Group control system and group control method

Abstract

To provide a group control system and a group control method that are able to appropriately control a plurality of moving bodies.SOLUTION: A group control system 10 includes: an acceleration data acquisition unit that acquires acceleration data of each moving body 200; a position estimation unit 102 that estimates a current position of each moving body 200; and a moving-body control unit 103 that repeatedly performs a process of inputting first state information including the acceleration data and the estimated position of each moving body 200 and second state information indicating a future state of each moving body 200, to a deep layer reinforcement learning model M1 and controlling an action of each moving body 200, based on output from the deep layer reinforcement learning model M1. The second state information input to the deep layer reinforcement learning model M1 indicates a future state of each moving object 200 predicted using an output obtained by inputting seed information to a hostile generation network M2.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a group control system and a group control method. [Background technology]

[0002] In recent years, the development of systems using multiple moving objects such as autonomously moving robots has progressed. Accordingly, various technologies for group control of multiple moving objects have been proposed. For example, Patent Document 1 discloses a group control system that can avoid collisions between moving objects. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2017-142778 A Summary of the Invention [Problem to be solved by the invention]

[0004] For example, when the error between the position of each moving object grasped by the device that controls the group of moving objects and the actual position of each moving object increases, the moving objects may not be controlled appropriately. In this context, there is a demand for a technology that can appropriately control multiple moving objects.

[0005] The present disclosure has been made in light of the above-mentioned circumstances, and has an object to provide a group control system and a group control method that are capable of appropriately controlling a plurality of moving objects. [Means for solving the problem]

[0006] One aspect of the present disclosure for achieving the above-mentioned object is a group control system for controlling a plurality of moving bodies capable of autonomously traveling within a specified area, comprising: an acceleration data acquisition unit that acquires, for each of the moving bodies, acceleration data detected by a sensor provided on the moving bodies; a position estimation unit that estimates a current position of each of the moving bodies; and a moving body control unit that inputs first state information including the acceleration data and the estimated position of each of the moving bodies and second state information representing a future state of each of the moving bodies into a deep reinforcement learning model, and repeatedly executes a process of controlling the behavior of each of the moving bodies based on the output from the deep reinforcement learning model, wherein the second state information input to the deep reinforcement learning model represents a future state of each of the moving bodies predicted using an output obtained by inputting seed information into a generative adversarial network, and the seed information includes information output from the deep reinforcement learning model in the previous processing.

[0007] In one aspect of the above, the deep reinforcement learning model may be a model in which deep reinforcement learning is performed in an environment configured so that a reward is awarded based on the difference between the position of each moving body estimated by the position estimation unit and the actual position of each moving body.

[0008] In the above aspect, the second state information may include position and acceleration data of each of the moving objects at a future time point.

[0009] In one of the above aspects, the generative adversarial network may output a plurality of images showing the positions of each of the moving bodies at different points in time in the future, and the second state information may include the acceleration data for a future point in time calculated from the positions of each of the moving bodies at different points in time shown in the plurality of images.

[0010] Another aspect of the present disclosure for achieving the above-mentioned object is a group control method for controlling a plurality of moving bodies capable of autonomously traveling within a specified area, comprising: acquiring acceleration data detected by a sensor provided on each of the moving bodies; estimating a current position of each of the moving bodies; inputting first state information including the acceleration data and the estimated position of each of the moving bodies and second state information representing a future state of each of the moving bodies into a deep reinforcement learning model; and repeatedly executing a process of controlling the behavior of each of the moving bodies based on an output from the deep reinforcement learning model; the second state information input to the deep reinforcement learning model represents a future state of each of the moving bodies predicted using an output obtained by inputting seed information into a generative adversarial network; and the seed information includes information output from the deep reinforcement learning model in the previous processing. Effect of the Invention

[0011] According to the present disclosure, it is possible to provide a group control system and a group control method that can appropriately control a plurality of moving objects. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram showing an example of a configuration of a group control system according to an embodiment; [Diagram 2] FIG. 13 is a schematic diagram showing an example of the configuration of a machine learning model used by a mobile object control unit. [Diagram 3] 10 is a flowchart showing an example of an operation flow of the information processing device. [Figure 4] FIG. 2 is a block diagram showing an example of a hardware configuration of an information processing device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Hereinafter, an embodiment will be described with reference to the drawings. Fig. 1 is a block diagram showing an example of a configuration of a group control system 10 according to an embodiment. As shown in Fig. 1, the group control system 10 includes a plurality of moving objects 200 and an information processing device 100 that manages the plurality of moving objects 200. The information processing device 100 and each moving object 200 are connected to each other so as to be able to communicate wirelessly.

