JP7572663B2 - Operation plan generation device, learning device, operation plan generation method, learning method, and program - Google Patents

Description

This disclosure relates to an operation plan generation device, a learning device, an operation plan generation method, a learning method, and a program.

In recent years, autonomously moving devices such as drones, self-driving cars, and self-driving ships (hereafter referred to as autonomous moving devices) have been actively researched and some of them are actually being used. An autonomous moving device is a device that autonomously controls its path to move along a given route, and is generally configured to autonomously correct its trajectory to move along the route even if there is a disturbance during movement.

A typical example of an autonomous mobile device is a drone that flies autonomously outside of visual line of sight. An autonomously flying drone does not receive sequential instructions from an operator, but moves while making its own decisions to follow a given route. If the autonomous mobile device is a device that operates in the air, such as a drone, or a ship that moves on the ocean, a typical example of a disturbance that may cause it to deviate from its route is wind. Even in the event of such a disturbance, an autonomous mobile device can autonomously control the aircraft to move as far as possible along its route.

Although autonomous mobile devices have a certain degree of autonomous navigation function, they are usually operated while being remotely monitored by a traffic management system in order to respond to unforeseen circumstances and ensure safe operation. Such monitoring of the operation route can also be said to be one of the functions of the traffic management system. Some autonomous mobile devices are equipped with a function to avoid collisions with each other using sensors, etc., but since they may be used in urban areas, it is necessary to avoid collisions between autonomous mobile devices in the unlikely event that they occur. Therefore, even if multiple autonomous mobile devices are operated at the same time, it is necessary to create an operation plan by adjusting the route and time in advance so that the autonomous mobile devices do not collide and come into contact with each other. Here, the operation plan shows the operation route and passing time for which an adjustment plan has been made.

The function of formulating (creating) an operation plan is one of the functions of an operation management system. As an example of using the operation plan creation function, Non-Patent Document 1 describes a successful interconnection test of a drone traffic management system in which general drone operators also participated.

As an example of an operation planning system (in the case of drones), Non-Patent Document 2 describes the current process from application to operation in a drone information infrastructure system. In the process described in Non-Patent Document 2, procedures are carried out manually via email.

Non-Patent Document 3 describes how, in a study on flight management for integrated traffic management functions, route design and collision detection are handled using a route with a uniform, fixed width. In Non-Patent Document 3, this route with width is called a protected airspace.

Patent Document 1 also describes a traffic control support system that includes a memory unit, an instruction responsiveness estimation unit, a geographic E-map generation unit, and a movement prediction unit, with the aim of improving the accuracy of ship route control while suppressing the amount of calculation. The memory unit has behavior data related to the behavior of a moving object, geographic data in which each section of a map divided into meshes is assigned with geographic attribute information, which is information related to the movement standard of the moving object, and responsiveness data related to the responsiveness of the moving object to an instruction. The instruction responsiveness estimation unit estimates the ideal behavior of the moving object based on the responsiveness data, calculates the difference between the estimated behavior and the behavior of the behavior data, and updates the responsiveness data based on the calculation result. The geographic E-map generation unit estimates the likelihood that the moving object exists at a certain coordinate at each time based on the behavior data, geographic data, and responsiveness data, and generates a geographic E-map. The movement prediction unit predicts the future coordinates of the moving object based on the geographic E-map.

JP 2018-036958 A

New Energy and Industrial Technology Development Organization, NEC Corporation, and eight other organizations, "Successful interconnection test of drone traffic management system with participation from general drone operators - 29 operators conduct flight tests -", [online], October 30, 2019, New Energy and Industrial Technology Development Organization, [June 1, 2020], Internet <URL: https://www.nedo.go.jp/news/press/AA5_101228.html> Ministry of Land, Infrastructure, Transport and Tourism, "Drone/UAS Information Platform System "Drone Information Platform System Operation Manual (Operation Manual Applicant Edition 3.0)" [online], March 26, 2020, Ministry of Land, Infrastructure, Transport and Tourism, [June 1, 2020], Internet <URL: https://www.dips.mlit.go.jp/portal/file_download> NEC Corporation and four other corporations, "Japan Drone 2018 DRESS Project Forum Project name: "Research and development of a traffic control system to realize safe, secure and efficient logistics services", pages 10-13, [online], September 2018, New Energy and Industrial Technology Development Organization, [June 1, 2020], Internet <URL: https://nedo-dress.jp/wp-content/uploads/2018/09/kenkyuu_006.pdf>

However, in the flow described in Non-Patent Document 2, procedures are merely performed manually via email. Note that the test described in Non-Patent Document 1 is merely an interconnection test of the drone traffic management system, and does not take into consideration the procedures related to creating an operation plan. Therefore, there is a need for a method that can automatically perform a series of procedures, such as designing a route and generating an operation plan that is coordinated with the routes of other autonomous mobile devices, simply by setting the start and end points.

As mentioned above, even if an autonomous mobile device has a set route, it may deviate from the planned route temporarily due to disturbances, and so the route is usually designed to allow a large margin of error to avoid collisions with other autonomous mobile devices. For example, in the case of an autonomous mobile device that flies in the air, such as a drone, a distance margin is provided that is large enough that the device will not extend beyond other operating sections even if it is blown away under the maximum wind speed at which operation is permitted. This large margin must also be secured in terms of time. For example, in the case of an autonomous mobile device that flies in the air, such as a drone, collisions will occur if multiple drones enter the same spatial section at the same time, so the time period during which a drone can occupy a certain space must be long, taking delays into account. This large margin reduces the efficiency of space utilization, and limits the number of autonomous mobile devices that can operate simultaneously in the same time period within a certain space.

However, the technology described in Non-Patent Document 3 handles route design and collision detection for routes of a uniform, fixed width, but does not sufficiently increase the density of autonomous mobile devices such as drones in space or the number of simultaneous operations. Therefore, it is desirable to propose a method for planning (designing) routes that improves the efficiency of space utilization (spatial efficiency, temporal efficiency) and enables multiple autonomous mobile devices to operate efficiently.

The technology described in Patent Document 1 was designed with humans on board and is based on the idea of allowing multiple ships to fit within one large mesh section. This is because, since humans are on board, any collisions that may occur within the section can be avoided by human judgment. As can be seen from the fact that the technology described in Patent Document 1 is designed with ships in mind, it not only uses meshes that divide a two-dimensional plane, but also has a uniform size for these meshes. Furthermore, because the technology described in Patent Document 1 is primarily designed to avoid collisions and is used in environments where human judgment is possible, the meshes are of a uniform size and are not based on the perspective of improving the accommodation density of ships or the number of simultaneous operations.

As such, the technology described in Patent Document 1 cannot automatically create operation plans for autonomous mobile devices that do not have humans on board, and cannot improve the accommodation density and number of simultaneous operations of the autonomous mobile devices.

The objective of the present disclosure is to provide an operation plan generation device, a learning device, an operation plan generation method, a learning method, and a program that can automatically create an operation plan for an autonomous mobile device, and can improve the accommodation density and number of simultaneous operations of the autonomous mobile devices.

The operation plan generation device according to the first aspect of the present disclosure includes a generation unit that generates an operation plan indicating a route along which the autonomous mobile device is scheduled to operate based on operation conditions input for the operation of the autonomous mobile device, and a prediction unit, the route includes information expressed as a series of sections that are areas for avoiding simultaneous or simultaneous coexistence with other autonomous mobile devices, the prediction unit predicts the difference between the operation plan and the operation performance, which is the actual operation performance based on the operation plan, for at least each section on the route included in the operation plan, and the generation unit generates an operation plan for the autonomous mobile device based on the operation conditions using new sections updated according to the difference for each section predicted by the prediction unit.

The learning device according to the second aspect of the present disclosure includes an input unit that inputs an operation plan indicating a planned route for an autonomous mobile device and operation results that are the results of actual operation based on the operation plan, and a learning unit, where the route includes information expressed as a series of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices, and the learning unit analyzes, for each section used to generate the operation plan, the difference between the operation plan and the operation results that are the results of actual operation based on the operation plan, and generates a learned model that predicts the difference that may occur in a new operation plan based on the analysis results.

The operation plan generation method according to the third aspect of the present disclosure includes a generation step of generating an operation plan indicating a route along which the autonomous mobile device is scheduled to operate based on operation conditions input for the operation of the autonomous mobile device, and a prediction step, the route including information expressed as a series of sections that are areas for avoiding simultaneous or simultaneous coexistence with other autonomous mobile devices, the prediction step predicts the difference between the operation plan and the operation performance, which is the actual operation performance based on the operation plan, for at least each section on the route included in the operation plan, and the generation step generates an operation plan for the autonomous mobile device based on the operation conditions using new sections updated according to the difference for each section predicted in the prediction step.

The learning method according to the fourth aspect of the present disclosure includes an input step of inputting an operation plan indicating a route along which an autonomous mobile device is scheduled to operate and operation results that are the results of actual operation based on the operation plan, and a learning step, in which the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or concurrent coexistence with other autonomous mobile devices, and the learning step analyzes, for each section used to generate the operation plan, the difference between the operation plan and the operation results that are the results of actual operation based on the operation plan, and generates a learned model that predicts the difference that may occur in a new operation plan based on the analysis results.

A program according to a fifth aspect of the present disclosure causes a computer to execute a generation step of generating an operation plan showing a route along which the autonomous mobile device is scheduled to operate based on operation conditions input for the operation of the autonomous mobile device, and a prediction step, the route including information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous coexistence with other autonomous mobile devices, the prediction step predicts the difference between the operation plan and the operation performance, which is the performance of actual operation based on the operation plan, for at least each section on the route included in the operation plan, and the generation step generates an operation plan for the autonomous mobile device based on the operation conditions using new sections updated according to the difference for each section predicted in the prediction step.

A program according to a sixth aspect of the present disclosure causes a computer to execute an input step of inputting an operation plan indicating a planned route for an autonomous mobile device and operation results that are the results of actual operation based on the operation plan, and a learning step, the route including information expressed as a succession of sections that are areas for avoiding simultaneous or concurrent presence with other autonomous mobile devices, and the learning step analyzes, for each section used to generate the operation plan, the difference between the operation plan and the operation results that are the results of actual operation based on the operation plan, and generates a learned model that predicts the difference that may occur in a new operation plan based on the analysis results.

The present disclosure makes it possible to provide an operation plan generation device, a learning device, an operation plan generation method, a learning method, and a program that can automatically create an operation plan for an autonomous mobile device, and can improve the accommodation density and number of simultaneous operations of the autonomous mobile devices.

