of sediment while preventing a calculation amount.SOLUTION: An information processing device (1) comprises a shape information acquisition unit (102) that acquires shape information indicating the surface shape of given sediment before the occurrence of a given event that changes the surface shape of the sediment, and a difference estimation unit (103) that estimates a difference of the surface shape before and after the occurrence of the given event using a shape difference model (111) in which the change in the surface shape by the given event is modeled and the acquired shape information.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to an information processing device or the like that estimates changes in the surface shape of deposits. [Background technology]

[0002] Various types of waste, such as household garbage, are collected at treatment facilities, temporarily stored in waste pits set up at the treatment facilities, and then disposed of by incineration, etc. In order to reduce the labor required to operate such waste pits, efforts are being made to automate the operation of cranes that transport waste within the waste pits.

[0003] For example, Patent Document 1 below discloses a conveying device that controls the operation of a crane using distance data indicating the distance to the surface of the waste measured by a distance measuring means that uses optical pulses. This conveying device makes it possible to realize short-term automatic operation control, such as grabbing waste at high heights with the crane bucket and dumping it into a waste hopper. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2000-143151 Summary of the Invention [Problem to be solved by the invention]

[0005] The conveying device of Patent Document 1 has room for improvement in that it is difficult to achieve optimal operation control over the long term. However, to achieve optimal operation control over the long term, it is necessary to estimate how the surface shape of the waste in the pit, which is expressed in a high-dimensional data format, will change. Furthermore, in order to accurately estimate the changes in the high-dimensional data, it is necessary to generate a highly accurate estimation model by performing learning using a large amount of training data. Even if an estimation model of the surface shape can be created through such learning, there is a problem in that the amount of calculation required for the estimation becomes excessive when using such an estimation model. This problem is not limited to waste in pits, but is a common problem when handling sediments whose surface shape changes due to some event.

[0006] An object of one aspect of the present invention is to provide an information processing device or the like that can accurately estimate changes in the surface shape of deposits while reducing the amount of calculation. [Means for solving the problem]

[0007] In order to solve the above problem, an information processing device according to one aspect of the present invention includes a shape information acquisition unit that acquires shape information indicating the surface shape of a specified deposit before the occurrence of a specified event that changes the surface shape of the specified deposit, and a difference estimation unit that estimates the difference in the surface shape before and after the occurrence of the specified event using a shape difference model that models the change in surface shape due to the specified event and the shape information.

[0008] In order to solve the above-mentioned problems, a shape difference model generation method according to one embodiment of the present invention is a shape difference model generation method executed by at least one information processing device, and includes: a data acquisition step of acquiring training data that associates shape information indicating the surface shape of a predetermined event that changes the surface shape of a predetermined deposit before the occurrence of the predetermined event with a difference in the surface shape before and after the occurrence of the predetermined event; and a learning step of generating, by machine learning using the training data, a shape difference model that outputs the difference in surface shape before and after the occurrence of the predetermined event according to the input shape information.

[0009] In order to solve the above problem, an estimation method according to one aspect of the present invention is an estimation method executed by at least one information processing device, and includes a shape information acquisition step of acquiring shape information indicating the surface shape of a specified deposit before the occurrence of a specified event that changes the surface shape of the specified deposit, and an estimation step of estimating the difference in the surface shape before and after the occurrence of the specified event using a shape difference model that models the change in surface shape due to the specified event and the shape information. [Effects of the Invention]

[0010] According to one aspect of the present invention, it is possible to accurately estimate changes in the surface shape of deposits while reducing the amount of calculation. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a diagram illustrating an example of the configuration of an estimation system according to an embodiment of the present invention. [Figure 2] 2 is a block diagram showing an example of a configuration of a main part of an information processing device included in the estimation system shown in FIG. 1. FIG. [Figure 3] 2 is a block diagram showing an example of a configuration of a main part of a learning device included in the estimation system shown in FIG. 1. FIG. [Figure 4] FIG. 10 is a diagram showing an example of calculation of the difference in surface shape of waste material. [Figure 5] FIG. 10 is a diagram illustrating an example of automatically generating an operation schedule. [Figure 6] 1 is a flowchart showing the flow of a method for generating a shape difference model according to an embodiment of the present invention. [Figure 7] 3 is a flowchart showing a flow of processing executed by the information processing device shown in FIG. 2 when estimating a difference in surface shape. [Figure 8] 3 is a flowchart showing a flow of processing executed by the information processing device shown in FIG. 2 when recommending an operation schedule. DETAILED DESCRIPTION OF THE INVENTION

[0012] [System Configuration] An overview of an estimation system 5 according to one embodiment of the present invention will be described with reference to FIG. 1. FIG. 1 is a diagram showing an example of the configuration of the estimation system 5. The estimation system 5 is a system equipped with a function for estimating the surface shape of a deposit. The estimation system 5 shown in FIG. 1 includes an information processing device 1, a learning device 2, and a display device 3. Both the information processing device 1 and the learning device 2 can be realized by a general-purpose computer (which can also be referred to as an information processing device). Note that if the information processing device 1 has a function for displaying information, the display device 3 can be omitted.

[0013] The information processing device 1 uses shape information indicating the surface shape of a given deposit before a given event that changes the surface shape of the given deposit occurs to estimate the difference in surface shape between before and after the event. For this estimation, a shape difference model that models the change in surface shape due to the given event is used.

[0014] Generally, changes in the surface shape of piles are not easily affected by the height at which the change occurs, and are localized. For example, when waste piled up in a pit is picked up by a crane bucket, the surface shape of the waste around the position where the picking operation occurred becomes concave, while the surface shape of the waste in other areas remains largely unchanged. Furthermore, assuming that the condition of the piled waste (e.g., the degree of compaction, etc.) remains the same, the effect of the pile height of the waste at the position where the picking operation occurred on the surface shape of the waste after the picking operation is almost nonexistent, or even if it is, it is negligible.

[0015] Therefore, as described above, the information processing device 1 employs a configuration in which a shape difference model that models changes in the surface shape of a deposit due to a predetermined event that changes the surface shape of the deposit is used to estimate the difference in the surface shape before and after the occurrence of the predetermined event. This configuration, which estimates the difference in the surface shape before and after a change rather than the surface shape after the change, makes it possible to accurately estimate changes in the surface shape of the deposit while reducing the amount of calculation.

[0016] When a predetermined event actually occurs, the information processing device 1 can estimate the difference in the surface shape before and after the occurrence of the predetermined event. Also, assuming that a predetermined event has occurred, the information processing device 1 can estimate the difference in the surface shape before and after the occurrence of the predetermined event, that is, predict the difference in the surface shape.

[0017] Below, an example of estimating changes in the surface shape of waste (for example, general household waste to be incinerated) stored in a pit installed in a waste treatment facility will be described. However, the deposits for which changes in surface shape are to be estimated are not limited to waste, as long as they consist of multiple objects (which may all be of the same type, or may include multiple types of objects) that form a continuous surface. Other examples of deposits for which changes in surface shape are to be estimated include, for example, incineration ash and gravel generated by incinerating waste. In the following description, "waste" can be read as any "deposit."

