Information processing device, remote control system, and information processing method

Abstract

To provide an information processing device, a remote control system, and an information processing method that can secure the required amount of learning data and obtain appropriate learning accuracy.SOLUTION: An information processing device 40 includes a CPU 41. The CPU 41 serves as a generation unit and generates a virtual control room that is a digital reproduction of a remote control room used to remotely control work machines at construction sites. The CPU 41 serves as a learning unit and learns the details of work machine operation control in response to monitoring data from input / output data including work machine operation control data output from the virtual control room and monitoring data that is virtually obtained by a work machine monitoring device on the basis of virtual operation of the work machine in accordance with the operation control data and input to the virtual control room.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to an information processing device, a remote control system, and an information processing method. [Background technology]

[0002] Because construction machinery at construction sites creates a dangerous and difficult environment for the operator, progress is being made in the development of technologies for remotely operating the machinery based on images captured by monitoring devices and measurement results. Furthermore, for the purpose of simulating work procedures at construction sites, simulated spaces are being created that digitally reproduce construction sites based on CAD data and captured images. Patent Document 1 invents a technology that determines work efficiency in a simulated space according to the operation pattern of construction machinery such as a dozer, and applies this to optimize work efficiency at the actual site.

[0003] Meanwhile, in light of the aging and declining workforce, attention is also being paid to the autonomous driving of construction machinery. After limited machine control has been put into practical use as autonomous driving technology, attempts are being made to utilize machine learning based on photographic images and sensor measurement results, with the aim of further expanding the conditions of application. While construction sites are diverse environments, the situation is often uncontrollable. For this reason, as an alternative to cases where machine learning cannot be completed due to a lack of learning data from on-site experience alone, it has been proposed to efficiently perform machine learning in a simulated space, such as that used in the simulation mentioned above. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2023-131859 Summary of the Invention [Problem to be solved by the invention]

[0005] However, to perform machine learning in a simulated space with enough accuracy to result in autonomous driving, the reality of the simulated space, that is, the reproducibility of the work environment in which autonomous driving is desired, is important. The more reproducibility is pursued, the higher the cost, but if reproducibility is compromised due to real-world cost constraints, the problem of learning accuracy not improving is faced. Therefore, the challenge was to create a simulated space that would improve learning accuracy at a limited cost.

[0006] In order to solve the above problems, the present invention provides an information processing device, a remote operation system, and an information processing method that perform machine learning on the remote operation of construction machinery in a simulated space, secure the necessary amount of learning data, and achieve appropriate learning accuracy. [Means for solving the problem]

[0007] One aspect of the present invention is a generation unit that generates a virtual control room that reproduces in a digital space a remote control room used for remotely controlling a work machine at a construction site; a learning unit that learns the details of operation control of the work machine in response to the monitoring data using input / output data including operation control data of the work machine output from the virtual operation room and monitoring data that is virtually obtained by a monitoring device of the work machine based on virtual operation of the work machine in accordance with the operation control data and input to the virtual operation room; The information processing device includes: [Effects of the Invention]

[0008] According to the present invention, it is possible to secure a necessary amount of learning data for remote control and obtain appropriate learning accuracy. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram showing the overall configuration of a remote control system. [Figure 2] FIG. 2 is a diagram illustrating a display panel. [Figure 3] FIG. 2 is a block diagram illustrating the functional configuration of a control device and an information processing device. [Figure 4] FIG. 10 is a diagram illustrating the positioning of a virtual model. [Figure 5] 10 is a flowchart showing the procedure of a learning control process for autonomous driving. [Figure 6] 10 is a flowchart showing a control procedure for automatic driving. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an embodiment will be described with reference to the drawings. FIG. 1 is a diagram showing the overall configuration of a remote control system 100 according to this embodiment. The remote operation system 100 performs operations related to, for example, tunnel construction.

[0011] The remote control system 100 includes a work machine 10 at a construction site, a monitoring device 20 that takes photographs and measures at the construction site, a remote control room 30 for remote control at a location away from the construction site, and an information processing device 40 that is connected to the remote control room 30 via a communication line.