1 is a block diagram showing a configuration example of an operation plan generation device according to a first embodiment. FIG. 2 is a block diagram showing an example configuration of a learning device that generates a trained model that can be used in the operation plan generation device of embodiment 1. FIG. 3 is a flow diagram for explaining an example of processing in the learning device of FIG. 2 . 2 is a flow diagram for explaining a processing example in the operation plan generating device of FIG. 1 . FIG. 11 is a block diagram showing a configuration example of a traffic management device according to a second embodiment. 6 is a schematic diagram showing an example of an operation plan held in the operation management device of FIG. 5 . 10 is a schematic diagram showing an example of an operation plan generated based only on static operation sections, which is stored in a traffic management device according to a comparative example. FIG. 6 is a schematic diagram showing an example of an operation plan generated by the operation management device of FIG. 5 . FIG. 9 is a schematic diagram showing an example of operation results based on the operation plan of FIG. 8 . 6 is a schematic diagram showing an example of a driving record held in the driving management device of FIG. 5 . 6 is a schematic diagram showing an example of a static operation section held in the operation management device of FIG. 5 . FIG. 12 is a schematic diagram for explaining the information contained in the static operation section of FIG. 11 . 6 is a block diagram showing an example of the configuration of a learning unit in the traffic management device of FIG. 5 . 14 is a schematic diagram showing an example of position difference information extracted by the learning unit in FIG. 13 . 14 is a schematic diagram showing an example of time difference information extracted by the learning unit in FIG. 13 . 14 is a schematic diagram for explaining an example of pre-learning processing executed by the learning unit in FIG. 13 . 14 is a schematic diagram showing an example of a result of execution of pre-learning processing held in the learning unit of FIG. 13 . FIG. 14 is a schematic diagram for explaining an example of a multiple regression analysis process executed in the learning unit of FIG. 13 . 14 is a schematic diagram showing an example of a result of execution of a multiple regression analysis process held in a learning unit in FIG. 13 . 14 is a schematic diagram showing another example of the execution result of the multiple regression analysis process held in the learning unit of FIG. 13 . FIG. 21 is a schematic diagram for explaining the execution results of FIGS. 19 and 20 . 6 is a block diagram showing an example of the configuration of a prediction unit in the traffic management device of FIG. 5 . 23 is a schematic diagram showing an example of a predicted value for each operation section predicted by the prediction unit of FIG. 22 . 23 is a schematic diagram showing an example of a dynamic operation section generated by the prediction unit of FIG. 22 and stored in the dynamic operation section storage unit. FIG. 25 is a schematic diagram for explaining the dynamic operation section of FIG. 24 . FIG. 25 is a schematic diagram for explaining the dynamic operation section of FIG. 24 . FIG. 25 is a schematic diagram for explaining the dynamic operation section of FIG. 24 . FIG. 25 is a schematic diagram for explaining the dynamic operation section of FIG. 24 . 6 is a schematic diagram showing an example of an operation plan generated by the operation management device of FIG. 5 . 6 is a schematic diagram for explaining an example of a method for generating an operation plan in the operation management device of FIG. 5 . 6 is a schematic diagram for explaining an example of a method for generating an operation plan in the operation management device of FIG. 5 . FIG. 32 is a schematic diagram showing an example of an operation plan generated by the generation method of FIGS. 30 and 31 . 6 is a flow diagram for explaining an example of a flow from an application for operation of an autonomous mobile device using the operation management device of FIG. 5 to arrival at a destination. 6 is a flow diagram for explaining an example of a learning process in the traffic management device of FIG. 5 . FIG. 6 is a flow diagram for explaining an example of a prediction process in the traffic management device of FIG. 5 . 6 is a flow diagram for explaining an example of an operation plan generation process in the operation management device of FIG. 5 . 23 is a schematic diagram for explaining another example of the prediction process in the prediction unit of FIG. 22 . 23 is a schematic diagram for explaining another example of the prediction process in the prediction unit of FIG. 22 . 23 is a schematic diagram for explaining another example of the prediction process in the prediction unit of FIG. 22 . FIG. 2 illustrates an example of a hardware configuration of the apparatus.

Hereinafter, the embodiments will be described with reference to the drawings. In the embodiments, the same or equivalent elements may be denoted by the same reference numerals, and duplicate descriptions will be omitted as appropriate.

<Embodiment 1>
An operation plan generating device according to the first embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the configuration of the operation plan generating device according to the first embodiment.

As shown in FIG. 1, the operation plan generation device 1 according to this embodiment is a device that generates an operation plan for an autonomous mobile device, and can include a generation unit 1a and a prediction unit 1b. The operation plan generation device 1 can also be said to generate (design) a route for the autonomous mobile device, and can therefore also be called a route generation device, route design device, etc. Note that the operation plan generation device 1 can be constructed as a standalone device, but can also be constructed as a distributed system in which functions are distributed.

We will explain the autonomous mobile device for which the operation plan is generated. An autonomous mobile device is a device that can move autonomously, and its movement according to the operation plan can be controlled based on the detection results of various sensors such as position sensors and attitude sensors equipped with it.

The autonomous mobile device may be any type of autonomous mobile device, such as a drone (unmanned aerial vehicle), an autonomous vehicle (including a manually operable vehicle), an aircraft or ship with an automatic navigation function, or a robotic electric wheelchair. Here, the drone may be operated in any environment, such as on the ground, in the air, on the water, underwater, or in space, and the robot may be moved by any method, such as running or walking. Furthermore, the autonomous mobile device may be used for any purpose, such as forklifts and unmanned guided vehicles (automated guided vehicles) used in factories and warehouses, construction vehicles such as construction machinery and heavy machinery, railroad vehicles, taxis, trucks, and other vehicles used in logistics, police vehicles, and fire engines. Here, the railroad vehicles may be moved by any method, such as light rail, steel wheel type, new transportation system, monorail, or magnetic levitation type.

As exemplified by the manually drivable automobile, an autonomous mobile device may be configured to allow a human to ride on it, and may also be configured to allow it to be driven by a passenger or by external control. In this way, an autonomous mobile device only needs to be capable of moving autonomously, and may also have the function of moving according to the operation of a passenger or according to control from an external controller.

The operation plan generation device 1 according to this embodiment can be used, for example, for a drone, to mediate, manage, and assign operation routes to businesses that fly beyond visual line of sight for tasks such as transporting goods, taking photographs, and monitoring. In Japan, such mediation and other tasks are performed by the Ministry of Land, Infrastructure, Transport and Tourism. Of course, tasks that can use the operation plan generation device 1 are not limited to drones, but also include the operation and management of various autonomous mobile devices as described above and tasks that use these. Tasks that use autonomous mobile devices are not limited to the transportation and manufacturing industries, but also include infrastructure inspection, maintenance, and management, construction and civil engineering, public (flood control, dam management, disaster prevention, disaster response, police, firefighting), and security.

The generation unit 1a generates an operation plan that indicates the planned route of operation of the autonomous mobile device based on the operation conditions input for the operation of the autonomous mobile device. It can be said that it is preferable to input the operation conditions by communication from an external device from the viewpoint of user convenience and reduction of the processing burden on the administrator of the operation plan generation device 1.

Here, the operating conditions can include temporal conditions and spatial conditions. The temporal conditions can be conditions indicating a desired departure date and time and/or a desired arrival date and time (either of which can be a time range with a certain degree of flexibility). The spatial conditions can include a desired departure point and a desired arrival point, and can also include conditions indicating a certain degree of a desired route, such as a desired stopover point. If the autonomous mobile device is a flying device, the desired route can be a route that includes altitude, and can also include the flight altitude allowed by the autonomous mobile device according to regulations and/or performance.

The route referred to in this embodiment includes information (section route information) expressed as a succession of sections, which are areas for avoiding simultaneous or simultaneous coexistence with other autonomous mobile devices (interference at the same time or time). These sections are set to exclude intrusions of other autonomous mobile devices at the same time or time, and can be set according to the margin of the autonomous mobile device's operating route. These sections can be two-dimensional plane sections, but in cases where the autonomous mobile device is an autonomous flying device that flies, they are sections that assume a three-dimensional space. As will be described later, in this embodiment, these sections are variable. In this way, the planned operating route in this embodiment can be expressed as continuous planned operating sections.

The prediction unit 1b predicts the difference between the operation plan and the operation results (actual operation results) based on the operation plan for at least each section on the route included in the operation plan. The above difference can also be called a difference, a deviation, etc. Of course, sections other than the route included in the operation plan can also be the subject of prediction. For example, a difference prediction may be performed for all sections between the desired departure point and the desired arrival point. In addition, when predicting the above difference for the first autonomous mobile device, it is because the routes of the second autonomous mobile device that wishes to operate during the same time period may cross even if they are avoided in terms of time. Note that the crossing of routes means that the route of the second autonomous mobile device includes sections unrelated to the route of the first autonomous mobile device. In this way, the prediction unit 1b can make predictions taking into account other operation plans during the same period.

Furthermore, the generation unit 1a generates an operation plan for the autonomous mobile device based on the above operation conditions, using new sections updated according to the differences between each section predicted by the prediction unit 1b. This operation plan can be presented to a user such as a business (a client who has requested the creation of an operation plan). Thus, in this embodiment, the sections are variable.

In addition, the generation unit 1a can generate an operation plan for the autonomous mobile device using the set sections, update the sections as necessary through prediction by the prediction unit 1b, and update the operation plan for the autonomous mobile device using the updated sections. However, the generation unit 1a can be configured to generate an operation plan based on the operation conditions using the sections predicted by the prediction unit 1b. Therefore, if such sections have already been set, the generation unit 1a can present the operation plan as is to the user as the final version without having to perform prediction again.

The operation plan generation device 1 may also include a control unit (not shown) that controls the entire device, and this control unit may be configured to execute the functions of the generation unit 1a and the prediction unit 1b. This control unit may be realized, for example, by a CPU (Central Processing Unit), a working memory, and a non-volatile storage device that stores a program. This program may be an operation plan generation program. This control unit may also be realized by including, for example, an integrated circuit.

It is also preferable that the prediction unit 1b predicts the above-mentioned difference using a trained model. Note that the trained model here may be a trained model that has been determined to be used in the operational phase, and can be updated successively to further improve accuracy, etc.

Next, such a trained model that can be used by the operation plan generation device 1 will be described with reference to FIG. 2. FIG. 2 is a block diagram showing an example of the configuration of a learning device that generates a trained model that can be used by the operation plan generation device 1.

As shown in FIG. 2, the learning device 2 can include a learning unit 2a and an input unit 2b. The learning device 2 can be constructed as a standalone device, but can also be constructed as a distributed system with distributed functions. In addition, the learning device 2 will be described on the assumption that it is provided outside the operation plan generation device 1, but it can also be provided inside the operation plan generation device 1 as a learning unit.

The input unit 2b inputs an operation plan indicating the planned route of operation of the autonomous mobile device, and operation results, which are the results of actual operation based on the operation plan. The operation results can be input at any time from the autonomous mobile device, for example, at any time or at the end of operation. The route is as described above, and the operation plan includes information on sections. Note that the operation results may not include section information at the time of input, but section information may be added after input based on the operation plan.

The learning unit 2a analyzes the difference between the operation plan and the actual operation results based on the operation plan for each section used to generate the operation plan. The data set to be analyzed here is data showing operation plans generated for multiple autonomous mobile devices for various time periods and the corresponding operation results, and preferably data that includes environmental information such as weather in the operation results, thereby improving the accuracy of the analysis.

Then, based on the analysis results, the learning unit 2a generates a trained model that predicts differences that may occur in the new operation plan (at least for the sections included in the route of the new operation plan). The difference to be predicted is the difference between the operation plan and the operation results for each section, as described above. For convenience, the analysis and model generation are described in stages, but the learning unit 2a is not limited to applying such staged processing, and may generate a trained model as a result of the analysis described above.

The trained model generated by the learning unit 2a can be a model obtained by a statistical learning method or a model obtained by a machine learning method such as deep learning, regardless of the algorithm or method. Of course, it can also be a model that falls into both categories.