[0018] Furthermore, the "predetermined event" may be any event that changes the surface shape of the pile. For example, in the case of waste stored in a pit, the "predetermined event" may be the action of lifting up the waste with the bucket of a crane (bucket crane) installed in the pit and dropping the lifted waste into the pit. In addition, for example, the predetermined event may be the dumping of waste into the pit from a compactor truck that transports waste to a waste treatment facility.

[0019] The manner in which the surface shape changes varies depending on the predetermined event. For this reason, it is preferable to prepare a shape difference model for each "predetermined event." This allows the information processing device 1 to obtain appropriate estimation results by applying a shape difference model that corresponds to the predetermined event that has occurred.

[0020] The learning device 2 generates the above-mentioned shape difference model through machine learning. Details of the shape difference model and the method for generating it will be described later.

[0021] The display device 3 displays various information related to the estimation system 5. For example, in the example of Fig. 1, information indicating the pile height of waste at each location in the pit after the grabbing operation with the bucket is performed is displayed on the display device 3.

[0022] [Configuration of information processing device] A more detailed configuration of the information processing device 1 will be described with reference to Fig. 2. Fig. 2 is a block diagram showing an example of the configuration of the main parts of the information processing device 1. As shown in the figure, the information processing device 1 includes a control unit 10 that controls each unit of the information processing device 1, and a storage unit 11 that stores various data used by the information processing device 1. The information processing device 1 also includes a communication unit 12 that enables the information processing device 1 to communicate with other devices, an input unit 13 that accepts input to the information processing device 1, and an output unit 14 that enables the information processing device 1 to output various data.

[0023] The control unit 10 also includes an event identification unit 101, a shape information acquisition unit 102, a difference estimation unit 103, a surface shape estimation unit 104, an operation schedule generation unit 105, an evaluation unit 106, and an operation schedule recommendation unit 107. The storage unit 11 stores a shape difference model 111 and a shape map 112. Of the components included in the control unit 10, the operation schedule generation unit 105 and the following components will be described later in the section "Generation of an Operation Schedule."

[0024] The event identification unit 101 identifies an event that has caused a difference in the surface shape to be estimated, or an event that will cause this difference, from among a plurality of predetermined events that change the surface shape of waste stored in the pit. In this embodiment, the event identification unit 101 identifies either an event in which waste in the pit is picked up by the bucket of a crane (hereinafter referred to as a "grabbing event"), or an event in which the waste picked up by the bucket of a crane is dropped into the pit (hereinafter referred to as a "dropping event").

[0025] As described above, the predetermined event can be set arbitrarily and is not limited to these examples. Furthermore, even with the same gripping operation, the manner and extent of change in the surface shape may differ depending on the state of the waste being gripped. For example, waste accumulated deep in a pit may be compacted by other waste piled on top of it. When a gripping operation is performed on such compacted waste, the change in the surface shape tends to be smaller than when a gripping operation is performed on uncompacted waste. For this reason, different shape difference models 111 may be generated depending on the state of the waste being gripped, the height of the crane bucket when the gripping operation is performed, and the like. By using multiple shape difference models 111 generated in this way, more accurate estimation can be performed.

[0026] The shape information acquisition unit 102 acquires shape information indicating the surface shape of the waste stored in the pit before the occurrence of a predetermined event that changes the surface shape of the waste. The shape information acquisition unit 102 may acquire shape information indicating the surface shape of the waste in the entire pit, or may acquire shape information indicating the surface shape of the waste in a portion of the pit. In the latter case, the shape information acquisition unit 102 may determine which area of the waste to acquire shape information indicating the surface shape of depending on the event that changed the surface shape. For example, if the event identified by the event identification unit 101 is a grabbing event, the shape information acquisition unit 102 may identify the location where the crane bucket is (or was) performing the action of grabbing the waste and set a region of a predetermined size centered on that location. Then, the shape information acquisition unit 102 may acquire shape information indicating the surface shape of the waste in the identified region.

[0027] The difference estimation unit 103 estimates the difference in surface shape before and after the occurrence of a predetermined event using a shape difference model 111 that models changes in the surface shape due to the predetermined event and the shape information acquired by the shape information acquisition unit 102. Specifically, the difference estimation unit 103 inputs the shape information acquired by the shape information acquisition unit 102 to the shape difference model 111. As a result, information indicating the difference in surface shape before and after the occurrence of the predetermined event is output from the shape difference model 111.

[0028] As described above, the shape difference model 111 is a model for estimating the difference in surface shape before and after the occurrence of a predetermined event, and is a model of the change in surface shape due to the predetermined event. As will be described in detail later, the shape difference model 111 is generated by the learning device 2. The shape difference model 111 may be stored in the storage unit 11 by the time estimation is performed by the difference estimation unit 103. Note that while one shape difference model 111 is shown in FIG. 1, it is preferable to prepare a shape difference model 111 for each predetermined event, as described above. When a shape difference model 111 is prepared for each predetermined event, the difference estimation unit 103 estimates the difference in surface shape before and after the occurrence of the predetermined event using the shape difference model 111 corresponding to the event identified by the event identification unit 101.

[0029] The surface shape estimation unit 104 estimates the surface shape of the waste after the occurrence of a predetermined event, using the difference between the shape information acquired by the shape information acquisition unit 102 and the surface shape estimated by the difference estimation unit 103. The estimation result by the surface shape estimation unit 104 is reflected in the shape map 112.

[0030] The shape map 112 is information indicating the surface shape of the waste at each position within the pit. The form in which the surface shape of the waste stored in the pit is represented is arbitrary. For example, the inside of the pit viewed from above may be divided into multiple grid-like sections, and the surface shape of the waste stored in the pit may be represented by the pile height of the waste in each section. For example, if the inside of the pit is divided into 150 sections (15 x 10), the surface shape of the waste in the pit is represented by the combination of the pile heights of the waste in each of these 150 sections. In this case, the shape map 112 becomes information indicating the pile height of the waste in each of the 150 sections. The shape map 112 can be generated, for example, from the measurement results of measuring the distance to the waste in the pit using a distance sensor placed above the pit.

[0031] Note that the smaller the size of one section, the more detailed the surface shape can be represented, but the cost of learning the shape difference model 111 and the cost of calculations using the shape difference model 111 also increase. For this reason, the size of one section can be set depending on the accuracy of the surface shape to be obtained, the calculation capacity of the information processing device 1, etc.

[0032] Here, in the operation of a crane in a pit for storing waste, conventionally, sections of a size corresponding to the size of the bucket are set within the pit, an address is assigned to each section, and commands specifying the address are sent to the crane to perform operations such as grabbing and dropping. In contrast, the estimation system 5 can manage the surface shape of the waste within the pit in units of sections that can be set arbitrarily, as described above. In other words, the estimation system 5 does not require the conventional transport control that depends on sections, and can more flexibly specify operation positions for grabbing and dropping. For example, in a small section that does not have an assigned address, as described above, if the surface shape of the waste is convex, the estimation system 5 can also control the grabbing operation to be performed on the convex part.