[0012] The work machines 10 include construction machines that drill or cut holes in the rock at the tunnel face, transport machines that transport (transport) excavated materials such as excavated rock and earth outside the tunnel, and environmental machines that illuminate the inside of the tunnel under construction, supply air, and suck up (collect) dust. The transport machines may include transport vehicles and belt conveyors, as well as power shovels and dozers that load materials onto these, or shovel loaders and wheel loaders that combine transport and loading functions. The environmental machines may include fixed machines and pipes connecting the inside and outside of the tunnel. At least a portion of the work machines 10 is remotely controlled, and can switch between operation in response to direct operation by the driver in the cab and operation in response to remote control signals received from outside.

[0013] The monitoring device 20 includes at least a camera 21 that captures images of the construction site, and may also include a measurement sensor that analyzes the air quality at the construction site, a vibration sensor that measures vibrations and sounds at the construction site, and a sound collection device (microphone). The camera 21 also includes a device that captures images of the surroundings visible to a driver sitting in the driver's seat of a work machine 10 that may be the target of remote operation, to obtain captured image data. The surroundings may include the parts of the work machine 10 that actually operate (mainly the front), both sides, and the rear (which may be equivalent to an image seen through a rearview mirror). The image capture resolution may be, for example, approximately full HD. The monitoring device 20 transmits the captured image data and measurement data to the remote control room 30.

[0014] The remote control room 30 is located away from the work area of the construction machinery, at least away from the tunnel face. The remote control room 30 may be located outside the work area within the tunnel under construction, or outside the tunnel (it may be in a work shaft). The remote control room 30 may be a dedicated facility or a temporary enclosure, or it may be part of a building that also serves as an office or a rest area for workers.

[0015] Inside the remote control room 30, there are located a display panel 31 that displays images captured by the imaging device 21 in approximately real time, a cab 32 that has a handle, levers, stick, pedals, and switches for operating the work machine 10 that is the target of remote operation, and a control device 33, which is the remote operation device of this embodiment. A monitor or display screen that displays measurement results from sensors other than captured images may be located next to the display panel 31 or above the cab 32. The control device 33 acquires captured data and measurement data from the monitoring device 20 and displays them on the display panel 31 or the like, and generates and outputs operation control signals in response to operations accepted by the cab 32. The control device 33 may be a dedicated device integrated with the cab 32, or a regular electronic computer (PC) may be connected to the cab 32 or the like. The remote control room 30 may have a seat in front of the cab 32 for the driver who performs remote operation. The interior of the remote control room 30 may be air-conditioned.

[0016] The information processing device 40 may be located in the remote control room 30, or may be located in a data center, office building, or the like that is completely separate from the construction site and the remote control room 30. The information processing device 40 is an electronic computer. The information processing device 40 can generate a virtual control room that models the remote control room 30 in a digital space.

[0017] FIG. 2 is a schematic diagram showing the display panel 31 of the remote control room 30. As shown in FIG. The display panels 31 may, for example, consist of eight or nine panels, and the sizes of the panels may differ from one another. For example, the largest panel showing the front of the work machine 10 may be located in the center, with panels corresponding to rearview mirrors or panels showing left and right being located on either side of it. The other panels may include, for example, a panel showing an image of an area above the work machine 10, a screen showing an enlarged view of the area to be operated, or a panel showing a view from an angle. These other panels may also include a captured image of the entire work machine 10 from above. Some of the panels may be tilted at an appropriate angle so as to face the driver directly. Panels that do not need to be displayed may be turned off depending on the type of work machine 10. The display panels 31 may not be multiple independent panels, but may instead display multiple images divided into sections on a common panel.

[0018] FIG. 3 is a block diagram illustrating the functional configuration of the control device 33 and the information processing device 40. As shown in FIG. The control device 33 shown in FIG. 3(a) includes a CPU (Central Processing Unit) 331 (control unit), a storage unit 332, a communication unit 333, an output unit 334, an operation reception unit 335, and the like.

[0019] The CPU 331 is a processor that performs arithmetic processing and provides overall control over the operation of the control device 33. The processor may be a general-purpose processor, or a dedicated processor specialized for display on the display panel 31 and operation control of the work machine 10 in response to input operations to the cab 32. There may be multiple processors. The multiple processors may execute the same process in parallel, or may each operate independently for each application. The CPU 331 outputs an operation control signal to the work machine 10 in response to an input operation to the operation reception unit 335.