For example, the trained model may be a model generated using multiple regression analysis as described later in the second embodiment. The algorithm of the untrained model that is the basis of the trained model generated by the learning device 2 may be, for example, a CNN (Convolutional Neural Network) having a convolutional layer, a pooling layer, and a fully connected layer. The algorithm may be an algorithm suitable for prediction having a recursive structure such as an RNN (Recurrent Neural Network). The RNN may be a neural network extended to have a long short-term memory (LSTM) block in order to alleviate the gradient vanishing problem. Of course, the untrained model and trained model are not limited to those using DNNs such as these, even when a machine learning method is used.

The accuracy of the generated trained model can be checked, and if the accuracy is good, it can be implemented as is. If the accuracy is poor, preprocessing such as convolution processing can be changed, or tuning can be performed. Trained models can then be generated and evaluated to determine which trained model to implement. Any hyperparameter can be tuned. Examples of hyperparameters include the number of layers in the NN, the number of units (number of nodes) in each layer, and the number of repeated training cycles (number of epochs) using the same dataset.

As described above, the process of generating a trained model may be an update process. In other words, the trained model can be updated at any time according to the operation method, such as when a certain amount of new training data has been accumulated.

The learning device 2 can be equipped with a storage unit (not shown) that stores the trained model generated by the learning unit 2a, and an untrained model can be stored in this storage unit and updated to become a trained model. The prediction unit 1b uses the trained model thus generated during the operation phase to perform difference prediction. Therefore, the learning device 2 only needs to be connected in a state accessible from the operation plan generation device 1. The learning device 2 can also be configured to transmit the generated trained model, for example, to the operation plan generation device 1 via a network, or to transmit the generated trained model via a network to a server device accessible from the operation plan generation device 1.

The learning device 2 may include a control unit (not shown) that controls the entire device, and this control unit may be configured to execute some of the functions of the learning unit 2a and the input unit 2b. This control unit may be realized, for example, by a CPU, a working memory, and a non-volatile storage device that stores a program. This program may be a learning program. This control unit may also be realized, for example, by including an integrated circuit.

In this way, the operation plan generation device 1 according to this embodiment can constitute an operation plan generation system together with the learning device 2 described above, and this operation plan generation system can also be equipped with a server device as described above.

Next, an example of the processing of the learning device 2 and the operation plan generating device 1 will be described with reference to Figures 3 and 4. Figure 3 is a flow diagram for explaining an example of the processing of the learning device 2, and Figure 4 is a flow diagram for explaining an example of the processing of the operation plan generating device 1.

First, in the learning stage (model generation process), the learning device 2 inputs an operation plan indicating the planned route of operation of the autonomous mobile device and operation results, which are the results of actual operation based on the operation plan (step S1). Next, the learning device 2 analyzes the difference between the operation plan and the corresponding operation results for each section used to generate the operation plan, and generates a learned model that predicts the above-mentioned differences that may occur in a new operation plan based on the analysis results (step S2). Note that the data to be analyzed in step S2 can be the data input in step S1 itself, but it is preferable to perform preprocessing if necessary. Note that the above steps S1 and S2 can be called the input step and the learning step, respectively.

In the operation stage (operation plan generation process), the operation plan generation device 1 generates an operation plan indicating the planned route of operation of the autonomous mobile device based on the operation conditions input for the operation of the autonomous mobile device (step S11). The operation plan generation device 1 predicts the difference between the operation plan and the operation results, which are the results of actual operation based on the operation plan, for at least each section on the route included in the operation plan (step S12).

Here, step S11 uses new sections updated according to the differences between sections predicted in step S12 to generate an operation plan for the autonomous mobile device based on the above operating conditions. In other words, in the operation phase, an operation plan based on the above operating conditions can be generated using sections based on the prediction. Note that steps S11 and S12 can be referred to as a generation step and a prediction step, respectively.

With the operation plan generation device 1 configured as described above, a series of procedures can be automatically performed, such as designing an operation route and generating an operation plan that is coordinated with the routes of other autonomous mobile devices, simply by setting, for example, the start point (desired departure point), end point (desired arrival point), and time conditions.

Furthermore, in the operation plan generation device 1, the sections that avoid the coexistence of other autonomous mobile devices at the same time or in the same time period are variable, and the sections used simultaneously can be made uneven in size (unevenly divided). Therefore, in the operation plan generation device 1, even if the size of the sections is variable, collisions can be avoided and the accommodation efficiency (accommodation density) per space can be increased without compromising safety, and the number of vehicles that can operate simultaneously can be increased. For example, in the operation plan generation device 1, for sections (small places) where the difference between the planned and actual operation routes is small, the number of vehicles that can be accommodated simultaneously per space (number of simultaneous operations) can be increased. Therefore, in the operation plan generation device 1, the utilization efficiency of the operation environment can be improved spatially and temporally, and multiple autonomous mobile devices can be operated efficiently. In addition, in the operation plan generation device 1, since the variable sections correspond to protected areas such as protected airspace, there is no need for a human being to board the autonomous mobile device and perform risk avoidance operations based on the judgment of the person on board.

In this way, the operation plan generation device 1 according to this embodiment can not only automatically create (generate) an operation plan for an autonomous mobile device that avoids collisions between the autonomous mobile devices, but also improve the accommodation density and the number of simultaneous operations of the autonomous mobile devices. In other words, this embodiment makes it possible to prevent collisions while improving the efficiency of the use of space and time. Furthermore, the learning device 2 as described above can generate a trained model that can be used in the operation plan generation device 1 so that the operation plan generation device 1 can be configured as a device that achieves such effects.

<Embodiment 2>
The second embodiment will be described with reference to Figures 5 to 39, focusing on the differences from the first embodiment, but various examples described in the first embodiment can be applied. For example, the second embodiment will be described mainly with an example applied to an operation plan of an autonomous mobile device such as a drone, but other examples can also be applied. Also, for example, the present embodiment will be described with an example in which multiple regression analysis is used in the learning unit and the prediction unit, but other statistical methods such as other multivariate analysis may be adopted as the learning method, and machine learning such as deep learning may also be used.

<Configuration example>
A configuration example of the present embodiment will be described with reference to Fig. 5 to Fig. 32. First, a configuration example of the operation management device according to the present embodiment will be described with reference to Fig. 5. Fig. 5 is a block diagram showing one configuration example of the operation management device according to the present embodiment.

As shown in FIG. 5, the operation management device 10 according to this embodiment can include a learning unit 11, a learning model storage unit 12, a prediction unit 13, a dynamic operation section storage unit 14, an operation record storage unit 15, a static operation section storage unit 16, an operation plan storage unit 17, and an operation plan generation unit 18.

As will be clear from the explanation below, the operation management device 10 is configured to have the functions of the operation plan generation device 1 in FIG. 1 and the functions of the learning device 2 in FIG. 2. Each of the storage units, the dynamic operation area storage unit 14, the operation record storage unit 15, the static operation area storage unit 16, and the operation plan storage unit 17, can have a common or separate storage device for storing information.

The operation management device 10 is also configured to be able to accept input from a terminal device such as a PC (Personal Computer) used by the user 4, and therefore may be equipped with a communication unit (not shown) for communicating with the terminal device. The user 4 is a person or business that wants to operate the autonomous mobile device 3. For example, in the case of a drone, the user 4 would be a person who wants to fly beyond visual line of sight for work such as transporting luggage, taking photographs, or monitoring. The user 4 applies to the operation management device 10 for the desired departure and arrival points and operating times of the autonomous mobile device 3, and is issued an operation plan. Of course, the user 4 is not limited to a person, and can also be another system for operating an autonomous mobile device.

As described in the first embodiment, the autonomous mobile device 3 to be managed by the traffic management device 10 refers to, for example, drones, robots, self-driving cars, automated transport vehicles in factories, ships and aircraft equipped with automatic navigation functions, etc. The autonomous mobile device 3 is configured to be capable of moving to a destination by autonomous control, using the equipped positioning function and sensors to correct for external disturbances such as wind, while following (adhering to) a given travel route and planned passing time as much as possible. Of course, the autonomous mobile device 3 may be partially controlled by human operation, and it is sufficient that the device is at least capable of autonomous movement control.

The autonomous mobile device 3 can also determine its own position, retain a history of its own operating history (coordinates passed and dates and times), and have the function of transmitting (reporting) these to the operation management device 10. For this reason, the autonomous mobile device 3 can be equipped with a communication unit (not shown) that communicates with the operation management device 10. This communication unit can be at least one of a wireless communication unit and a wired communication unit, depending on the type of autonomous mobile device 3, the timing of sending and receiving data with the operation management device 10, etc. If this communication unit is a wired communication unit, communication will usually be performed when movement based on the operation plan is completed.

The various components of the operation management device 10 will be explained. The operation management device 10 can be said to be a device that supports the safe and reliable operation of the autonomous mobile device 3, and its main functions are to remotely monitor the operation status (operation location, status, etc.) of the autonomous mobile device 3 during operation, and to generate an operation plan for the autonomous mobile device 3 (operation plan generation function).

When the traffic management device 10 performs remote monitoring, the target autonomous mobile device 3 will have a communication unit that communicates with the traffic management device 10. Unless the autonomous mobile device 3 is a device that moves over an infrastructure that allows wired communication, such as a railway, this communication unit will be limited to a wireless communication unit.

The operation plan generation function corresponds to the function of the operation plan generation device 1 in FIG. 1, and is a function that arbitrates and plans routes and times so that autonomous mobile devices do not collide with each other. In addition, the operation management device 10 can have a function to present (dispatch) the generated operation plan to the target autonomous mobile device 3 and its user 4. Presentation to the autonomous mobile device 3 can be performed by communication, and presentation to the user 4 can be performed by communication with the user 4's terminal device.

The operation plan generation unit 18 receives an operation application for the autonomous mobile device 3 from the user 4 and generates an operation plan. The operation plan generation unit 18 normally receives multiple operation applications. The operation plan generation unit 18 generates an operation plan so that multiple autonomous mobile devices (including the autonomous mobile device 3) do not enter the same area at the same time or time period. In other words, even when there are multiple autonomous mobile devices 3 wishing to move along a route where there is a possibility of collision, the operation plan generation unit 18 generates an operation plan for each of the multiple operation applications received so as to avoid collisions.

In this way, the operation plan generation unit 18 receives the starting point, end point, desired departure and arrival times, etc. of the operation from the user 4, and generates an operation plan that does not cause conflicts with other autonomous mobile devices in the same area and that meets the user's wishes as much as possible. Details of the operation plan generation unit 18 will be described later as a related element in the explanation of the prediction unit 13 with reference to FIG. 22.

The operation plan storage unit 17 stores the operation plans generated (planned) and issued by the operation plan generation unit 18. The stored operation plans can be referenced when generating new operation plans in order to avoid conflicts between autonomous mobile devices in the same area at the same time or in the same time period, and to avoid duplication of operation plans.

Now, with reference to Figures 6 and 7, an example of an operation plan and an example of the format of information stored as the operation plan will be described. Figure 6 is a schematic diagram showing an example of an operation plan stored in the operation plan storage unit 17 of the operation management device 10. Figure 7 is a schematic diagram showing an example of an operation plan generated based only on static operation sections in which all sections have the same size, stored in a operation management device according to a comparative example.

One operation plan is generated for each application from the user 4 (one operation of the autonomous mobile device 3). One operation plan can be a set of the following information arranged in the order that the autonomous mobile device 3 should follow when operating, as exemplified in FIG. 6. The set can include a zone ID indicating the zone in which the autonomous mobile device 3 may operate, a target arrival coordinate (e.g., the zone's center of gravity coordinate) to be targeted when passing through the zone, and a target arrival time that is the target time for passing through the target arrival coordinate.