[0033] As described above, the information processing device 1 includes a shape information acquisition unit 102 that acquires shape information indicating the surface shape of the waste stored in the pit before the occurrence of a predetermined event that changes the surface shape of the waste, and a difference estimation unit 103 that estimates the difference in surface shape before and after the occurrence of the predetermined event using a shape difference model 111 that models the change in surface shape due to the predetermined event and the shape information acquired by the shape information acquisition unit 102. This makes it possible to accurately estimate the change in the surface shape of the waste while reducing the amount of calculation.

[0034] As described above, the information processing device 1 also includes a surface shape estimation unit 104 that estimates the surface shape of the waste after the occurrence of a predetermined event, using the shape information acquired by the shape information acquisition unit 102 and the difference in the surface shape estimated by the difference estimation unit 103. This makes it possible to display the surface shape of the waste after the occurrence of a predetermined event, and to create a crane operation plan based on that surface shape.

[0035] [Configuration of the learning device] A more detailed configuration of the learning device 2 will be described with reference to FIG. 3. FIG. 3 is a block diagram showing an example of the configuration of the main parts of the learning device 2. As shown in the figure, the learning device 2 includes a control unit 20 that controls each unit of the learning device 2, and a memory unit 21 that stores various data used by the learning device 2. The learning device 2 also includes a communication unit 22 that enables the learning device 2 to communicate with other devices, an input unit 23 that accepts input to the learning device 2, and an output unit 24 that enables the learning device 2 to output various data. The control unit 20 also includes a data acquisition unit 201 and a learning unit 202. The memory unit 21 shown in FIG. 3 stores a shape difference model 111 generated by the learning unit 202.

[0036] The data acquisition unit 201 acquires training data that corresponds shape information indicating the surface shape before a specified event occurs that changes the surface shape of waste stored in a pit with the difference in surface shape before and after the occurrence of the specified event.

[0037] The learning unit 202 generates a shape difference model that outputs the difference in surface shape before and after the occurrence of a predetermined event according to the input shape information, through machine learning using the training data acquired by the data acquisition unit 201. The generated shape difference model is stored in the storage unit 21 as a shape difference model 111. This shape difference model 111 is then provided to the information processing device 1 and used to predict differences in surface shape.

[0038] As described above, a shape difference model 111 may be prepared for each predetermined event. In this case, the data acquisition unit 201 acquires training data corresponding to each event. Then, the learning unit 202 generates a shape difference model 111 corresponding to each event using the training data corresponding to each event.

[0039] For example, the data acquisition unit 201 may acquire training data that associates the difference in surface shape before and after the occurrence of the above-mentioned gripping event with shape information indicating the surface shape before the occurrence of the gripping event. The learning unit 202 can generate a shape difference model 111 (hereinafter also referred to as the "shape difference model 111 for gripping event") that estimates the difference in surface shape before and after the occurrence of the gripping event through machine learning using this training data.

[0040] Furthermore, for example, the data acquisition unit 201 may acquire training data that associates the difference in surface shape before and after the drop event with shape information indicating the surface shape before the drop event. The learning unit 202 can generate a shape difference model 111 (hereinafter also referred to as a "shape difference model for a drop event") that estimates the difference in surface shape before and after the drop event through machine learning using this training data.

[0041] As described above, the learning device 2 includes a data acquisition unit 201 that acquires training data that associates shape information indicating the surface shape of waste stored in a pit before the occurrence of a specified event that changes the surface shape of the waste with the difference in the surface shape before and after the occurrence of the specified event, and a learning unit that generates, through machine learning using the acquired training data, a shape difference model 111 that outputs the difference in surface shape before and after the occurrence of the specified event according to the input shape information.

[0042] Here, as described above, it is possible to accurately estimate differences in the surface shapes of waste while reducing the amount of calculations by using the shape difference model 111. Therefore, the learning device 2 has the effect of being able to accurately estimate differences in the surface shapes of waste while reducing the amount of calculations.

[0043] Any algorithm, including known algorithms, can be applied as the machine learning algorithm of the shape difference model 111. For example, the shape difference model 111 may be a convolutional neural network model. In this case, the learning unit 202 can generate the shape difference model 111 by, for example, a gradient method or the like.

[0044] [Differences in the surface shape of waste] The difference in the surface shape of the waste can be quantified by subtracting the numerical representation of the surface shape of the waste after the occurrence of a specified event from the numerical representation of the surface shape of the waste before the occurrence of the specified event. This will be explained with reference to Figure 4. Figure 4 is a diagram showing an example of calculating the difference in the surface shape of the waste.

[0045] 401a and 401 in Fig. 4 show the state of the pit before a predetermined event (specifically, a grabbing event) occurs. More specifically, 401a shows the pit as viewed from above (looking down from above). Meanwhile, 401 shows the height of the waste pile in each section included in the row indicated by the white arrow in 401a by the height of a rectangle placed at the position corresponding to each section.

[0046] As shown in 401a, in the example of Figure 4, the inside of the pit is divided into 150 rectangular sections, each 15 wide and 10 high. Note that these sections are used for convenience in representing the surface shape, and no physical divisions actually exist within the pit. The surface shape of the waste in the pit is represented by the pile height of the waste in each of these sections (the distance from the surface of the waste to the bottom of the pit; this can also be expressed as depth). As mentioned above, information indicating the pile height of the waste in each of these 150 sections may be used as the shape map 112.

[0047] Here, assume that a grabbing event occurs at position P1 shown in 401a and 401. In this case, a target area AR1 of a predetermined size is set at position P1 where the grabbing event occurred. The target area AR1 is an area where the surface shape is likely to change due to the grabbing event. In the example of Figure 4, a 7 x 7 area centered on the section of position P1 where the grabbing event occurred is set as the target area AR1. The surface shape of the waste in the target area AR1 before the grabbing event occurs is represented by 7 x 7 = 49 pile height sets. The size and position of the target area to be set can be determined appropriately depending on the event being targeted.

[0048] 402 in Figure 4 shows the state of the pit after the grabbing event, similar to 401. As shown in 402, after the grabbing event, the pile height of the waste in each section within the target area AR1 has decreased, with the degree of decrease being greatest in particular near position P1. The surface shape of the waste after the grabbing event in the target area AR1 is also represented by 7 x 7 = 49 pile height sets, just like before the grabbing event.

[0049] Here, a set of 49 pile heights representing the difference in the surface shape of the waste before and after the grabbing event is obtained by subtracting the 49 pile heights representing the surface shape of the waste after the grabbing event from the 49 pile heights representing the surface shape of the waste before the grabbing event. More specifically, the difference in the surface shape of the waste before and after the grabbing event is represented by the set of pile heights in each of the 49 sections, as shown in 403a of Figure 4.

[0050] Furthermore, the difference in the pile height of waste in each compartment included in the row indicated by the white arrow in 403a can be expressed as the height of a rectangle placed at a position corresponding to each compartment, as shown in 403. From 403, it can be seen that the pile height of waste decreases the most in the compartment near position P1, and the further the compartment is from position P1, the smaller the decrease in the pile height of waste.