[0020] The storage unit 332 has a RAM 3321 (Random Access Memory) and a non-volatile memory 3322. The RAM 3321 provides a working memory space for the CPU 331 and stores temporary data. The RAM 3321 may be, for example, a DRAM. The non-volatile memory 3322 stores various programs, setting data, and the like. The non-volatile memory 3322 may be a HDD (Hard Disk Drive), a flash memory, or the like.

[0021] The programs stored in the non-volatile memory 3322 include a remote operation program P1 and an automatic driving program P2. The remote operation program P1 generates a control signal according to the content of an input operation to the operation reception unit 335 and transmits the control signal to the target work machine 10 via the communication unit 333. The automatic driving program P2 is a program that determines appropriate operation content for the work machine 10 to be remotely operated based on image data that is the display content on the display panel 31, and outputs the operation content to the remote operation program P1. The CPU 331 that executes the remote operation program P1 and the automatic driving program P2 can receive this operation content information and automatically output an operation control signal to the work machine 10. That is, in this case, the CPU 331 does not need to accept input operations to the operation reception unit 335. The automatic driving program P2 may include a trained model P21 that outputs operation content in response to input of image data. The trained model P21 is obtained as a learning result of the machine learning model P41, which will be described later. The trained model P21 will be described later.

[0022] The communication unit 333 controls communication with external devices. The communication unit 333 has a network card or the like, and is capable of sending and receiving communication data via, for example, a LAN (Local Area Network) and / or a wireless LAN. Communication connection destinations include the above-mentioned work machine 10. Furthermore, communication with the information processing device 40 is possible in real time or for a specific period when communication connection is established.

[0023] The output unit 334 includes a display unit that outputs display data to the display panel 31 under the control of the CPU 331. The output unit 334 may also have a speaker and output sound measured by a sound collection device of the monitoring device 20. The output unit 334 may also have a mechanism that vibrates the driver's seat. The CPU 331 may cause the output unit 334 to output vibrations corresponding to vibrations detected by a vibration sensor of the monitoring device 20. The output volume and vibration amount of the speaker may not be equal to the actual measured values. The output volume and vibration amount may be a magnitude obtained by reducing the actual magnitude by an appropriate ratio.

[0024] The operation reception unit 335 receives input operations related to the operation of the work machine 10 to be remotely controlled from the levers and the like on the cab 32, and outputs the input operations as electrical signals to the CPU 331. The cab 32 may replicate the operation mechanism of the work machine 10 to be remotely controlled, and may also have an additional operation mechanism for remote control in addition to this.

[0025] In the remote control room 30, it may be possible to switch between the work machines 10 to be remotely operated. For example, the operation reception unit 335 may have a selector switch or the like for switching between the work machines 10 to be remotely operated. Switching may be performed by operating a touch panel at hand or the like. The cab 32 may have operation reception mechanisms corresponding to all of the work machines 10 to be remotely operated, and may not accept input operations from unnecessary operation reception mechanisms. Alternatively, the cab 32 may have a general-purpose operation mechanism common to multiple work machines 10, and may accept operations that are different from direct operation of each work machine 10.

[0026] The output unit 334, the operation reception unit 335, etc. may be separate from the main body of the control device 33 and may be connected to the control device 33 as peripheral devices. The non-volatile memory 3322 is also an auxiliary storage device and may be externally attached to the main body of the control device 33.

[0027] Additionally, the CPU 331 of the control device 33 may be capable of detecting an abnormality based on data acquired from the monitoring device 20 and performing an alarm operation according to the nature of the abnormality. The alarm operation may be performed by displaying a warning on the display panel 31, outputting a sound, turning on an LED lamp, or the like. For this purpose, the remote control room 30 may have a speaker, an LED lamp, or the like that performs the alarm operation.

[0028] The information processing device 40 shown in FIG. 3(b) includes a CPU 41, a storage unit 42, a communication unit 43, an output unit 44, an operation reception unit 45, and the like.

[0029] The CPU 41 is a processor that performs arithmetic processing and controls the overall operation of the information processing device 40. The processor may be a general-purpose product or a dedicated product specialized for the information control of the present invention. There may be multiple processors. The multiple processors may execute the same process in parallel, or may operate independently for each purpose.