In the example of FIG. 6, the section is a dynamic section, and the section is indicated by a section ID and a division ID, but this will be described in detail later in the explanation of the prediction unit 13 with reference to FIG. 22 etc., and in particular in the explanation of the division mechanism with reference to FIG. 24 to FIG. 28. On the other hand, in the traffic management device according to the comparative example, an operation plan is generated based on a static operation section in which all sections are the same size, and as a result, as shown in FIG. 7, a set of information as shown below can be arranged in the order to be followed when the autonomous mobile device operates. The set in FIG. 7 can include a section ID indicating a fixed section, a target arrival coordinate to be targeted when passing through the section (e.g., the coordinates of the center of gravity of the section), and a target arrival time that is the target time for passing through the target arrival coordinate.

In the examples of Figures 6 and 7, the destination coordinates are shown in three dimensions, but they may be two-dimensional when applied to the management of an autonomous mobile device that travels on land or sea surface. In addition, three-dimensional space is expressed using latitude, longitude, and altitude, but even in three-dimensional space, any information that can uniquely indicate a point within the space is sufficient, and is not limited to expressions using latitude, longitude, and altitude.

In addition, in the example of Figure 6, the coordinates of the center of gravity of the section are shown as an example of the target coordinates to be reached, but this does not necessarily have to be the center of gravity depending on the application. For example, if the section is treated in two dimensions, the coordinates of the center of gravity of the section may be used. Also, for example, in the case of an autonomous mobile device that operates in the sky, such as a drone, if the policy is to be more cautious about sudden descents due to wind or other factors than upwards for safety reasons, it is possible to set the coordinates vertically higher than the center of gravity. In this way, the coordinates to be reached can be determined depending on the application.

The operation record storage unit 15 stores the operation status record of the autonomous mobile device 3. The operation status record, i.e., the operation record, can be a history of information transmitted one by one from the autonomous mobile device 3 by remotely monitoring the operation status of the autonomous mobile device 3. As described above, the autonomous mobile device 3 can accumulate a status history while in operation, and register it collectively in the operation record storage unit 15 after the operation is completed, thereby collecting the operation record.

Now, with reference to Figures 8 to 10, examples of operation records will be described along with examples of the format of information stored as operation records. Figure 8 is a schematic diagram showing an example of an operation plan generated by the operation management device 10, and Figure 9 is a schematic diagram showing an example of operation records operated based on the operation plan of Figure 8. Also, Figure 10 is a schematic diagram showing an example of operation records stored in the operation management device 10.

Operation records indicate the actual operation results according to the operation plan, and one is generated for each operation of the autonomous mobile device. One operation record can be a set of information as shown in the example of FIG. 9 for one operation plan as shown in the example of FIG. 8. The set can include the result arrival time, which is the time when the position of the autonomous mobile device 3 was recorded, and the result arrival coordinates, which indicate the position where the autonomous mobile device 3 was located at that time.

In Figures 8 and 9, the notation hh:mm indicates the target arrival time and the actual arrival time, respectively. Furthermore, the black circles (●) in Figure 8 indicate the target arrival coordinates at the target arrival time, and the black circles (●) in Figure 9 indicate the actual arrival coordinates recorded at the target arrival time. Furthermore, the white circles (◯) in Figure 9 indicate the actual arrival coordinates recorded at regular intervals or regular travel distances for the convenience of operation monitoring.

The timing for recording the position of the autonomous mobile device 3 may be any timing required for operational management, and various methods can be adopted, such as recording at regular intervals or recording when there is a change in the position.

However, in the recording method of this embodiment, it is basically necessary to leave a record of where the autonomous mobile device 3 was at the target arrival time set in the operation plan. It is advisable to record this record together with the section ID so that it can be distinguished from records of other positions. Note that the section ID recorded together here is not the section ID of the location where the autonomous mobile device 3 was, but the planned section ID, which indicates which section ID the autonomous mobile device 3 was planned to be in according to the operation plan at the same planned arrival time as the actual arrival time.

The format of the information stored as operation results can be, for example, as shown in Figure 10, in which section IDs, actual arrival coordinates, and actual arrival times are associated with each other. Figure 10 shows an example in which the section ID column is missing, but it is also possible to record all section IDs without omissions. In that case, it is a good idea to add and record a column of flag information to distinguish that this is a record of actual arrival coordinates that correspond to the planned arrival time.

In the example of FIG. 10, the destination coordinates are shown in three dimensions, but they may be two-dimensional when applied to the management of an autonomous mobile device that travels on land or sea surface. In addition, the three-dimensional space is expressed using latitude, longitude, and altitude, but even in the case of a three-dimensional space, any information that can uniquely indicate a point within that space is sufficient, and the expression is not limited to using latitude, longitude, and altitude.

In the example of Figure 10, the centroid coordinates of the section are shown as an example of the target coordinates to be aimed for, but this does not necessarily have to be the centroid depending on the application. For example, if the section is treated in two dimensions, the central coordinates of the section may be used. It is preferable that the actual coordinates are the same type of coordinates as the target coordinates, as this makes it easier to calculate the difference, but even if they are different, it is possible to calculate the difference by conversion.

If there is information on the sections in the operation plan and the positions and times in the operation records, it is possible to record the two so that they ultimately correspond to each other. For example, even if only the white circles (○) in Figure 9 are recorded, it is possible to approximately calculate the positions of the black circles (●) in Figure 9 by using an interpolation process.

The static operation section storage unit 16 divides the space in which the autonomous mobile device 3 can move into box-shaped spaces of fixed, uniform size, and stores this information as coordinate information for each section. The sections here are called static operation sections. These static operation sections are of a predetermined size. The operation plan generated and output in this embodiment is not an operation route that consists of a string of static operation sections, but rather an operation route that consists of a string of dynamic operation sections (variable sections).

Now, with reference to Figures 11 and 12, we will explain an example of a static operation section in this embodiment, as well as an example of the format of information stored as the static operation section. Figure 11 is a schematic diagram showing an example of a static operation section stored in the operation management device 10, and Figure 12 is a schematic diagram for explaining the information contained in the static operation section of Figure 11.

The information indicating the static operation section shown in FIG. 11 is associated with a section ID, a first vertex (vertex 1), a second vertex (vertex 2), and an altitude. A static operation section is a section that uniformly divides a three-dimensional space as shown in the left diagram of FIG. 12, and can be represented by vertex 1, vertex 2, and the lower and upper altitude limits shown in the right diagram of FIG. 12.

In the examples of Figures 11 and 12, the destination coordinates are shown in three dimensions, but they may be two-dimensional when applied to the management of an autonomous mobile device that travels on land or sea surface, and in that case, for example, upper and lower altitude limits are not necessary. In addition, three-dimensional space is expressed using latitude, longitude, and altitude, but even in three-dimensional space, any information that can uniquely indicate a point within that space is sufficient, and expression using latitude, longitude, and altitude is not limited to this.

In the examples of Figures 11 and 12, the divisions are expressed by two vertices on a plane and the upper and lower limits of the altitude, but this is not limiting and any method can be used to express a division in space. Examples include a method of expressing it using the coordinates of eight vertices, a method of specifying the length of the side of each division and the direction that divides the space and expressing it using the coordinates of one representative vertex, and a method of specifying the direction that divides the space and expressing it using the coordinates of three vertices.

The learning unit 11 analyzes the schedule and performance from the operation plan and operation performance for each static operation section, and generates a learned model. The learning model holding unit 12 holds the learned model that is the result of learning by the learning unit 11. Predictions are performed using the learned model held here, and dynamic operation sections are determined. Details of the learning unit 11 and the learning model holding unit 12 will be described later with reference to FIG. 13.

Based on the learned model, the prediction unit 13 predicts the difference between the operation plan and the actual operation performance during the period of operation conditions provided by the operation plan generation unit 18, and calculates the dynamic operation section. Details of the prediction unit 13, the dynamic operation section storage unit 14, and the operation plan generation unit 18 will be described later with reference to FIG. 22.

The learning unit 11 will be described in detail together with its related elements with reference to Figures 13 to 21. Figure 13 is a block diagram showing an example of the configuration of the learning unit 11 in the traffic management device 10.

As shown in FIG. 13, the learning unit 11 can include a planned performance difference storage unit 11a, a pre-learning processing unit 11b, a target variable explanatory variable storage unit 11c, a planned performance difference extraction unit 11d, and a main learning processing unit 11e. The learning unit 11 is also connected to the operation plan storage unit 17, the operation performance storage unit 15, and the static operation section storage unit 16 at the planned performance difference extraction unit 11d, and is connected to the learning model storage unit 12 at the main learning processing unit 11e.

The planned performance difference extraction unit 11d finds the difference between the operation plan and the operation results after processing the data, such as by aligning the number of data as necessary, so that the operation plan stored in the operation plan storage unit 17 can be compared with the corresponding operation results stored in the operation result storage unit 15. The processing results of the planned performance difference extraction unit 11d are stored in the planned performance difference storage unit 11a.

The method for finding the difference may be to align the times and find the position difference, or to align the positions and find the time difference, or both. By finding the position difference (vector difference) between the operation plan and the operation results, that is, by finding the vector difference between the target arrival coordinate at the target arrival time and the actual arrival coordinate at the same time as the target arrival time, the position deviation distance and position deviation direction can be obtained. At this time, since the section ID is always recorded in each row of the plan information at the target arrival time and the actual information at the same time as the target arrival time, information can be obtained as to which section the deviation occurred when the vehicle should have been in. For the position deviation direction, for example, information expressed as a direction or a reference direction (such as magnetic north) in the case of two dimensions, or a three-dimensional vector in the case of three dimensions can be obtained.

In addition, by calculating the time difference (lag/lead time) between the operation plan and the actual operation results, the target arrival time, the target arrival coordinates, and the lag or lead time relative to the target arrival time can be obtained.

Examples of information items obtained as a result of such processing will be described with reference to Figs. 14 and 15. Figs. 14 and 15 are schematic diagrams showing an example of position difference information and an example of time difference information, respectively, extracted by the planned-actual difference extraction unit 11d of the learning unit 11 in Fig. 13.

As shown in FIG. 14, the result of determining the position difference (position deviation distance) between the operation plan and the operation results can be information that associates the position deviation distance, the position deviation direction, the target arrival coordinates (latitude, longitude, altitude), the target arrival time, and the section ID. Also, as shown in FIG. 15, the result of determining the time difference (lag/lead time) between the operation plan and the operation results can be information that associates the lag/lead time, the target arrival coordinates (latitude, longitude, altitude), and the target arrival time.

The planned performance difference holding unit 11a holds the difference between the operation plan and the operation results obtained by the planned performance difference extraction unit 11d. For example, the planned performance difference holding unit 11a can hold the results of processing such as the planned performance difference extraction unit 11d obtaining the position difference (vector difference) between the operation plan and the operation results. The results held here are the target arrival time, the section ID planned to be at that target arrival time, the target arrival coordinates, and the position deviation distance and position deviation direction relative to the target arrival coordinates. More specifically, the planned performance difference holding unit 11a can hold, for example, the information in FIG. 14 as the difference. The held position deviation direction can also be expressed as a direction from a reference direction (such as magnetic north) in the case of two dimensions, or as a three-dimensional vector in the case of three dimensions.