[0051] Now, consider generating a shape difference model 111 for a grasping event. In this case, it is sufficient to calculate the difference between the pile heights in 49 sections included in the target area AR1 before the occurrence of a grasping event and the pile heights in the 49 sections included in the target area AR1 after the occurrence of a grasping event. Then, the calculated differences in pile heights in the 49 sections are associated with the sets of pile heights in the 49 sections before the occurrence of a grasping event, and these are used as training data. By performing machine learning using such training data, the learning unit 202 can generate a shape difference model 111 for a grasping event that infers the difference in the surface shape of the 49 sections included in the area where the grasping event will occur (or occurred) from the pile heights in the 49 sections.

[0052] When generating training data, it is preferable to represent the pile height in each partition as a relative value relative to a reference value (typically zero) set to a pile height in any one of the partitions. For example, the pile height in each partition may be represented as zero, with the pile height in the partition located at position P1, i.e., the center of the target area AR1. In this case, for example, the pile height of a partition that is 1 meter higher than the partition at position P1 is represented as "+1," and the pile height of a partition that is 1 meter lower than the partition at position P1 is represented as "-1."

[0053] By expressing the pile height in each section as a relative value in this way, the amount of data required for training can be reduced. For example, if a gripping operation is performed on waste piled 3 m high and a gripping operation is performed on waste piled 1 m high, if the surface shape before the operation is the same, the difference in the surface shape before and after the operation will be the same. Therefore, it is sufficient to use data observed in only one of these cases as training data. In contrast, if the pile height in each section is the actual height, it is necessary to use training data generated from the results obtained from gripping operations performed at different pile heights.

[0054] In addition, when the difference estimation unit 103 performs inference using a shape difference model 111 generated by machine learning using training data in which the pile height in each section is expressed as a relative value, the shape information acquisition unit 102 acquires shape information in which the pile height in each section is expressed as a relative value.

[0055] The same applies when generating a shape difference model 111 for a drop event. In this case, a target area is set with the location where the drop event occurred as its center, and the difference between the pile height of waste in each section of the target area before the drop event occurs and the pile height of waste in each section of the target area after the drop event occurs is calculated. The calculated difference is then associated with the pile height of waste in each section of the target area before the drop event occurs (preferably expressed as a relative value as described above) and used as training data. The learning unit 202 performs machine learning using such training data, thereby generating a shape difference model 111 that outputs the difference in surface shape before and after the drop event occurs.

[0056] Furthermore, when agitating waste using a crane bucket, the waste can be picked up by the bucket and then scattered along the bucket's path by moving the bucket horizontally and adjusting the bucket opening. This allows for effective agitation that spreads the waste over a wide area. A shape difference model 111 can also be generated for such an event (i.e., an event in which waste picked up by the bucket is scattered along the bucket's path). In this case, a target area is set with a line segment connecting the start and end positions of the scattering operation as its center, and the difference between the pile height of waste in each section of the target area before the event occurs and the pile height of waste in each section of the target area after the event occurs is calculated. The calculated difference is then associated with the pile height of waste in each section of the target area before the event occurs (preferably expressed as a relative value as described above) and used as training data.

[0057] In addition, a shape difference model 111 that outputs the difference in surface shape before and after the occurrence of other events, such as dumping waste from a compactor truck into a pit, can also be generated in the same manner as the above examples.

[0058] As described above, the shape difference model 111 may be a model generated by learning the relationship between the surface shape before a predetermined event occurs and the difference between the surface shapes before and after the occurrence of the predetermined event in a target area of a predetermined size set at the position where the predetermined event occurred in the pit. In this case, the shape information acquisition unit 102 acquires shape information indicating the surface shape of the waste in the above-mentioned area of a predetermined size at the position where the predetermined event occurred. In this way, by using the shape difference model 111 generated by learning targeting a target area of a predetermined size rather than the entire area in the pit, the amount of calculation required for estimation can be further reduced.

[0059] [About the estimated target] The estimation target of the shape difference model 111, in other words, the response variable, may be the pile height in all sections included in the target area, or may be the pile height in some sections included in the target area.

[0060] For example, the shape difference model 111 may be a trained model generated by learning the relationship between the pile height of waste before the occurrence of a predetermined event in each of a plurality of compartments obtained by dividing the target area and the difference in the pile height of waste before and after the occurrence of the predetermined event in one of the plurality of compartments. In this case, the difference estimation unit 103 uses the shape difference model 111 to estimate the difference in pile height of the waste for each of a plurality of compartments obtained by dividing an area of a predetermined size at the location where the predetermined event occurred.

[0061] The above configuration simplifies the configuration of the shape difference model compared to estimating the surface shape of deposits over the entire area of a predetermined size, thereby reducing the amount of training data required for learning (data that associates information indicating the surface shape before an event that changes the surface shape with information indicating the difference between the surface shapes before and after the event), thereby improving data efficiency.

[0062] In this case, the explanatory variables of the shape difference model 111 include information indicating the section to be estimated (hereinafter referred to as section designation information). For example, in the example of the grabbing event in FIG. 4, there are 49 sections to be estimated. Therefore, in this example, training data corresponding to each of the 49 sections must be generated. For example, for the section at position P1, the training data can be generated by associating the difference in waste pile height before and after the grabbing event in the section at position P1 with a combination of information indicating the surface shape of the waste in the target area AR1 before the grabbing event occurred (indicating the pile height of each of the 49 sections) and the section designation information indicating the position P1. Training data can be generated in the same manner for the other 48 sections, resulting in 49 training data. Then, by inputting the information indicating the surface shape of the waste in the target area before the grabbing event occurred and the section designation information into the shape difference model 111 generated using such training data, an estimation result of the difference in waste pile height before and after the grabbing event in the section indicated by the section designation information can be obtained.

[0063] [Adding explanatory variables] The explanatory variables of the shape difference model 111 may include other information depending on the target event, as long as the other information affects the surface shape after the target event occurs.

[0064] For example, when waste material picked up by a crane bucket is dropped in a pit, the weight of the dropped waste material affects the surface shape of the waste material at the location where the waste material is dropped. In other words, generally, the heavier the weight of the dropped waste material, the greater the change in the surface shape of the waste material at the location where the waste material is dropped. For this reason, the explanatory variables of the shape difference model 111 for drop events may include information indicating the weight of the dropped waste material.

[0065] In other words, the shape difference model 111 for a drop event may be a trained model generated by learning the relationship between the surface shape of the waste at the location before the waste is dropped, the weight of the dropped waste, and the difference in the surface shape before and after the drop for a predetermined event (i.e., a drop event) in which waste grabbed by a crane bucket is dropped. In this case, the difference estimation unit 103 estimates the difference in the surface shape before and after the drop event using this shape difference model 111, shape information indicating the surface shape of the waste before the drop event, and weight information indicating the weight of the dropped pile. This makes it possible to accurately predict the difference in the surface shape of waste grabbed by a crane bucket when it is dropped, taking into account the weight of the dropped waste.

[0066] [Simulation of crane work results] By using the information processing device 1, it is also possible to perform a simulation of what the surface shape of the waste in the pit will be after the crane has been operated according to a certain operation schedule.

[0067] In this case, for each operation position in the operation schedule showing a series of operations in which the crane bucket repeatedly performs the operations of grabbing and dropping waste, the information processing device 1 repeatedly performs a process of estimating the difference in surface shape before and after the operation at that operation position using the difference estimation unit 103, and a process of updating the shape map 112 using the surface shape estimation unit 104.