[0030] The storage unit 42 has a RAM 421 and a nonvolatile memory 422. The RAM 421 provides a working memory space for the CPU 41 and stores temporary data. The RAM 421 may be, for example, a DRAM. The nonvolatile memory 422 stores various programs, setting data, and the like. The nonvolatile memory 422 may be, for example, a hard disk drive (HDD) or a flash memory.

[0031] The programs stored in the non-volatile memory 422 include a virtual model P4. The virtual model P4 generates a virtual control room that recreates the remote control room 30 in a digital space (virtual space). The virtual model P4 that recreates a real space in this way is also called a digital twin. The virtual model P4 has a machine learning model P41. The machine learning model P41 is a model that performs learning related to autonomous driving. The trained machine learning model P41 is copied to the control device 33 and becomes the trained model P21.

[0032] The communication unit 43 controls communication with external devices. The communication unit 43 has a network card or the like, and is capable of transmitting and receiving communication data via, for example, a LAN (Local Area Network) and / or a wireless LAN. The communication connection destination includes the control device 33. It is not necessary for the communication unit 43 to be constantly connected to the control device 33.

[0033] The output unit 44 performs monitor display from at least the monitoring device reproduced in the virtual space under the control of the CPU 41. The output unit 44 may also have a speaker and be capable of outputting a reproduced sound of the sound measured by the sound collection device of the monitoring device 20.

[0034] The operation reception unit 45 receives an input operation and outputs a signal according to the content of the received operation to the CPU 41. The operation reception unit 45 may include, for example, a keyboard and a pointing device.

[0035] The output unit 44, the operation reception unit 45, etc. may be separate from the main body of the information processing device 40 and may be connected to the information processing device 40 as peripheral devices. The nonvolatile memory 422 is also an auxiliary storage device and may be externally attached to the main body of the information processing device 40.

[0036] Next, the virtual model P4 of the information processing device 40 will be described. FIG. 4 is a diagram for explaining the positioning of the virtual model P4.

[0037] As described above, the virtual model P4 is a model that reproduces the remote control room 30 in a virtual space. In reality, at a construction site, the site conditions are acquired by the imaging device of the monitoring device 20 and sent to the control device 33. The control device 33 displays the captured images in real time on the display panel 31. In addition, operation control signals according to operations by the driver in the remote control room 30 are sent to the work machine 10, causing the work machine 10 to operate and changes to the construction site. These operations of the work machine 10 and changes in the construction site are detected by the monitoring device 20 and sent again to the control device 33.

[0038] The information processing device 40 may receive photographed data and measurement data from such monitoring device 20 and change the display on a display panel in the virtual control room. The information processing device 40 also operates as a simulator that virtually generates changes in the remote control room 30 in accordance with construction work. Specifically, the information processing device 40 virtually operates the work machine 10 and the like in accordance with an artificially generated operation control signal. The information processing device 40 generates photographed image data, measurement data, and the like in accordance with the virtual operation of the work machine 10 and inputs them into the virtual control room. As a result, the monitoring data is displayed on a display panel or the like in the virtual control room.

[0039] In this way, the display image reproduced and displayed by the virtual model P4 in the information processing device 40 conforms to the imaging range and imaging resolution of the imaging device of the monitoring device 20 and the display resolution of the display panel 31. As a result, compared to a reproduction of an actual construction site, information outside the field of view, resolution, and sensitivity of the imaging device, and the resolution and color reproducibility of the display panel 31 is omitted or simplified. In other words, parts that do not appear in the monitoring results of the monitoring device 20 or parts whose impact is considered to be negligible do not need to be reproduced in the virtual model P4. Parts whose impact is considered to be negligible include, for example, information that an operator actually riding on the construction machine 10 could not recognize without using senses other than sight. In light of the current situation in which remote operation has been realized, the content of this omission and the degree of simplification are considered to cover the range of information necessary for remote operation. In this embodiment, learning related to the autonomous operation of the construction machine 10 and the like is performed using this virtual control room simulation.

[0040] When displayed on the display panel of the virtual control room, the tunnel shape and the like can be acquired in advance as separate design information. Note that physical property values corresponding to the constituent materials of the tunnel face, road surface, etc., which are difficult to measure before the actual work, i.e., hardness distribution, can be simply represented based on pre-measurement data, etc. When operating as a simulator, a virtual image that is the result of the simulation can be displayed on the display panel of the virtual control room.