When the time difference (lag/lead time) between the operation plan and the actual operation is calculated, the planned/actual difference storage unit 11a can store the target arrival time, the target arrival coordinates, and the lag or lead time relative to the target arrival time. More specifically, the planned/actual difference storage unit 11a can store, for example, the information shown in FIG. 15 as the difference.

In addition, in the examples of Figures 14 and 15, the destination coordinates are shown in three dimensions, but they may be two-dimensional when applied to the management of an autonomous mobile device that operates on land or sea surface, and in that case, for example, upper and lower altitude limits are not necessary. In addition, three-dimensional space is expressed using latitude, longitude, and altitude, but even in three-dimensional space, any information that can uniquely indicate a point within that space will suffice, and expression using latitude, longitude, and altitude is not limited to this.

Also, Figures 14 and 15 show examples in which positional deviation information and lead/delay time information are stored in table format. Such positional deviation tables and lead/delay time tables can be stored in separate tables for each operation (each operation), for each model or type of autonomous mobile device, or for each weather condition. As in this example, the operation of storing the difference between the operation plan and the operation results can be performed with a variety of ways to group the data, and it is particularly preferable to perform the operation so that the data is optimal for the processing method in the learning unit 11 in the subsequent process.

As a result, for example, the prediction unit 13 using the learning results can perform prediction of the above-mentioned difference for at least one of the type of autonomous mobile device, the model of autonomous mobile device, and the operating environment.

The pre-learning processing unit 11b processes and replaces data as pre-processing for information indicating the difference between the operation plan and the operation results stored in the plan-performance difference storage unit 11a, so that machine learning can be performed in the subsequent main learning processing unit 11e. An example of the pre-learning processing will be described with reference to FIG. 16. FIG. 16 is a schematic diagram for describing an example of the pre-learning processing executed by the pre-learning processing unit 11b of the learning unit 11.

The learning pre-processing unit 11b can store the objective variables and explanatory variables in the objective variable explanatory variable storage unit 11c. In the learning pre-processing illustrated in FIG. 16, the position difference (position deviation distance, position deviation direction) between the operation plan and the operation results is calculated, and the position deviation distance and position deviation direction are extracted as objective variables and stored in the objective variable explanatory variable storage unit 11c.

In this pre-learning process, the date of the target arrival time corresponding to the objective variable is converted into a season band number, and the hour and minute of the target arrival time are converted into a time zone number, and are stored as explanatory variables in the objective variable explanatory variable storage unit 11c. For example, the following pre-processing is performed on the target arrival time, and it is divided into columns of season numbers and time zone numbers. For seasons, if 3≦M≦5, it is set to 0, if 6≦M≦8, it is set to 1, if 9≦M≦11, it is set to 2, and otherwise it is set to 3. For time zones, if 0≦H≦12, it is set to 0, if 13≦H≦17, it is set to 1, and otherwise it is set to 2. As a result, in the example of FIG. 16, the seasons, which are columns of data, can be expressed as 0 for spring, 1 for summer, 2 for autumn, and 3 for winter, and the time zones can be expressed as 0 am, 1 for afternoon, and 2 for evening. Note that the example shown in FIG. 16 does not intend to limit the number of classifications, such as dividing the seasonal bands into four classifications and dividing the time zones into two classifications, and more sparse or dense classifications may be used.

Among autonomous mobile devices, those that fly in the air or sail on the ocean are particularly susceptible to differences in the time and coordinates of arrival due to the influence of wind, so other information that is expected to correlate with wind speed may be used as explanatory variables. For example, in addition to time periods and seasonal zones, numbers corresponding to the altitude zone of the ground surface can be assigned to the target coordinates and sections to be used as explanatory variables. The type of ground surface and the temperature zone of the air temperature for each section can also be used to convert them into explanatory variables or added as items of explanatory variables. The type of ground surface can be, for example, a number assigned to distinguish between land and sea, or a number assigned to distinguish between forest, desert, rocky area, city, etc.

In addition, if the time lag (advance, lag) is used as the objective variable instead of the positional shift distance and positional shift direction, it is possible to learn the tendency for the position to be advanced or delayed. Furthermore, by using the positional shift and time lag as the objective variables, it is possible to learn both the tendency for the positional shift and the tendency for the position to be advanced or delayed within the section.

The objective variable explanatory variable storage unit 11c stores the objective variables and explanatory variables that are the processing results of the learning pre-processing unit 11b. An example of the format of the information stored here will be explained with reference to FIG. 17. FIG. 17 is a schematic diagram showing an example of the execution result of the learning pre-processing stored in the objective variable explanatory variable storage unit 11c of the learning unit 11.

In the example of pre-learning processing shown in FIG. 16, information is stored in a format as shown in FIG. 17. In the example of FIG. 17, information extracted as objective variables, namely the positional deviation distance and the positional deviation direction, and information converted as explanatory variables into the season band number for the target arrival time and into the time zone number for the hour and minute for the target arrival time are stored in association with each other.

In the example shown in FIG. 17, like the example shown in FIG. 16, there is no intention to limit the number of classifications, such as dividing seasonal zones into four categories or dividing time periods into two categories, and classifications may be made more coarsely or more densely. Other examples of application of explanatory variables are also given as examples corresponding to the information stored in the objective variable explanatory variable storage unit 11c.

That is, among autonomous mobile devices, especially those that fly in the sky and those that navigate on the ocean, the target arrival time and target arrival coordinates tend to differ due to the influence of wind, so other information that is expected to correlate with wind speed may be used as explanatory variables. For example, in addition to time periods and seasonal zones, numbers corresponding to the altitude zone of the ground surface may be assigned to the target arrival coordinates and the section as explanatory variables, or the type of the ground surface, such as the temperature zone of the air temperature for each section, may be used or added as an item. The type of ground surface may be, for example, a number assigned to distinguish between land and sea, or a number assigned to distinguish between forests, deserts, rocky areas, cities, etc. In addition, if the time shift (advance, delay) is used as the objective variable instead of the position shift distance and the position shift direction, the tendency of advance and delay can also be learned. Furthermore, by using the position shift and time shift as the objective variables, both the tendency of position shift and the tendency of advance and delay within the section can be learned.

The processing of the learning main processing unit 11e will be explained based on an example in which the objective variable (positional deviation distance) and explanatory variables (time period number, season number) stored in the objective variable and explanatory variable storage unit 11c are input, but other examples can be similarly applied. The learning main processing unit 11e can perform multiple regression analysis for each section using the objective variable (positional deviation distance) and explanatory variables (time period number, season number) as input, and perform processing to obtain a multiple regression equation for each section. The learning main processing unit 11e stores this generated multiple regression equation and the coefficients of the equation in the learning model storage unit 12 as a learned model.

The mechanism of the multiple regression analysis process applicable to this embodiment will be described with reference to FIG. 18. FIG. 18 is a schematic diagram for explaining an example of the multiple regression analysis process executed by the learning unit 11 in FIG. 13.

The learning processing unit 11e can obtain data corresponding to the black circles (●) in the left graph of FIG. 18 for each section from the data on positional deviation distance, seasonal zone, and time period in the table shown in FIG. 17 (the table in the right graph of FIG. 18), and calculates a multiple regression equation based on this data. The left graph of FIG. 18 is plotted with the positional deviation distance on the vertical axis, the seasonal zone on the first horizontal axis, and the time period on the second horizontal axis.

As a template for a multiple regression equation, for example, the following function can be given.
y = w + _ax1 + _bx2 (linear function)
y = w + _ax1 ² + _bx1 + _cx2 ² + _dx2 (quadratic function)

The main learning processing unit 11e prepares templates for multiple regression equations of several orders (for example, linear to sextic equations) and varies the coefficients (w, a, b, c, d in the above example) for each equation. The main learning processing unit 11e then finds an equation and coefficients that represent a plane (hyperplane) that passes close to the most black circles (●) and is as smooth as possible. This is done for each piece of data with each block ID. The main learning processing unit 11e then stores the obtained set of equations and coefficients for each block in the learning model storage unit 12. Note that while the vertical axis is the positional shift distance, the concept is the same when the vertical axis is the positional shift direction or time advance/delay.

In this way, the learning model holding unit 12 holds the multiple regression equations and their coefficients (learned models) for each section generated by the main learning processing unit 11e. Examples of the format of the information held here will be described with reference to Figures 19 to 21. Figure 19 is a schematic diagram showing an example of the execution results of a multiple regression analysis process held in the learning model holding unit 12, and Figure 20 is a schematic diagram showing another example of the execution results of a multiple regression analysis process held in the learning model holding unit 12. Also, Figure 21 is a schematic diagram for explaining the execution results of Figures 19 and 20.

In the storage example shown in FIG. 19, a table is provided for each variation of the formula and coefficients are stored. In the storage example shown in FIG. 20, a table for formulas and a table for storing coefficients are stored separately. Although FIG. 19 and FIG. 20 show two examples of the format for storing information, the method is not limited to these, and any method that can store multiple regression formulas and the coefficients of those formulas in association with a section ID can be used.

In both the storage examples in Figures 19 and 20, as shown in Figure 21, a multiple regression equation including coefficients is stored for each partition (the partition ID is shown inside each partition). In Figure 21, the partitions are shown on a plane for convenience of explanation, but the partitions are not limited to planes, and the same idea applies in three dimensions. Each partition has a function, and the function will have a different form, such as a linear equation or a quadratic equation. There are as many functions as there are fixed meshes, and in the example of Figure 21, there are 16 functions. In Figure 21, z is the response variable (positional deviation), w is a constant, a, b, c, and d are partial regression coefficients, and x and y are explanatory variables (seasonal zone number, time zone number).

As exemplified here, the trained model can be a model that inputs information including at least one of the operating time periods exemplified by the time period numbers, the operating seasons exemplified by the season numbers, and other operating environments, and outputs a predicted value of the above difference.

Next, the details of the prediction unit 13 will be described together with related elements with reference to Figures 22 to 32. Figure 22 is a block diagram showing an example of the configuration of the prediction unit 13 in the traffic management device 10.

As shown in FIG. 22, the prediction unit 13 can include a planned performance difference prediction unit 13a and a dynamic operation section generation unit 13b. The prediction unit 13 is connected to the learning model storage unit 12 at the planned performance difference prediction unit 13a, and is connected to the static operation section storage unit 16 and the dynamic operation section storage unit 14 at the dynamic operation section generation unit 13b.

The planned-performance difference prediction unit 13a calculates the position deviation that is expected to occur for each section (the deviation of the actual arrival coordinate from the planned arrival coordinate that will occur in the future when passing through each section). The planned-performance difference prediction unit 13a calculates the position deviation that will occur for each section by predicting it using the trained model (stored in the trained model storage unit 12) generated by the main learning processing unit 11e.

The planned performance difference prediction unit 13a obtains the multiple regression equation for each section stored in the learning model storage unit 12, and substitutes the values of the seasonal band (seasonal band number) and time period (time period number) to be predicted into the variables. This allows the planned performance difference prediction unit 13a to obtain a predicted value of the deviation of the actual arrival coordinate from the planned arrival coordinate that will occur in the specified seasonal band and time period in the corresponding operation section.

The position deviations for all time periods and seasons may be recalculated every time the learning dataset is updated (every time the operation record is changed). In addition, if the required seasonal bands and time periods are fixed, such as for operating hours limited to daytime or for issuing operation plans up to two months in advance, the differences may be recalculated and updated for a limited period.