[0068] According to the above configuration, the surface shape of the waste in the pit after the crane has performed the series of operations indicated in the operation schedule can be reflected in the shape map 112, that is, a simulation can be performed when the operation schedule is applied. The results of such a simulation can be used to evaluate the operation schedule and can also be used to generate an optimal operation schedule.

[0069] The results of estimating the difference in surface shape by the information processing device 1, or the results of estimating the surface shape based on the estimation results, can also be used for purposes other than simulation. For example, based on the results of estimating the difference in surface shape or the surface shape, it is possible to predict the time required to transport waste piled up in a specified area, or the amount of power consumption of a crane during such transportation.

[0070] [Regarding the generation of operation schedules] The automatic generation of an operation schedule based on the above-mentioned simulation will be described below. The automatic generation of an operation schedule is realized by the operation schedule generation unit 105, the evaluation unit 106, and the operation schedule recommendation unit 107 shown in FIG.

[0071] The operation schedule generating unit 105 generates an operation schedule indicating a series of operations in which the operations of grabbing and dropping waste are repeatedly performed using the crane bucket. The method of generating the operation schedule is not particularly limited. For example, the operation schedule generating unit 105 may receive input of constraints such as the number of times each operation of grabbing and dropping waste is repeated and the area in which each operation is performed, and generate an operation schedule that satisfies the constraints.

[0072] For example, when generating an operation schedule for optimally transferring waste from a first area set in a pit to a second area also set in the pit, each of these areas and the number of repetitions of each operation of grabbing and dropping the waste may be input as constraints. In this case, the operation schedule generation unit 105 may generate an operation schedule by, for example, randomly determining a grab position within the first area and randomly determining a drop position within the second area for the waste grabbed at the grab position, repeating this process a specified number of times.

[0073] The operation schedule generated in the above manner is unlikely to be appropriate. Therefore, the operation schedule generation unit 105 repeats the above process to generate a large number of operation schedules. There is a high possibility that an appropriate operation schedule will be included among the large number of operation schedules generated in this manner. The evaluation unit 106 and operation schedule recommendation unit 107, which will be described below, are configured to select an appropriate operation schedule from the large number of generated operation schedules.

[0074] The evaluation unit 106 evaluates the operation schedule based on a shape map 112 that shows the surface shape of the waste in the pit after the series of operations shown in the operation schedule have been performed. The shape map 112 that shows the surface shape of the waste in the pit after the series of operations shown in the operation schedule have been performed can be generated by the method described in the section entitled "Simulation of work results by crane."

[0075] There are no particular limitations on the evaluation method applied by the evaluation unit 106. For example, the evaluation unit 106 may evaluate the operation schedule based on the difference between the surface shape shown in the shape map 112 and a pre-input ideal surface shape. Specifically, the evaluation unit 106 may use, as the evaluation result, an index value (for example, the mean square error or the root mean square error) that indicates the error between the waste pile height in each section shown in the shape map 112 and the ideal waste pile height in each section.

[0076] The operation schedule recommendation unit 107 determines an operation schedule to recommend based on the evaluation results of the multiple operation schedules by the evaluation unit 106. For example, the operation schedule recommendation unit 107 may recommend an operation schedule that has the best evaluation result by the evaluation unit 106 (for example, the smallest mean square error). Furthermore, for example, the operation schedule recommendation unit 107 may recommend an operation schedule that has the evaluation result by the evaluation unit 106 satisfying a predetermined condition (for example, the mean square error is within an allowable range). In this case, multiple operation schedules may be recommended.

[0077] Fig. 5 is a diagram showing an example of automatically generated operation schedules. In Fig. 5, 501, the surface shape of the waste in the pit before the transfer work by the crane is shown by a dashed line. This surface shape is reflected in the shape map 112 based on, for example, the measurement results by a distance sensor. Also in 501, the ideal surface shape after the transfer work is completed is shown by a dashed-dotted line. The ideal surface shape can be set appropriately by the user of the information processing device 1 depending on the amount of waste already piled up, the target pile height set for each area in the pit, etc.

[0078] To recommend an optimal operation schedule, first, the operation schedule generation unit 105 generates a plurality of operation schedules. Three of the generated operation schedules, SC1 to SC3, are shown in Fig. 5. The first stage of the operation schedule shown in SC1 is to perform an operation of grabbing waste at position P2 and drop the grabbed waste at position P3.

[0079] The difference in the surface shape of the waste before and after these actions is estimated by the difference estimation unit 103, and the surface shape estimation unit 104 reflects the estimation result in the shape map 112. More specifically, the event identification unit 101 identifies that a grab event will occur at position P2 by referring to the action schedule, and the shape information acquisition unit 102 acquires shape information indicating the surface shape of the waste around position P2 (specifically, the pile height of the waste in each section included in a predetermined size range centered on position P2) from the shape map 112. The difference estimation unit 103 then inputs this shape information into the shape difference model 111 for the grab event to estimate the difference in the surface shape before and after the grab event at position P2, and the surface shape estimation unit 104 reflects the estimated difference in the shape map 112. For a drop event at position P3, the difference in surface shape is estimated and the shape map 112 is updated in a similar manner. Note that the shape difference model 111 for the drop event is used to estimate the difference at position P3.

[0080] 502a in Fig. 5 shows the surface shape of the waste after the gripping operation at position P2 and the dropping operation at position P3 are completed, using dashed lines. As can be seen from the figure, these operations have caused a depression at position P2, and the height of the waste pile at position P3 has increased.

[0081] By performing the above process for each operation shown in the operation schedule in the order shown in the operation schedule, the shape map 112 will ultimately show the surface shape of the waste in the pit after the series of operations shown in the operation schedule have been performed. The surface shape of the waste in the pit after the series of operations shown in the operation schedule of SC1 have been performed will be the surface shape shown by the dashed line in 503a of Figure 5.

[0082] Next, the evaluation unit 106 evaluates the operation schedule of SC1 based on the updated shape map 112. The surface shape indicated by the dashed line in 503a is significantly different from the ideal surface shape indicated by the dashed line, and therefore the evaluation result by the evaluation unit 106 is low.

[0083] Similarly, the shape map 112 is updated and evaluated by the evaluation unit 106 for SC2, SC3, and other operation schedules not shown. In the example of Fig. 5, the surface shape of the waste in the pit after the series of operations shown in the operation schedule of SC2 is performed is nearly identical to the ideal surface shape. Therefore, the evaluation result of the operation schedule of SC2 by the evaluation unit 106 is high, and the operation schedule recommendation unit 107 determines the operation schedule of SC2 as the recommended operation schedule.

[0084] Alternatively, one operation schedule may be generated and evaluated, and if the evaluation result satisfies a predetermined condition, the operation schedule may be recommended, and if the predetermined condition is not satisfied, the operation schedule may be updated. In this case, the operation schedule generation unit 105 may generate one operation schedule, and if the operation schedule recommendation unit 107 determines that the evaluation result of the evaluation unit 106 on the generated operation schedule does not satisfy the predetermined condition, the operation schedule may be updated. The method for updating the operation schedule may be determined in advance.