[0041] The information processing device 40 trains a machine learning model P41 on the operation control of the work machine 10 using input / output data of a virtual model P4, which includes operation control data for virtually remotely operating the work machine 10 that is the target of automatic driving, and monitoring data, which includes simulated results of the operation of the work machine 10 based on the operation control data, received by the monitoring device 20 in the form of virtual image data and measurement data. For example, reinforcement learning using deep learning may be used for the learning. The specific reinforcement learning algorithm is not particularly limited and may be selected arbitrarily. Alternatively, multiple algorithms may be switched during learning, each used for the simulation, and the most appropriate one may be selected. For example, the machine learning model P41 applies an appropriate image recognition algorithm to images displayed on a display panel to determine the positional relationship between the work machine 10 that is the target of automatic driving and its surrounding objects. The surrounding objects include other work machines 10, the monitoring device 20, and the tunnel walls and face. If other workers are present, the other workers are also included in the surrounding objects.

[0042] During learning, an operation control signal is output according to acceptable operations so that the work machine 10 executes an operation toward the goal while appropriately avoiding these surrounding objects. A reward is then determined according to the operation of the work machine 10 according to the operation control signal. The reward can be determined based on the time required for execution, the amount of operation, the number and type of troubles that occurred during the operation, etc. A large reward is given if the operation is performed efficiently in a short time without any trouble, and the reward is reduced if a trouble occurs. Rewards are determined according to the results of the operations executed while changing environmental conditions, and the operation that obtains the maximum reward is selected as the optimal operation (champion). As described above, the object being simulated is the remote control room 30, not an actual construction site. Therefore, the amount of calculations required for simulation is significantly less than that required for simulating a construction site.

[0043] The machine learning model P41 trained in this way is copied to the trained model P21 of the control device 33 in the remote control room 30 and is used by the control device 33. In other words, the trained model P21 outputs an operation control signal corresponding to optimal operation based on the learning results in response to the input of display content to the actual display panel 31. When the output operation control signal is received by the work machine 10, the work machine 10 operates in accordance with the operation control signal.

[0044] Note that such learning depends on the work content and the specific type of work machine 10. Therefore, learning of the machine learning model P41 may be performed individually for each work unit. Furthermore, the machine learning model P41 may also be changed if the work machine 10 is changed, for example.

[0045] FIG. 5 is a flowchart showing the procedure of the learning control process for autonomous driving in the information processing device 40. The CPU 41 of the information processing device 40 generates a virtual remote control room in a virtual space (memory space of the information processing device 40) based on the virtual model P4 (S1; generation unit, generation step). The CPU 41 sets an operation target for the work machine to be automatically driven in the virtual space (S2).

[0046] The CPU 41 sets reward conditions for actions according to the goals (S3). The CPU 41 performs reinforcement learning by operating the target work machine in a virtual space while determining environmental conditions and determining rewards according to the actions (S4; learning unit, learning step).

[0047] The CPU 41 establishes a communication connection with the control device 33 in the remote control room 30, transmits the machine learning model for which the optimal solution has been determined by reinforcement learning to the control device 33 as a trained model P21, and stores it in the non-volatile memory 3322 (S5). Then, the CPU 41 ends the autonomous driving learning control process.

[0048] FIG. 6 is a flowchart showing the control procedure for automatic driving in the control device 33. This automatic driving control is executed, for example, by calling the automatic driving program P2 from the remote control program P1 during automatic driving.

[0049] The CPU 331 executes the automatic driving program P2 (S11). The CPU 331 acquires monitoring data from the monitoring device 20, inputs the data into the automatic driving program P2, and applies the learned model P21 (S12).

[0050] The CPU 331 acquires the output of the trained model P21 (S13). The CPU 331 causes the communication unit 333 to output the acquired content to the work machine 10 as an operation control signal for automatic remote control (automatic driving) (S14).

[0051] The CPU 331 determines whether an instruction to end autonomous driving has been received (S15). The reception of the instruction to end autonomous driving may be received by the operation reception unit 335, as described above. If it is determined that an instruction to end autonomous driving has not been received (S15; NO), the processing of the CPU 331 returns to S12. If it is determined that an instruction to end autonomous driving has been received (S15; YES), the CPU 331 ends the autonomous driving control processing.