In addition, after a user 4 requests the operation plan generation unit 18 to generate a plan, the plan performance difference prediction unit 13a may narrow down the range, seasonal zone, and time period required for operation that meets the operation conditions included in the request, and perform calculation of the predicted values as described above.

The planned performance difference prediction unit 13a passes the predicted deviation value for each section (operation section) to the dynamic operation section generation unit 13b. An example of the data items of the processing results passed here will be explained with reference to FIG. 23. FIG. 23 is a schematic diagram showing an example of a predicted value for each operation section predicted by the prediction unit 13 of FIG. 22. As illustrated in FIG. 23, the planned performance difference prediction unit 13a can pass, for example, data associating the obtained predicted value (position deviation distance), explanatory variables (time period number and season number), and section ID to the dynamic operation section generation unit 13b.

The dynamic operation section generation unit 13b determines the number of divisions of the section so that even if a deviation occurs, the section will not go beyond the size and the section will be as small as possible, based on the predicted value of the positional deviation of the actual arrival coordinates from the planned arrival coordinates obtained by the planned performance difference prediction unit 13a. The size that does not go beyond the size can also be a size that further includes a margin distance for safety. The dynamic operation section generation unit 13b then assigns a division ID to the divided section, calculates its coordinates, and stores them in the dynamic operation section storage unit 14.

The dynamic operation section generation unit 13b and the dynamic operation section storage unit 14 are an example of a section update unit that generates new sections based on the differences between sections predicted by the planned performance difference prediction unit 13a, and sets the new sections for use in generating an operation plan.

Now, with reference to Figures 24 and 25, we will explain the concept of this ID assignment and the data items output as a result of the process. Figure 24 is a schematic diagram showing an example of a dynamic operation section generated by the prediction unit 13 in Figure 22 and stored in the dynamic operation section storage unit 14, and Figure 25 is a schematic diagram for explaining the dynamic operation section in Figure 24.

In the table illustrated in FIG. 24, a dynamic partition ID including a partition ID, the first row of partition IDs, and the second row of partition IDs is stored in association with vertex 1, vertex 2, altitude, and explanatory variables (time zone number and season number). For example, the partition indicated by the dynamic partition ID in the first record in FIG. 24 represents the partition indicated by partition ID 1, first row ID 3, and second row ID 4 illustrated in FIG. 25.

Note that although Figs. 24 and 25 show an example in which the division is limited to two stages, this is merely an example, and the number of stages can be increased or decreased depending on the difference in the autonomous driving device or the difference in the use. Also, although Figs. 24 and 25 show an example in which the number of divisions in the horizontal and vertical directions is the same, this is merely an example. For example, depending on the difference in the autonomous driving device or the difference in the use, various examples can be given, such as dividing only in the horizontal direction, dividing only in the vertical direction, or dividing the horizontal and vertical divisions into different numbers of stages.

Here, the division mechanism will be explained with reference to Figs. 26 to 28. Figs. 26 to 28 are schematic diagrams for explaining the dynamic operation section of Fig. 24. Note that the specific numerical values exemplified in Figs. 26 to 28 are given for the convenience of explanation, and are not limited to these specific numerical values.

First, if we consider the static operation section to be a cube with each side being 100m long, the number of divisions is set so that the length of each side of the section is at least twice the expected positional deviation distance and the size of each section is minimized. Also, to allow for a safety margin, the length of each side of the section can be set to twice the expected positional deviation distance plus an additional margin distance.

As shown in Figure 26, if the predicted position shift distance for section ID 1 is 17 m and the predicted position shift distance for section ID 8 is 11 m, the number of divisions can be determined for section ID 1 and section ID 8 as shown in Figures 27 and 28, respectively.

As shown in FIG. 27, for section ID 1, when a cube with sides of 100 m is divided into 8, 1/2 of a side (to take into account the distance from the center of the section to the edge of the neighboring section) is 25 m, which is larger than the expected position shift distance of 17 m, so dividing it into 8 is not a problem. On the other hand, when divided into 64 sections, 1/2 of a side is 12.5 m, which is smaller than the expected position shift distance of 17 m, so using a section divided to this extent could result in a collision with another autonomous mobile device, which is a problem. Therefore, section ID 1 is divided into 8 sections. As shown in FIG. 28, for section ID 8, when a cube with sides of 100 m is divided into 8 and 64 sections, 1/2 of a side is 25 m and 12.5 m, respectively, which are both larger than the expected position shift distance of 11 m, so dividing it into 8 and 64 is not a problem. On the other hand, if it is divided into 512 parts, each side will be 6.25 m, which is smaller than the expected position shift distance of 11 m. Therefore, if a section divided to this extent is used, collisions with other autonomous mobile devices may occur, which is problematic. Therefore, section ID 8 will be divided into 64 parts.

As described above, the dynamic operation section storage unit 14 stores the dynamic operation section that is the result of the dynamic operation section generation unit 13b. Here, information on sections that are divided more finely than the static operation sections is stored. The format of the dynamic operation section stored here may be one of those exemplified in Figures 24 and 25. By the operation plan generation unit 18 generating an operation plan using this dynamic operation section, it is possible to generate an operation plan that utilizes space more efficiently than when a static operation section is used.

In this way, it is preferable that the section update unit exemplified by the dynamic operation section generation unit 13b and the dynamic operation section storage unit 14 set new sections so as to minimize at least one of the spatial and temporal aspects of the sections.

In addition, for example, when the difference between the operation plan and the operation results includes a spatial difference (spatial difference, distance difference) between the section included in the operation schedule and the section (section in the operation results) that was actually operated at the same time, the section update unit can set a new section as follows. That is, in that case, the section update unit can set a new section so that the section with a small spatial difference is smaller in at least one of space and time than the section with a large spatial difference among the sections included in the operation schedule.

In addition, for example, when the difference between the operation plan and the operation results includes a time difference, which is the difference between the planned time to pass through a section included in the operation schedule and the time when the section is actually passed through (the time indicated by the operation results), the section update unit can set a new section as follows. That is, in this case, the section update unit can set a new section such that sections with a small time difference among sections included in the operation schedule are smaller in at least one of space and time than sections with a large time difference.

As in these examples, in this embodiment, when designing a route, the margin and exclusion time of the entire space (all sections) are not set to a fixed uniform size or length. As in these examples, in this embodiment, in places where there is little deviation from the planned route based on operational performance, it is preferable to make the divisions smaller, and conversely, if this deviation is large, to make the divisions larger. Also, in this embodiment, if there is a large deviation between the planned arrival time and the actual arrival time, it is preferable to make the time for excluding the section longer, and conversely, if the deviation is small, to make it shorter.

As described above, the operation plan generation unit 18 receives an operation application for the autonomous mobile device 3 from the user 4 and generates an operation plan. The operation plan generation unit 18 generates an operation plan so that multiple autonomous mobile devices do not enter the same area at the same time or time period. The operation plan generation unit 18 receives operation conditions such as the starting point and end point of the operation and the desired departure and arrival times from the user 4, and generates an operation plan so that there is no competition between autonomous mobile devices in the same area and so that the desired departure and arrival times are met as much as possible.

Here, the reason why it is said that the request will be met as much as possible is that, for example, there may be an operation plan for another autonomous mobile device that has been issued in the past, and a situation may arise in which it is not possible to generate an operation route that matches the operation conditions. To explain this more specifically, there may be cases where the same start point and end point at the same time have already been requested in an operation plan issued in the past, or where it is necessary to detour a section used in an operation plan that has already been issued. Also, instead of or in addition to such a detour, it may be necessary to shift the time of operation in that section. Therefore, even when operation conditions are input, there may be cases in which it is not possible to completely meet the desired departure and arrival times indicated by the operation conditions, and therefore, it may not be possible to generate an operation route that matches the input operation conditions, and an operation route that meets the request as much as possible may be generated.

The operation plan will be described with reference to FIG. 29. FIG. 29 is a schematic diagram showing an example of an operation plan generated by the operation management device 10. As illustrated in FIG. 29, the operation plan is expressed as information in which operation sections (dynamic operation sections) stored in the dynamic operation section storage unit 14 are linked from a start point to an end point based on the wishes (operation conditions) of the user 4. In FIG. 29, the route connected by black circles (●) constitutes the operation plan for one operation, and the operation plan generation unit 18 can create an operation plan with a route based on the dynamic operation sections whose size varies depending on the location, as shown in FIG. 29.

An example of a method for generating an operation plan so that other autonomous mobile devices do not enter the same section at the same time or in the same time period will be described with reference to Figs. 30 and 31. An example of an operation plan generated as a result will be described with reference to Fig. 32. Figs. 30 and 31 are schematic diagrams for explaining an example of a method for generating an operation plan in the operation management device 10. Fig. 32 is a schematic diagram showing an example of an operation plan generated by the generation method of Figs. 30 and 31.

The operation plan storage unit 17 stores operation plans that have already been assigned for other autonomous mobile devices, as shown in the right diagram of FIG. 30. In these assigned operation plans, the target arrival coordinates (latitude, longitude, altitude) and the target arrival time are associated with each section ID (here, section IDs 15, 11, and 7).

In this state, an example will be given of a case where the user 4 inputs operation conditions including the coordinates (latitude, longitude, altitude) of the desired start point and end point for operation of the autonomous mobile device 3 and the desired arrival time. In this case, the operation plan generation unit 18 searches for the section ID corresponding to the coordinates of the start point (in this example, section ID 2) and the section ID corresponding to the end point (in this example, section ID 12) from the dynamic operation section stored in the dynamic operation section storage unit 14 in accordance with the operation conditions. Then, as shown in the left diagram of Figure 30, the operation plan generation unit 18 associates the section ID as the search result with those coordinates and the desired arrival time, and temporarily stores them in the operation plan storage unit 17 as information on the start point and end point of the operation plan.

The next step in generating an operation plan for the autonomous mobile device 3 will be explained with reference to the example in Figure 31. Figure 31 shows meshes (rectangular shapes) that exist in the operation area (the delivery area if the user 4 is a delivery company). In this example, for simplicity of explanation, it is assumed that there are 16 meshes in total. The numbers in the bottom right of the meshes in Figure 31 are waypoint IDs.

The operation plan generation unit 18 searches for a route from the start coordinates to the end coordinates, as exemplified in the left diagram of FIG. 31. For example, if the route proceeds through waypoint IDs 2, 6, 10, 11, and 12, there will be a conflict with the operation plan already issued and held in the operation plan holding unit 17 in section 11 (waypoint ID = 11). Note that a conflict refers to a state in which two or more autonomous mobile devices exist in the same section at the same time period.

The operation plan generation unit 18 therefore re-searches for a route other than this one that is the shortest. For example, as shown in the right diagram of FIG. 31, if the route is to proceed through waypoint IDs 2, 3, 4, 8, and 12, this route is determined as there is no conflict with the operation plan that has already been issued (the routes do not overlap). In this way, once the route from the start point (section 2) to the end point (section 12) has been determined (the conflict is resolved), the operation plan is completed. The operation plan generation unit 18 stores (updates) the completed operation plan for the autonomous mobile device 3 in the operation plan storage unit 17.

In addition, if the autonomous mobile device 3 is a device that can stop and wait, it can take into consideration the possibility of waiting outside the competing section and shifting the time period, and therefore not perform a re-search, or it can perform a re-search while taking into consideration such a shift in the time period.