[0085] As described above, the information processing device 1 includes an evaluation unit 106 that evaluates the operation schedule based on a shape map 112 that shows the surface shape of the waste in the pit after the series of operations shown in the operation schedule have been performed, and an operation schedule recommendation unit 107 that determines a recommended operation schedule based on the evaluation results of one or more operation schedules by the evaluation unit 106.

[0086] According to the above configuration, a desirable operation schedule can be recommended. The recommended operation schedule may be presented to the user to decide whether to adopt it or not, or the recommended operation schedule may be automatically applied to automatically operate the crane.

[0087] [Flow of the method for generating a shape difference model] The process executed by the learning device 2, that is, the flow of the method for generating the shape difference model 111, will be described with reference to Fig. 6. Fig. 6 is a flowchart showing the flow of the method for generating the shape difference model 111.

[0088] In S1 (data acquisition step), the data acquisition unit 201 acquires training data for generating the shape difference model 111. As described above, this training data is data in which the difference in surface shape before and after the occurrence of a predetermined event that changes the surface shape of the waste in the pit is associated with shape information indicating the surface shape before the occurrence of the predetermined event.

[0089] In S2 (learning step), the learning unit 202 generates a shape difference model 111 that outputs the difference in surface shape before and after the occurrence of a specified event according to the input shape information through machine learning using the training data acquired in S1.

[0090] In S3, the learning unit 202 records the shape difference model 111 generated in S2 in the storage unit 21. This ends the processing in Fig. 6. Note that instead of or in addition to recording the generated shape difference model 111 in the storage unit 21, the learning unit 202 may transmit the generated shape difference model 111 to the information processing device 1.

[0091] As described above, the method for generating a shape difference model according to this embodiment includes a data acquisition step (S1) for acquiring training data that associates the difference in surface shape before and after a specific event that changes the surface shape of waste with shape information indicating the surface shape before the occurrence of the specific event, and a learning step (S2) for generating a shape difference model 111 that outputs the difference in surface shape before and after the occurrence of the specific event according to the input shape information through machine learning using the acquired training data.

[0092] As described above, it is possible to accurately estimate differences in the surface shapes of waste while reducing the amount of calculations by using the shape difference model 111. Therefore, with the above configuration, it is possible to accurately estimate differences in the surface shapes of waste while reducing the amount of calculations.

[0093] Furthermore, according to the above configuration, training data is obtained that associates the difference in surface shape before and after a predetermined event that changes the surface shape of the waste with shape information indicating the surface shape before the occurrence of the predetermined event. Such training data is relatively easy to collect, and the above generation method has the advantage of being able to reduce the amount of data required for learning.

[0094] In other words, if we were to generate an estimation model that estimates the height of waste piles at the location where a specific event occurred, we would need to prepare training data for each height. Therefore, we would need to observe the surface shapes of waste piled at various heights before and after the changes, and generate training data based on the observation results.

[0095] In contrast, when observing differences in surface shape, it is not necessary to consider the height of the waste pile. For example, if a crane bucket grabs waste piled 3 meters high and a crane bucket grabs waste piled 1 meter high, if the surface shape before the operation is the same, the difference in surface shape before and after the operation will be similar. Therefore, it is sufficient to use only one of these cases as training data.

[0096] [Flow of estimation method] The flow of processing executed by the information processing device 1 when estimating the difference in surface shape will be described with reference to Fig. 7. Fig. 7 is a flowchart showing the flow of processing executed by the information processing device 1 when estimating the difference in surface shape. This flowchart includes each step of the estimation method according to this embodiment.

[0097] In S11, the event identification unit 101 identifies, from among a plurality of predetermined events that change the surface shape of the waste, the event that caused the difference in the surface shape to be estimated or the event that caused the difference. The event identification unit 101 also identifies the location where the identified event occurred. For example, suppose that a user of the information processing device 1 wants to estimate the surface shape of waste after a crane bucket is used to lift up waste at a certain position in a pit. In this case, the user inputs to the information processing device 1 via the communication unit 12 or the input unit 13, information that a grabbing event has occurred and the location in the pit where the grabbing event occurred. This allows the event identification unit 101 to identify the fact that a grabbing event has occurred and the location in the pit where the grabbing event occurred.

[0098] In S12 (shape information acquisition step), the shape information acquisition unit 102 acquires shape information that indicates the surface shape of the waste around the generation location identified in S11. This shape information indicates the surface shape before the occurrence of a predetermined event that changes the surface shape of the waste, and can be acquired from the shape map 112.

[0099] In S13 (estimation step), the difference estimation unit 103 estimates the difference in surface shape before and after the occurrence of a predetermined event using a shape difference model 111 that models changes in the surface shape due to the predetermined event and the shape information acquired in S12. As described above, the explanatory variables of the shape difference model 111 may include variables other than the surface shape before the occurrence of the predetermined event. In that case, the difference estimation unit 103 also uses such variables to estimate the difference in surface shape before and after the occurrence of the predetermined event.

[0100] In S14, the surface shape estimation unit 104 estimates the surface shape of the waste after the occurrence of the predetermined event using the difference between the shape information acquired in S12 and the surface shape estimated in S13, and updates the shape map 112. Specifically, the surface shape estimation unit 104 identifies the pile height of waste in each section located around the position where the predetermined event occurred, as indicated in the shape information acquired in S12, from among the pile heights of waste in each section shown in the shape map 112. The surface shape estimation unit 104 then adds the difference in the surface shape of each section estimated in S13 (specifically, the difference in the pile height of waste before and after the occurrence of the predetermined event) to the pile height of waste in each identified section. This updates the pile height of waste in each section located around the position where the predetermined event occurred in the shape map 112.

[0101] In S15, the surface shape estimation unit 104 displays the shape map 112 updated in S14 on the display device 3. This completes the illustrated processing. The display format in S15 is not particularly limited. For example, the surface shape estimation unit 104 may display the pile height of waste in each section shown in the shape map 112 in a table format. Note that it is not essential to display the shape map 112. For example, the surface shape estimation unit 104 may display an image showing the pile height of waste in a portion of the sections that were the subject of estimation (for example, a row of sections including the section where the grab or drop action was performed), as in the example of FIG. 1.

[0102] As described above, the estimation method according to this embodiment includes a shape information acquisition step (S12) of acquiring shape information indicating the surface shape of the waste before the occurrence of a predetermined event that changes the surface shape of the waste, and an estimation step (S13) of estimating the difference in the surface shape before and after the occurrence of the predetermined event using the acquired shape information and a shape difference model 111 that models the change in surface shape due to the predetermined event. This makes it possible to accurately estimate the change in the surface shape of the waste while minimizing the amount of calculation.

[0103] [Flow of recommended operation schedule method] The flow of processing executed by the information processing device 1 when recommending an operation schedule will be described with reference to Fig. 8. Fig. 8 is a flowchart showing the flow of processing executed by the information processing device 1 when recommending an operation schedule. This flowchart includes each step of the method for recommending an operation schedule according to this embodiment.