[0014] The moving body 200 is an autonomously movable device, and is also called an autonomous mobile robot. For example, the moving body 200 may be an AMR (Autonomous Mobile Robot) and transport luggage and the like. The moving body 200 executes various tasks involving movement. For example, the executed task may be a transport task of moving from one point to another point to transport luggage, or a charging task of moving to a point where a charging facility is located to charge a rechargeable battery for driving the moving body 200. In this embodiment, as shown in FIG. 1, each moving body 200 includes, in particular, an IMU (Inertial Measurement Unit) 201, a LiDAR (Light Detection And Ranging) sensor 202, a position detection unit 203, a travel control unit 204, and a communication processing unit 205.

[0015] The IMU 201 is a sensor that detects the acceleration of the moving body 200, and detects the acceleration in each direction of three orthogonal axes (X-axis, Y-axis, and Z-axis) and the angular acceleration around each of these three axes.

[0016] The LiDAR sensor 202 is a sensor that uses laser light to measure the shape and distance of objects around the moving body 200.

[0017] The position detection unit 203 detects the position of the moving body 200 by comparing the measurement data of the LiDAR sensor 202 with map data of the moving environment of the moving body 200. More specifically, the position detection unit 203 detects the position of the moving body 200 at the time of measurement by a sensor for position detection (the LiDAR sensor 202).

[0018] The travel control unit 204 controls the autonomous travel of the mobile body 200. The travel control unit 204 controls the travel of the mobile body 200 based on the position of the mobile body 200 detected by the position detection unit 203 and the travel route to the destination point for executing the task, which is assigned by the information processing device 100. That is, the travel control unit 204 controls the travel of the mobile body 200 so that the mobile body 200 moves to the destination point. Note that when the travel route assigned by the information processing device 100 is a set of a series of waypoints to the destination point, that is, when only the discrete waypoints are notified to the mobile body 200 as the travel route and the travel route to each waypoint is not notified, the travel control unit 204 may create a specific travel route to the next waypoint to be visited and cause the mobile body 200 to travel along the created travel route.

[0019] The communication processing unit 205 performs processing for transmitting and receiving information with the information processing device 100. In this embodiment, the communication processing unit 205 transmits the position of the moving object 200 detected by the position detection unit 203 to the information processing device 100. The communication processing unit 205 also transmits acceleration data detected by the IMU 201 to the information processing device 100. That is, the communication processing unit 205 transmits the acceleration data of the moving object 200 to the information processing device 100. Note that the transmission of the position of the moving object 200 detected by the position detection unit 203 and the acceleration data detected by the IMU 201 to the information processing device 100 is performed periodically.

[0020] Next, the information processing device 100 will be described. The information processing device 100 is a device that controls a plurality of moving objects 200 that can travel autonomously within a predetermined area. In particular, the information processing device 100 assigns a task to the moving object 200 that exists within the predetermined area, and notifies the moving object 200 of a moving route for executing the task. As shown in FIG. 1, in this embodiment, the information processing device 100 includes, in particular, a moving object information acquisition unit 101, a position estimation unit 102, a moving object control unit 103, a model storage unit 104, and a model learning unit 105. Note that, in the configuration shown in FIG. 1, the information processing device 100 includes the model learning unit 105, but the model learning unit 105 may be realized by another device. That is, a learning process to be described later may be executed by a device other than the information processing device 100.

[0021] The mobile object information acquisition unit 101 acquires information from each mobile object 200. Specifically, the mobile object information acquisition unit 101 acquires information transmitted by the communication processing unit 205 of the mobile object 200. In this embodiment, the mobile object information acquisition unit 101 acquires the position of the mobile object 200 detected by the position detection unit 203. Focusing on the function of acquiring the position of the mobile object 200, the mobile object information acquisition unit 101 may be referred to as a position acquisition unit. Furthermore, the mobile object information acquisition unit 101 acquires acceleration data of each mobile object 200. Focusing on the function of acquiring acceleration data, the mobile object information acquisition unit 101 may be referred to as an acceleration data acquisition unit.

[0022] The position estimation unit 102 estimates the current position of each moving object 200. The moving object information acquisition unit 101 acquires the position information of the moving object 200, but the position indicated by the acquired position information is not necessarily the current position of the moving object 200. In other words, if the position acquired by the moving object information acquisition unit 101 is the position of the moving object 200 detected at a time point preceding the current time point, the current position of the moving object 200 differs from the position acquired by the moving object information acquisition unit 101 due to the movement of the moving object 200 during the period from the position detection time point to the current time point.