The completed operation plan is stored with the section IDs arranged in the order of the route associated with the target arrival coordinates and the target arrival time, as shown in the example of FIG. 32. For convenience, these section IDs are shown in a single column in FIG. 32, but they are dynamic section IDs, as shown in the examples of FIG. 24 and in the balloon table in FIG. 32.

<Example of operation>
Next, an overall flow of processing using the traffic management device 10 according to this embodiment will be described with reference to Fig. 33. Fig. 33 is a flow diagram for explaining an example of a flow from an operation application of an autonomous mobile device using the traffic management device 10 to arrival at a destination.

First, the user 4 registers application information for operating the autonomous mobile device 3 with the operation management device 10 (step S21). Details of the application information are as described above in the explanation of the operation plan generation unit 18.

Next, the operation management device 10 determines an operation plan (operation route and operation schedule) based on the registered application information, and issues the operation plan to the autonomous mobile device 3 (step S22). The operation schedule is information related to the date and time of operation. Alternatively, the operation route and operation schedule can be issued to the user 4, who can then input them into the autonomous mobile device 3. Details of the operation route and operation schedule are as described above in the explanation of the operation plan generation unit 18.

The autonomous mobile device 3 starts moving to the destination by autonomous control based on the route and schedule issued (or input by the user 4) (step S23). Details of the movement to the destination by autonomous control are as described above in the explanation of the autonomous mobile device 3.

The autonomous mobile device 3 notifies the operation management device 10 of operation record information (operation record) such as its own position information at a predetermined timing (step S24). Alternatively, the operation record information can be recorded in the autonomous mobile device 3 at a predetermined timing, and the operation record information can be retrieved after arriving at the destination or at the end of a day's operation, and submitted to the operation management device 10 and stored in the operation record storage unit 15. Details of the notification of operation record are as described above in the explanation of the operation record storage unit 15.

The autonomous mobile device 3 arrives at the destination by autonomous control based on the issued (or input by the user 4) operating route and operation schedule (step S25). The movement to the destination by autonomous control is as described above in the explanation of the autonomous mobile device 3, and the operating route and operation schedule are as described above in the explanation of the operation plan storage unit 17.

Next, the operation of the traffic management device 10 will be described. The procedures described here do not need to be performed in order, and each process may be performed when the input data is updated, or may be performed periodically or at a fixed time. An overview of the three main processes will be described here, but the details will be described later with reference to Figures 34 to 36.

(First Process)
The learning unit 11 generates a learned model based on the information stored in the operation plan storage unit 17, the operation record storage unit 15, and the static operation section storage unit 16, and stores the model in the learning model storage unit 12.

(Second Processing)
The prediction unit 13 predicts where and how much deviation will occur in the operation route of the autonomous mobile device operated in the future, using the trained model stored in the learning model storage unit 12. The prediction unit 13 then determines the size of the operation section in which no collision or the like will occur even if the deviation occurs, and stores the result in the dynamic operation section storage unit 14 as a dynamic operation section.

(Third Process)
The operation plan generation unit 18 selects and connects dynamic operation sections so as to satisfy the operation conditions given by the user 4 as much as possible and not overlap with operation plans that have already been issued and are stored in the operation plan storage unit 17, and generates a route along which the autonomous mobile device 3 should operate. Here, the dynamic operation sections are stored in the dynamic operation section storage unit 14. Then, the operation plan generation unit 18 provides the generated route to the autonomous mobile device 3 and the user 4 as an operation plan.

The first process, i.e., the process in the learning unit 11, will be described with reference to FIG. 34. FIG. 34 is a flow diagram for explaining an example of the learning process in the operation management device 10. The process described here can be executed each time the operation record storage unit 15 is updated, or when a predetermined number of records have been accumulated in the operation record storage unit 15, or at regular or scheduled times, and the learned model is updated by this process.

First, the plan-performance difference extraction unit 11d compares the operation plan stored in the operation plan storage unit 17 with the operation performance stored in the operation performance storage unit 15 to calculate the deviation between the operation plan and the operation performance (step S31). Details of the process of calculating this deviation are as described above in the explanation of the plan-performance difference extraction unit 11d.

Furthermore, the plan/performance difference extraction unit 11d performs correspondence between the difference between the operation plan and the actual operation to determine in which operation section stored in the static operation section storage unit 16 the difference occurred (in which operation section the difference occurred) (step S32). If the operation results are based on an operation plan using a dynamic operation section, the static operation section storage unit 16 may be updated using the dynamic operation section, but the correspondence in step S32 may also be performed using a static operation section of a certain size.

The planned performance difference extraction unit 11d then stores the result in the planned performance difference storage unit 11a. Details of the matching process for determining which operational section stored in the static operational section storage unit 16 the deviation occurred in (which operational section the deviation occurred in) are as described above in the explanation of the planned performance difference extraction unit 11d.

Next, the pre-learning processing unit 11b processes and replaces data as pre-processing to enable machine learning to be performed in the subsequent main learning processing unit 11e on the information on the difference between the operation plan and the operation results stored in the plan-performance difference storage unit 11a (step S33). Details of the data processing and replacement as pre-learning processing are as described above in the explanation of the pre-learning processing unit 11b.

Next, the learning main processing unit 11e performs multiple regression analysis for each section, obtains a multiple regression equation for each section, and stores the obtained multiple regression equation and the coefficients of the equation in the learning model storage unit 12 (step S34). Details of the storage of the generated multiple regression equation and the coefficients of the equation are as described above in the explanation of the learning main processing unit 11e and the learning model storage unit 12.

The second process, i.e., the process in the prediction unit 13, will be described with reference to FIG. 35. FIG. 35 is a flow diagram for explaining an example of prediction processing in the operation management device 10. This process is executed when the learning model is updated by the learning unit 11, but it does not need to be executed in conjunction with the update process of the learned model as a series of processes, and is executed at a timing when it is acceptable to update the dynamic operation section. For example, it can be executed when formulating an operation plan for nighttime processing, regular execution, etc.

If this process is executed when an operation plan has already been issued but has not yet been operated, and then a new operation plan is generated, the underlying sections will change depending on the operation route. Therefore, when executing this process at such a time, the operation plan generation unit 18 needs to perform a determination process to respond to the change in the division of sections.

First, the planned performance difference prediction unit 13a predicts the amount of route deviation that occurs within the static operation section for each time period stored in the learning model storage unit 12 (step S41). A prediction is made for each time period. For details on the method for predicting the amount of route deviation, see the explanation of the planned performance difference prediction unit 13a. For details on moving to the destination by autonomous control, see the explanation of the autonomous mobile device 3 above.

Next, the dynamic operation section generation unit 13b determines the number of divisions (including the number of connections, prohibited use, and no connections) for the static operation section based on the deviation of the route for each time period calculated in step S41, and stores the number of divisions in the dynamic operation section storage unit 14 (step S42). Details of the method for determining the number of divisions for the static operation section are as described above in the explanation of the dynamic operation section generation unit 13b and the dynamic operation section storage unit 14.

The third process, i.e., the process in the operation plan generation unit 18, will be described with reference to FIG. 36. FIG. 36 is a flow diagram for explaining an example of the operation plan generation process in the operation management device 10. This process is executed in response to an application by the user 4.

First, the operation plan generation unit 18 receives the starting point, end point, and desired departure and arrival times of the operation from the user 4, generates an operation plan that avoids conflicts between autonomous mobile devices within the area and meets the desired departure and arrival times as much as possible, and stores the operation plan in the operation plan storage unit 17 (step S51). Details of the method for generating the operation plan are as described above in the explanation of the operation plan generation unit 18.

Next, the operation plan generating unit 18 issues the generated operation plan to the user 4 and the autonomous mobile device 3 (step S52). Details of the issuance of the operation plan are as described above in the explanation of the operation plan generating unit 18.

<Other application examples>
In this embodiment, an example of an operation plan in a three-dimensional space is mainly shown, but it can also be applied to an operation plan in a two-dimensional plane for an autonomous mobile device traveling on the ground. In that case, it is not necessary to use information on the upper limit altitude, the lower limit altitude, the altitude of the target arrival coordinates, and the actual arrival coordinates for the section. Even if it is applied to a traveling autonomous mobile device, when it is used in a situation with a hierarchical structure such as a multi-story parking lot, a large logistics center that vehicles can enter, or a factory, it may be expressed by floor height information in addition to expressing by altitude.

In addition, in this embodiment, an example is given in which the capacity per space is increased by using dynamic operation sections that divide static operation sections based on the learning results. However, this is not limited to this, and if the learning result shows that a positional deviation or time deviation occurs beyond the static operation section, conversely, instead of dividing it, one section may be made larger by connecting it with adjacent operation sections. This can be done by adding a column that holds multiple adjacent section IDs to the information on the dynamic operation section, and when the predicted deviation in the dynamic operation section generation unit 13b exceeds the size of the static operation section, the ID of the adjacent section is registered in the dynamic operation section storage unit 14. This has the effect of eliminating the need to assign an adjacent section to another drone when a section that is prone to deviation from the operation route and going outside the section has already been assigned, thereby avoiding collisions.

In addition, in this embodiment, the examples mainly use time zone numbers and season zone numbers as explanatory variables, but these items can be increased or replaced with other items. In other words, if the information is expected to correlate with the objective variable, not only can the data to be learned be obtained from the operation history alone, but items from other information sources can also be used. For example, when applied to an autonomous mobile device that operates on the ground, the ground surface condition type of each section may be added with a number (paved, unpaved, gravel, sand, etc.), and if applied to an autonomous mobile device that operates on the ocean, the wave height for each section may be added. Even in the case of an autonomous mobile device that operates in the sky, weather information for each section may be added. It is also possible to increase the number of items of information obtained from the operation history. For example, a sensor or the like may be installed in the autonomous mobile device, and information such as vibration information, temperature information, fluctuations in the output of the motor or engine used for propulsion, inclination, and fuel consumption status may be added to the objective variable. Through such various applications, it is possible to improve the prediction accuracy and build a trained model that is more in line with reality.

In addition, in this embodiment, the example in which the objective variable is the positional shift distance has been mainly given, but the positional shift direction can also be used as the objective variable. By using the positional shift direction in addition to the positional shift distance, it is possible to divide or combine only in a specific direction when dividing a section, rather than dividing all of the sections equally, which has the effect of further increasing the efficiency of space utilization.

In addition, in this embodiment, the time margin can be shortened or lengthened depending on the operation record. Such an example will be described with reference to Figs. 37 to 39. Figs. 37 to 39 are schematic diagrams for explaining other examples of prediction processing in the prediction unit of Fig. 22.

The left diagram in Figure 37 shows data indicating an operation plan, and the right diagram in Figure 37 shows data indicating operation results. Here, the data surrounded by dashed lines in the right diagram in Figure 37 indicates data between coordinates that are not in the operation plan. When data with coordinates that are not in the operation plan exists like this, it is necessary to take into account its presence in terms of time delays and advances. This is because a correspondence cannot be made due to the difference in the amount of data.

Therefore, when processing to find the difference between the operation plan and the operation results, first the number of data in the operation plan is increased by calculation to match the number of data in the operation results. For example, as shown in Figure 38, coordinates and time between two meshes (between mesh number 2 and mesh number 3 in this example) are calculated at equal intervals to match the increased number of data. This makes it possible to increase the amount of data in the same way as in the right diagram of Figure 39 (a diagram showing the operation results) as exemplified in the left diagram of Figure 39 (a diagram showing the operation plan), and find the advance or lag in time between the operation plan and the operation results.