[0104] In S21, the operation schedule generating unit 105 generates a plurality of operation schedules indicating a series of operations in which the operations of grabbing and dropping waste are repeatedly performed by the crane bucket. For example, as described above, the operation schedule generating unit 105 may receive input of constraints such as the number of times each operation of grabbing and dropping waste is repeated and the area in which each operation is performed, and generate an operation schedule that satisfies the constraints.

[0105] In S22, the event identifying unit 101 acquires one of the multiple operation schedules generated in S21. The operation schedule acquired here is an unevaluated operation schedule that has not been evaluated in S28, which will be described later.

[0106] In S23, the event identification unit 101 identifies, from among the multiple actions shown in the action schedule acquired in S22, an action for which a difference in surface shape is to be estimated and the action position where that action is performed. Specifically, from among the multiple actions shown in the action schedule acquired in S22, the event identification unit 101 identifies, from among the multiple actions shown in the action schedule acquired in S22, an action for which a difference in surface shape before and after that action has not been estimated and which is the earliest in the execution order, and the action position where that action is performed. For example, when performing the processing of S23 for the first time for a certain action schedule, the event identification unit 101 identifies the action that is specified to be performed first in the action schedule and the action position where that action is performed. Note that the action identified in S23 can be rephrased as a predetermined event that changes the surface shape, and the action position identified in S23 can be rephrased as the occurrence position of the event that changes the surface shape.

[0107] In S24 (shape information acquisition step), the shape information acquisition unit 102 acquires shape information indicating the surface shape of the waste around the operation position identified in S23. As described above, the shape information can be acquired from the shape map 112. Note that which section's shape information is to be acquired for the operation position may be determined in advance depending on the type of operation (which can also be rephrased as the type of predetermined event that changes the surface shape).

[0108] In S25 (estimation step), the difference estimation unit 103 estimates the difference in surface shape before and after the action identified in S23 using a shape difference model 111 that models the change in surface shape due to the action and the shape information acquired in S24. For example, if the action identified in S23 is a grasping action, the difference estimation unit 103 inputs the shape information acquired in S24 into the shape difference model 111 for the grasping event to obtain an estimation result of the difference in surface shape.

[0109] In S26, the surface shape estimation unit 104 estimates the surface shape of the waste after the operation specified in S23 using the difference between the shape information acquired in S24 and the surface shape estimated in S25, and updates the shape map 112. Note that the shape map 112 is updated individually for each operation schedule generated in S21. For this reason, the surface shape estimation unit 104 does not update the shape map 112 stored in the memory unit 11 (which shows the surface shape of the waste at various locations in the pit before the operation indicated in the operation schedule is performed). Then, the surface shape estimation unit 104 records the updated shape map 112 for each operation schedule in the memory unit 11, etc., separately from the shape map 112 that is stored.

[0110] In S27, the event identification unit 101 determines whether or not the shape map 112 has been updated for all the operations indicated in the operation schedule acquired in S22. If the determination in S27 is NO, the process returns to S23, and if the determination in S27 is YES, the process proceeds to S28.

[0111] In S28, the evaluation unit 106 evaluates the operation schedule based on the shape map 112 determined to have been updated in S27. The shape map 112 determined to have been updated in S27 indicates the surface shape of the waste in the pit after the series of operations indicated in the operation schedule acquired in S22 have been performed. The evaluation method is as described above, so a description thereof will not be repeated here.

[0112] In S29, the evaluation unit 106 determines whether evaluation of all the operation schedules generated in S21 has been completed. If the determination in S29 is NO, the process returns to S22. On the other hand, if the determination in S29 is YES, the process proceeds to S30.

[0113] In S30, the operation schedule recommendation unit 107 determines an operation schedule to recommend to the user based on the evaluation result of S28 for each operation schedule generated in S21. For example, the operation schedule recommendation unit 107 may recommend the operation schedule that has the most favorable evaluation result of S28 as the operation schedule.

[0114] In S31, the operation schedule recommendation unit 107 displays the operation schedule determined in S30 on the display device 3. This ends the processing in FIG. 8. It should be noted that displaying the recommended operation schedule is not essential. For example, instead of or in addition to displaying the recommended operation schedule, the operation schedule recommendation unit 107 may transmit the recommended operation schedule to the crane control device and start automatic control of the crane applying the operation schedule.

[0115] [Modification] The execution entity of each process described in the above-described embodiments may be any entity and is not limited to the above-described examples. That is, the functions of the information processing device 1 or the learning device 2 can be realized by a plurality of information processing devices (which may also be called processors) that can communicate with each other. For example, the execution of each process described in the flowcharts of FIG. 6, FIG. 7, or FIG. 8 can be shared among a plurality of information processing devices. That is, the execution entity of the shape difference model generation method and estimation method in the above-described embodiments may be one information processing device or a plurality of information processing devices.

[0116] [Reference example] As described above, the shape difference model 111 may be a model generated by learning the relationship between the surface shape before a predetermined event occurs and the difference in the surface shape before and after the occurrence of the predetermined event in a target area of a predetermined size set at the position where the predetermined event occurred. In this way, using a model that makes an estimation for a target area that is a part of the area where deposits are deposited, rather than the entire area, is effective not only for estimating the difference in surface shape, but also for making any estimation related to the surface shape of deposits.

[0117] The information processing device according to this reference example includes an inference unit that performs inference regarding the surface shape of a predetermined deposit that has changed due to the occurrence of a predetermined event. The content of the inference may be anything related to the surface shape of the deposit. For example, as in the above embodiment, the inference may be the difference in the surface shape before and after the occurrence of the predetermined event, or the surface shape after the occurrence of the predetermined event. Furthermore, this surface shape may be represented by the height of the deposit in each section obtained by dividing the area where the deposit has accumulated, or may be represented by other functions, etc.

[0118] The inference unit according to this embodiment performs the inference using an inference model generated by learning a target area of a predetermined size set within a range where the surface shape has changed or may change within the area where the deposits have accumulated, thereby reducing the amount of calculation required for the inference compared to when inference is performed for the entire area where the deposits have accumulated.

[0119] [Software implementation example] The functions of the information processing device 1 or learning device 2 (hereinafter referred to as "device") can be realized by a program (learning program / estimation program) that causes a computer to function as the device, and that causes a computer to function as each control block of the device (particularly each part included in the control unit 10 or 20).

[0120] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The control device and storage device execute the program, thereby realizing the functions described in each of the above embodiments.

[0121] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0122] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0123] 〔summary〕 An information processing device according to aspect 1 of the present invention includes a shape information acquisition unit that acquires shape information indicating the surface shape of a specified deposit before the occurrence of a specified event that changes the surface shape of the specified deposit, and a difference estimation unit that estimates the difference in the surface shape before and after the occurrence of the specified event using a shape difference model that models the change in surface shape due to the specified event and the shape information.

[0124] In an information processing device according to aspect 2 of the present invention, in aspect 1, the shape difference model is a model generated by learning the relationship between the surface shape before the occurrence of the specified event and the difference in the surface shape before and after the occurrence of the specified event in a target area of a specified size set at the position where the specified event occurred within the pit, and the shape information acquisition unit acquires the shape information indicating the surface shape of the deposit in the area of the specified size at the position where the specified event occurred.