[0052] As described above, the information processing device 40 of this embodiment includes a CPU 41. The CPU 41 serves as a generation unit and generates a virtual control room that is a digital model of the remote control room 30 used to remotely operate the work machine 10 at a construction site. The CPU 41 serves as a learning unit and learns the details of the operation control of the work machine 10 in response to the monitoring data from input / output data including operation control data of the work machine 10 output from the virtual control room and monitoring data that is virtually obtained by the monitoring device 20 of the work machine 10 based on virtual operation of the work machine 10 in accordance with the operation control data and input to the virtual control room. In this way, by virtualizing the remote control room 30 and learning the operational control of the work machine 10 using input / output data from the remote control room 30, the learning process is confined to a virtual space. Therefore, the monitoring data only needs to be reproduced with the required accuracy within the range of the resolution and time resolution of the monitoring data (particularly the photographed data). In other words, the information processing device 40 does not need to reproduce the actual construction site with high accuracy, nor does it need to perform trial and error or optimization to determine the criteria for simplifying the construction site. Therefore, the information processing device 40 of this embodiment does not need to perform work to actually generate learning data, and can easily obtain the required amount of input / output data with appropriate accuracy as learning data in the digital space. This allows the information processing device 40 to obtain accurate learning data related to autonomous driving.

[0053] The input / output data may also include photography data relating to the work machine 10 at the construction site. While photography information includes a lot of visual information, the data is compressed to a range corresponding to the number of pixels in real space, so the amount of data is compressed efficiently, making it easy to achieve both high learning accuracy and reduced load.

[0054] The photographed data may also include images of the surroundings of the work machine 10 that are visible from the driver's seat of the work machine 10. By obtaining photographed images that correspond to the visible range from the driver's seat, it is possible to carry out learning using the images that the driver uses for remote operation as they are. Therefore, the information processing device 40 can carry out more stable learning related to autonomous driving based on images that contain the amount of information the driver needs for remote operation.

[0055] The construction site may also be a tunnel. Tunnel construction sites are prone to high disaster risk, and automation of work is highly desirable, but it is difficult to obtain data for learning related to autonomous driving with sufficient accuracy in advance. By learning according to this embodiment, control information for autonomous driving can be obtained more efficiently and accurately than before.

[0056] Furthermore, reinforcement learning may be used for learning. In the operation of the work machine 10, there are actions that should be avoided and targets for desirable actions, but an optimal solution is not always easily determined due to the many environmental conditions, etc. In such cases, the information processing device 40 can use reinforcement learning to efficiently obtain a more desirable solution.

[0057] The remote operation system 100 of this embodiment also includes a control device 33 having a memory unit 332 that stores a trained model P21, which is the learning result obtained by the information processing device 40, a CPU 331, an output unit 334, and an operation reception unit 335, and an image capture device 21 that captures images of a construction site to obtain image capture data related to monitoring the work machine 10. The CPU 331 outputs an operation control signal to the work machine 10 in response to an input operation received by the operation reception unit 335. The CPU 331, as an output unit, can display the image capture data on the display panel 31 or the like and automatically output an operation control signal to the work machine 10 based on the image capture data and the trained model P21 stored in the memory unit 332. In other words, the remote operation system 100 can automatically operate the work machine 10 based on the monitoring data using the trained model P21. Therefore, the remote operation system 100 can easily reduce the labor required at a construction site without significantly changing the system configuration related to remote operation.

[0058] The information processing method of this embodiment also includes a generation step of generating a virtual control room that is a digital model of a remote control room 30 used for remotely controlling a work machine 10 at a construction site, and a learning step of learning the details of the operation control of the work machine 10 in response to the monitoring data from input / output data that includes operation control data for the work machine 10 output from the virtual control room and monitoring data that is virtually obtained by the monitoring device 20 of the work machine 10 based on the virtual operation of the work machine 10 in accordance with the operation control data and input to the virtual control room. According to this information processing method, the remote control room 30 is virtualized and the operational control of the work machine 10 is learned using input / output data from the remote control room 30, thereby confining the learning process to a virtual space. Therefore, there is no need to reproduce the actual construction site with high accuracy, and there is also no need to perform trial and error or optimization to determine what criteria should be used to simplify the construction site. Therefore, the information processing method of this embodiment can easily obtain the required amount of input / output data with appropriate accuracy as learning data. This allows for accurate acquisition of learning data related to autonomous driving.