By using this as the objective variable, it becomes possible to set the time period during which an autonomous mobile device occupies an operating section as a short period in accordance with past trends when planning operations, resulting in an even more efficient use of space.

<Other embodiments>
[a]
In each embodiment, the functions of the operation plan generation device, the functions of the operation management device, the functions of the autonomous mobile device, and the functions of the learning device have been described, but each device is not limited to the exemplified configuration examples, and it is sufficient if each device can realize these functions.

[b]
Each device according to each embodiment may have the following hardware configuration. Fig. 40 is a diagram showing an example of the hardware configuration of the device. The same applies to the other embodiment [a] above.

The device 100 shown in FIG. 40 may have a processor 101, a memory 102, and an interface 103. The processor 101 may be, for example, a microprocessor, an MPU (Micro Processor Unit), or a CPU. The processor 101 may include multiple processors. The memory 102 is, for example, configured by a combination of a volatile memory and a non-volatile memory. The functions of each device described in each embodiment are realized by the processor 101 reading and executing a program stored in the memory 102. At this time, input and output of information can be performed via an interface 103 such as a communication interface that communicates with other internal parts or other external devices.

In the above example, the program can be stored and supplied to the computer using various types of non-transitory computer readable media. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives) and magneto-optical recording media (e.g., magneto-optical disks). Further examples include CD-ROM (Read Only Memory), CD-R, and CD-R/W. Further examples include semiconductor memory (e.g., mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory)). The program may also be supplied to the computer by various types of transitory computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The temporary computer-readable medium can provide the program to the computer via a wired communication path, such as an electric wire or optical fiber, or via a wireless communication path.

[c]
Furthermore, just as the processing procedures in the operation plan generation device and the processing procedures in the learning device are exemplified in each of the above-mentioned embodiments, the present disclosure may also take the form of an operation plan generation method in an operation plan generation device or a learning method in a learning device. This operation plan generation method may include a generation step and a prediction step as described in FIG. 4, and other examples are as described in each of the above-mentioned embodiments. Furthermore, this learning method may include an input step and a learning step as described in FIG. 3, and other examples are as described in each of the above-mentioned embodiments.

In addition, a program related to an operation plan generation device (operation plan generation program) can be said to be a program for causing a computer (control computer) mounted on an operation plan generation device or functioning as an operation plan generation device to execute a generation step and a prediction step. A program related to a learning device (learning program) can be said to be a program for causing a computer (control computer) mounted on a learning device or functioning as a learning device to execute an input step and a learning step to generate a trained model.

Note that this disclosure is not limited to the above-described embodiments, and can be modified as appropriate without departing from the spirit and scope of the present disclosure. In addition, this disclosure can be implemented by combining the respective embodiments as appropriate.

A part or all of the above-described embodiments can be described as, but is not limited to, the following supplementary notes.
<Additional Notes>
(Appendix 1)
A generation unit that generates an operation plan indicating a route along which the autonomous mobile device is to be operated based on operation conditions input for the operation of the autonomous mobile device;
A prediction unit;
Equipped with
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The prediction unit predicts a difference between the operation plan and an operation record, which is an actual operation record based on the operation plan, for at least each section on a route included in the operation plan;
the generation unit generates an operation plan of the autonomous mobile device based on the operation conditions by using new sections updated according to the differences for each section predicted by the prediction unit.
Operation plan generation device.
(Appendix 2)
A section update unit that generates a new section according to the difference for each section predicted by the prediction unit and sets the new section for use in generating an operation plan.
2. An operation plan generating device according to claim 1.
(Appendix 3)
The difference includes a spatial difference between a section included in the operation schedule and a section that was actually operated at the same time,
the section update unit sets the new section such that a section having a small spatial difference among sections included in the operation schedule is smaller in at least one of space and time than a section having a large spatial difference.
3. An operation plan generating device according to claim 2.
(Appendix 4)
The difference includes a time difference between a scheduled time to pass through a section included in the operation schedule and a time to actually pass through the section,
the section update unit sets the new section such that a section having a small time difference among sections included in the operation schedule is smaller in at least one of space and time than a section having a large time difference among sections included in the operation schedule.
4. The operation plan generating device according to claim 2 or 3.
(Appendix 5)
the section update unit sets the new section so as to minimize at least one of spatial and temporal areas of a section for avoiding simultaneous or simultaneous coexistence with other autonomous mobile devices;
An operation plan generating device according to any one of appendix 2 to 4.
(Appendix 6)
The difference is at least one of a spatial difference, which is a spatial difference between a section included in the operation schedule and a section that was actually operated at the same time, and a temporal difference, which is a difference between a scheduled time to pass through the section included in the operation schedule and a time to actually pass through the section.
3. The operation plan generating device according to claim 1 or 2.
(Appendix 7)
the prediction unit predicts the difference for at least one of a type of autonomous mobile device, a model of autonomous mobile device, and an operating environment;
7. An operation plan generating device according to any one of claims 1 to 6.
(Appendix 8)
The prediction unit performs prediction of the difference using a trained model.
An operation plan generating device according to any one of appendix 1 to 7.
(Appendix 9)
A learning unit is provided that analyzes a difference between the operation plan and the operation record of the operation that is actually performed based on the operation plan for each section used for generating the operation plan, and generates a trained model that predicts the difference that may occur in a new operation plan based on the analysis result,
The prediction unit performs prediction of the difference using the trained model generated by the learning unit.
An operation plan generating device according to any one of appendix 1 to 7.
(Appendix 10)
The trained model is a model that inputs information including at least one of an operating time period, an operating season, and an operating environment, and outputs a predicted value of the difference.
10. The operation plan generating device according to claim 8 or 9.
(Appendix 11)
an input unit for inputting an operation plan indicating a route along which the autonomous mobile device is to be operated and an operation record indicating an actual operation performed based on the operation plan;
The learning department and
Equipped with
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The learning unit analyzes a difference between the operation plan and an operation record, which is an actual operation record based on the operation plan, for each section used to generate the operation plan, and generates a trained model that predicts the difference that may occur in a new operation plan based on the analysis result.
Learning device.
(Appendix 12)
A generation step of generating an operation plan indicating a planned route for operation of the autonomous mobile device based on operation conditions input for the autonomous mobile device;
A prediction step;
Equipped with
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The prediction step predicts a difference between the operation plan and an operation result, which is an actual operation result based on the operation plan, for at least each section on a route included in the operation plan;
the generation step uses new sections updated according to the differences for each section predicted in the prediction step to generate an operation plan for the autonomous mobile device based on the operation conditions.
How to generate operation plans.
(Appendix 13)
an input step of inputting an operation plan indicating a route along which the autonomous mobile device is to be operated and an operation record indicating an actual operation performed based on the operation plan;
The learning steps,
Equipped with
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The learning step includes analyzing a difference between the operation plan and an operation record, which is an actual operation record based on the operation plan, for each section used to generate the operation plan, and generating a trained model that predicts the difference that may occur in a new operation plan based on the analysis result.
How to learn.
(Appendix 14)
A computer is caused to execute a generating step of generating an operation plan indicating a route along which the autonomous mobile device is to be operated based on operation conditions input for the operation of the autonomous mobile device, and a predicting step;
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The prediction step predicts a difference between the operation plan and an operation result, which is an actual operation result based on the operation plan, for at least each section on a route included in the operation plan;
the generation step uses new sections updated according to the differences for each section predicted in the prediction step to generate an operation plan for the autonomous mobile device based on the operation conditions.
program.
(Appendix 15)
causing a computer to execute an input step of inputting an operation plan indicating a route along which the autonomous mobile device is to be operated and an operation record indicating an actual operation based on the operation plan, and a learning step;
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The learning step includes analyzing a difference between the operation plan and an operation record, which is an actual operation record based on the operation plan, for each section used to generate the operation plan, and generating a trained model that predicts the difference that may occur in a new operation plan based on the analysis result.
program.

1 Operation plan generation device 1a Generation unit 1b, 13 Prediction unit 2 Learning device 2a, 11 Learning unit 2b Input unit 3 Autonomous mobile device 4 User 12 Learning model storage unit 14 Dynamic operation section storage unit 15 Operation record storage unit 16 Static operation section storage unit 17 Operation plan storage unit 18 Operation plan generation unit 11a Plan performance difference storage unit 11b Pre-learning processing unit 11c Objective variable explanatory variable storage unit 11d Plan performance difference extraction unit 11e Main learning processing unit 13a Plan performance difference prediction unit 13b Dynamic operation section generation unit 100 Device 101 Processor 102 Memory 103 Interface

Claims (7)

A generation unit that generates an operation plan indicating a route along which the autonomous mobile device is to be operated based on operation conditions input for the operation of the autonomous mobile device;
A prediction unit;
Equipped with
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The prediction unit predicts a difference between the operation plan and an operation record, which is an actual operation record based on the operation plan, for at least each section on a route included in the operation plan;
the generation unit generates an operation plan of the autonomous mobile device based on the operation conditions by using new sections updated according to the differences for each section predicted by the prediction unit.
Operation plan generation device. A section update unit that generates a new section according to the difference for each section predicted by the prediction unit and sets the new section for use in generating an operation plan.
The operation plan generating device according to claim 1 . an input unit for inputting an operation plan indicating a route along which the autonomous mobile device is to be operated and an operation record indicating an actual operation performed based on the operation plan;
The learning department and
Equipped with
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The learning unit analyzes a difference between the operation plan and an operation record, which is an actual operation record based on the operation plan, for each section used to generate the operation plan, and generates a trained model that predicts the difference that may occur in a new operation plan based on the analysis result.
Learning device. A generation step of generating an operation plan indicating a planned route for operation of the autonomous mobile device based on operation conditions input for the autonomous mobile device;
A prediction step;
Equipped with
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The prediction step predicts a difference between the operation plan and an operation result, which is an actual operation result based on the operation plan, for at least each section on a route included in the operation plan;
the generation step uses new sections updated according to the differences for each section predicted in the prediction step to generate an operation plan for the autonomous mobile device based on the operation conditions.
How to generate operation plans. an input step of inputting an operation plan indicating a route along which the autonomous mobile device is to be operated and an operation record indicating an actual operation performed based on the operation plan;
The learning steps,
Equipped with
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The learning step includes analyzing a difference between the operation plan and an operation record, which is an actual operation record based on the operation plan, for each section used to generate the operation plan, and generating a trained model that predicts the difference that may occur in a new operation plan based on the analysis result.
How to learn. A computer is caused to execute a generating step of generating an operation plan indicating a route along which the autonomous mobile device is to be operated based on operation conditions input for the operation of the autonomous mobile device, and a predicting step;
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The prediction step predicts a difference between the operation plan and an operation result, which is an actual operation result based on the operation plan, for at least each section on a route included in the operation plan;
the generation step uses new sections updated according to the differences for each section predicted in the prediction step to generate an operation plan for the autonomous mobile device based on the operation conditions.
program. causing a computer to execute an input step of inputting an operation plan indicating a route along which the autonomous mobile device is to be operated and an operation record indicating an actual operation performed based on the operation plan, and a learning step;
the route includes information expressed as a succession of sections that are areas for avoiding simultaneous or simultaneous co-existence with other autonomous mobile devices;
The learning step includes analyzing a difference between the operation plan and an operation record, which is an actual operation record based on the operation plan, for each section used to generate the operation plan, and generating a trained model that predicts the difference that may occur in a new operation plan based on the analysis result.
program.

Images