[0125] An information processing device according to aspect 3 of the present invention is, in aspect 2, a trained model generated by learning the relationship between the pile height of the deposit before the occurrence of the specified event in each of a plurality of sections divided into the target area and the difference in the pile height of the deposit before and after the occurrence of the specified event in one of the plurality of sections, and the difference estimation unit uses the shape difference model to estimate the difference in the pile height of the deposit for each of a plurality of sections divided into the area of a specified size at the location where the specified event occurred.

[0126] An information processing device according to aspect 4 of the present invention is any of aspects 1 to 3, wherein the shape difference model is a trained model generated by learning the relationship between the surface shape of the pile at the position where it will be dropped before the pile is dropped, the weight of the pile to be dropped, and the difference between the surface shape before and after the pile is dropped, for the specified event of the pile being grabbed by a crane bucket being dropped, and the difference estimation unit estimates the difference in surface shape before and after the occurrence of the specified event using the shape difference model, the shape information, and weight information indicating the weight of the pile to be dropped.

[0127] An information processing device according to aspect 5 of the present invention, in any of aspects 1 to 4, includes a surface shape estimation unit that estimates the surface shape of the deposit after the occurrence of the specified event using the shape information and the difference in surface shape estimated by the difference estimation unit.

[0128] In an information processing device according to a sixth aspect of the present invention, in the fifth aspect, the deposit is waste stored in a pit, and the surface shape estimation unit updates a shape map showing the surface shape of the waste at each position within the pit based on the estimation results of the differential estimation unit, and for each operation position in an operation schedule showing a series of operations in which the bucket of a crane transporting the waste within the pit repeatedly performs the operations of grabbing and dropping the waste, the differential estimation unit estimates the difference in surface shape at that operation position before and after the operation, and the surface shape estimation unit updates the shape map.

[0129] The information processing device of aspect 7 of the present invention, in aspect 6, comprises an evaluation unit that evaluates the operation schedule based on the shape map that shows the surface shape of the waste in the pit after performing the series of operations shown in the operation schedule, and an operation schedule recommendation unit that determines a recommended operation schedule based on the evaluation results of one or more operation schedules by the evaluation unit.

[0130] A shape difference model generation method according to aspect 8 of the present invention is a shape difference model generation method executed by at least one information processing device, and includes a data acquisition step of acquiring training data that associates shape information indicating the surface shape of a predetermined event that changes the surface shape of a predetermined deposit before the occurrence of the predetermined event with the difference in the surface shape before and after the occurrence of the predetermined event, and a learning step of generating, by machine learning using the training data, a shape difference model that outputs the difference in surface shape before and after the occurrence of the predetermined event according to the input shape information.

[0131] An estimation method according to aspect 9 of the present invention is an estimation method executed by at least one information processing device, and includes a shape information acquisition step of acquiring shape information indicating the surface shape of a specified deposit before the occurrence of a specified event that changes the surface shape of the specified deposit, and an estimation step of estimating the difference in the surface shape before and after the occurrence of the specified event using a shape difference model that models the change in surface shape due to the specified event and the shape information.

[0132] An estimation program according to a tenth aspect of the present invention is an estimation program for causing a computer to function as the information processing device according to the first aspect, and causes the computer to function as the shape information acquisition unit and the difference estimation unit.

[0133] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0134] 1. Information processing equipment 102 Shape information acquisition unit 103 Difference estimation part 104 Surface shape estimation section 106 Evaluation Department 107 Operation Schedule Recommendation Section 111 Shape Difference Model 112 Shape Map

Claims

1. a shape information acquisition unit that acquires shape information indicating a surface shape of a predetermined deposit before a predetermined event that changes the surface shape of the predetermined deposit occurs; an information processing device comprising: a difference estimation unit that estimates a difference in the surface shape before and after the occurrence of the specified event using a shape difference model that models a change in the surface shape due to the specified event and the shape information.

2. the shape difference model is a model generated by learning a relationship between a surface shape before the occurrence of the predetermined event and a difference in the surface shape before and after the occurrence of the predetermined event in a target area of a predetermined size set at a position where the predetermined event occurred, The information processing apparatus according to claim 1 , wherein the shape information acquisition unit acquires the shape information indicating a surface shape of the deposit in an area of the predetermined size at a position where the predetermined event occurred.

3. the shape difference model is a trained model generated by learning the relationship between the accumulation height of the deposit before the occurrence of the predetermined event in each of a plurality of sections divided into the target area and the difference in accumulation height of the deposit before and after the occurrence of the predetermined event in one of the plurality of sections, The information processing device according to claim 2 , wherein the difference estimation unit estimates the difference in the accumulation height of the deposit using the shape difference model for each of a plurality of sections obtained by dividing the area of the predetermined size at the position where the predetermined event occurred.

4. the shape difference model is a trained model generated by learning the relationship between the surface shape of the pile at the position where the pile is to be dropped before the pile is dropped, the weight of the pile to be dropped, and the difference in the surface shape before and after the pile is dropped, for the predetermined event of the pile being dropped by the bucket of a crane; 4. The information processing device according to claim 1, wherein the difference estimation unit estimates the difference in the surface shape before and after the occurrence of the specified event using the shape difference model, the shape information, and weight information indicating the weight of the deposited material.

5. 4. The information processing device according to claim 1, further comprising a surface shape estimation unit that estimates the surface shape of the deposit after the occurrence of the predetermined event by using the shape information and the difference in the surface shape estimated by the difference estimation unit.

6. the deposit is waste contained in a pit; the surface shape estimation unit updates a shape map indicating the surface shape of the waste at each position within the pit based on the estimation result of the difference estimation unit; An information processing device as described in claim 5, which repeatedly performs a process of estimating the difference in surface shape before and after an operation at each operation position in an operation schedule showing a series of operations in which the bucket of a crane transporting the waste within the pit repeatedly grabs and drops the waste, using the difference estimation unit, and a process of updating the shape map using the surface shape estimation unit.

7. an evaluation unit that evaluates the operation schedule based on the shape map that indicates the surface shape of the waste in the pit after the series of operations indicated in the operation schedule have been performed; The information processing apparatus according to claim 6 , further comprising: an operation schedule recommendation unit that determines a recommended operation schedule based on a result of evaluation of one or more operation schedules by the evaluation unit.

8. A method for generating a shape difference model executed by at least one information processing device, comprising: a data acquisition step of acquiring training data in which a difference in the surface shape before and after the occurrence of a predetermined event that changes the surface shape of a predetermined deposit is associated with shape information indicating the surface shape before the occurrence of the predetermined event; a learning step of generating a shape difference model that outputs a difference in surface shape before and after the occurrence of the specified event according to input shape information through machine learning using the training data.

9. An estimation method executed by at least one information processing device, comprising: a shape information acquisition step of acquiring shape information indicating a surface shape of a predetermined deposit before the occurrence of a predetermined event that changes the surface shape of the predetermined deposit; an estimation step of estimating a difference in the surface shape before and after the occurrence of the specified event using a shape difference model that models a change in the surface shape due to the specified event and the shape information.

10. 2. An estimation program for causing a computer to function as the information processing device according to claim 1, the estimation program causing a computer to function as the shape information acquisition unit and the difference estimation unit.

Images