[0059] The above embodiment is merely an example, and various modifications are possible. For example, in the above description, the learning data, especially the monitoring data, are all obtained as a result of simulation, but this is not limiting. The learning data may also include the display state in the remote control room 30 of the monitoring data actually acquired by the monitoring device 20 at the construction site.

[0060] Furthermore, in the above, the trained model P21 does not have to be completely different for each work site or each work machine 10. For example, the trained models P21 of similar work machines 10 may be trained in common partway using a machine learning model P41 with a common algorithm, and then each may be fine-tuned (trained) in the end.

[0061] Although the above description has been given using a tunnel construction site as an example, the present invention is not limited to this and may be applied to construction sites of other structures or buildings.

[0062] In the above description, the machine learning model P41 for autonomous driving is trained by reinforcement learning, but this is not limited to this. It may be obtained by a machine learning model with an appropriate algorithm. Furthermore, the monitoring data that is input data to the machine learning model P41 does not have to be limited to captured images. Other sensor measurement data may also be included in the monitoring data.

[0063] Furthermore, parts of the remote control room 30 that are not related to monitoring and remote control do not need to be reproduced in the virtual space.

[0064] Furthermore, automatic driving control may be performed simultaneously on a plurality of construction machines 10. In this case, the types of construction machines 10 that are the targets of automatic driving may be different from one another. In addition, the specific details of the structure, configuration, control content, and procedures shown in the above embodiments can be modified as appropriate without departing from the spirit of the present invention. The scope of the present invention includes the scope of the invention described in the claims and its equivalents. [Explanation of symbols]

[0065] 10. Work Machinery 20 Monitoring equipment 21 Imaging equipment 30 Remote Control Room 31 Display panel 32 Driver's cab 33 Control device 331 CPU 332 Storage section 3321 RAM 3322 Non-volatile Memory 333 Communications Department 334 Output Section 335 Operation reception section 40 Information processing equipment 41 CPU 42 Storage section 421 RAM 422 Non-volatile memory 43 Communications Department 44 Output section 45 Operation reception section 100 Remote Control System P1 Remote Control Program P2 Autonomous Driving Program P21 Trained model P4 Virtual Model P41 Machine Learning Model

Claims

1. a generation unit that generates a virtual control room that reproduces in a digital space a remote control room used for remotely controlling a work machine at a construction site; a learning unit that learns the details of operation control of the work machine in response to the monitoring data using input / output data including operation control data of the work machine output from the virtual operation room and monitoring data that is virtually obtained by a monitoring device of the work machine based on virtual operation of the work machine in accordance with the operation control data and input to the virtual operation room; An information processing device comprising:

2. The information processing device according to claim 1 , wherein the input / output data includes photographic data relating to the work machine at the construction site.

3. The information processing device according to claim 2 , wherein the photographed data includes images of the surroundings of the work machine that are visible from the driver's seat of the work machine.

4. The information processing apparatus according to claim 1 , wherein the monitoring data includes information corresponding to a result of a virtual execution in the digital space based on the operation control data.

5. The information processing device according to claim 1 , wherein the construction site is a tunnel.

6. The information processing device according to claim 1 , wherein reinforcement learning is used for the learning.

7. a remote control device including a storage unit that stores a learning result obtained by the information processing device according to claim 1, a control unit, an output unit, and an operation reception unit; an imaging device that photographs the construction site and obtains imaging data related to monitoring the work machine; Including, The control unit outputting an operation control signal to the work machine in response to the input operation received by the operation receiving unit; the output unit displays the photographing data; The operation control signal can be automatically output to the work machine based on the photographed data and the learning results stored in the storage unit. Remote control system.

8. a generating step of generating a virtual control room in which a remote control room used for remotely controlling a work machine at a construction site is reproduced in a digital space; a learning step of learning the details of the operation control of the work machine in response to the monitoring data using input / output data including operation control data of the work machine output from the virtual operation room and monitoring data that is virtually obtained by a monitoring device of the work machine based on virtual operations of the work machine in accordance with the operation control data and input to the virtual operation room; An information processing method including:

Images