[0023] In this embodiment, the position estimation unit 102 estimates the current position of each moving body 200 based on a moving speed planned in advance for the moving body 200 and the position of the moving body 200 at the time of position detection notified from the moving body 200. The moving speed planned in advance (hereinafter referred to as the planned moving speed) is a constant speed, which may be different for each moving body 200, or may be a common speed for each moving body 200. In this embodiment, the moving body 200 accelerates until it reaches the planned moving speed, and when the speed of the moving body 200 reaches the planned moving speed, it moves while maintaining the planned moving speed. More specifically, the position estimation unit 102 estimates the current position of the moving body 200 based on the position of the moving body 200 at the time of position detection, assuming that the moving body 200 has moved at the planned moving speed along the moving route assigned to the moving body 200.

[0024] However, when the moving body 200 moves from point A to another point B, it cannot move at the planned moving speed for the entire section from point A to point B. That is, acceleration is required near point A, and the moving speed of the moving body 200 at that time is equal to or lower than the planned moving speed. Similarly, deceleration is required near point B, and the moving speed of the moving body 200 at that time is also equal to or lower than the planned moving speed. For this reason, the position estimated by the position estimation unit 102 may deviate from the actual position of the moving body 200. The possibility of such a deviation occurring increases as the moving environment becomes smaller. This is because the traveling section in which the planned moving speed can be maintained becomes shorter. In this embodiment, as described later, the behavior (movement) of the moving body 200 is controlled by the moving body control unit 103 so that the moving body 200 acts (moves) so as to suppress the deviation between the position estimated by the position estimation unit 102 and the actual position of the moving body 200. As a result, even if the moving environment is a small area, it is possible to suppress the difference between the position grasped by the information processing device 100 and the actual position of each moving body 200, and to appropriately manage the positions of the multiple moving bodies 200. In other words, it is possible to appropriately operate the moving bodies 200 in a small area.

[0025] The mobile object control unit 103 controls the behavior of each mobile object 200. Specifically, the mobile object control unit 103 assigns tasks to the mobile object 200 and assigns a movement route for executing the task. The mobile object control unit 103 then notifies each mobile object 200 of the assigned task and the movement route. Note that the mobile object control unit 103 may notify only the movement route. Note that in this embodiment, the movement route is composed of a series of waypoints to the destination point.

[0026] The mobile object control unit 103 assigns various tasks occurring in the environment to the mobile object 200, and also assigns a movement route for executing the task to the mobile object 200. Any of a plurality of predetermined tasks is assigned to the mobile object 200. Also, any of a plurality of predetermined movement routes is assigned to the mobile object 200. In this embodiment, the mobile object control unit 103 first assigns a task occurring in the environment to the mobile object 200 based on the current position of each mobile object 200, etc. Then, the mobile object control unit 103 adjusts the task and the movement route assigned to each mobile object 200 using a learned machine learning model stored in the model storage unit 104. That is, the mobile object control unit 103 controls the behavior of each mobile object 200 using the machine learning model. Specifically, the mobile object control unit 103 repeatedly executes the following process. This process is a process of inputting the first state information and the second state information into a deep reinforcement learning model, and controlling the behavior of each mobile object 200 based on the output from the deep reinforcement learning model. Here, the first state information is information including the acceleration data of each moving object 200 acquired by the moving object information acquisition unit 101 and the position of each moving object 200 estimated by the position estimation unit 102. The acceleration data in the first state information is specifically the most recently acquired acceleration data (latest acceleration data) and may further include acceleration data acquired before that. Similarly, the position in the first state information is specifically the most recently estimated position (latest estimated position) and may further include a position estimated before that. In addition, the second state information is information representing a future state of each moving object 200 predicted using the output of the generative adversarial network. Here, the future state may be, for example, a state several minutes ahead, or preferably, a state several seconds ahead. In this embodiment, the second state information includes the position and acceleration data of each moving object 200 at a future time point. Note that the second state information may be only the position of each moving object 200 at a future time point, or may include other information (for example, the speed of each moving object 200 at a future time point).

[0027] FIG. 2 is a schematic diagram showing an example of the configuration of the machine learning model M used by the mobile object control unit 103. The mobile object control unit 103 repeats a process of acquiring the output of the deep reinforcement learning model M1 in the machine learning model M over time. Then, every time the mobile object control unit 103 acquires the output of the deep reinforcement learning model M1, it controls the behavior of each mobile object 200 based on this output. The deep reinforcement learning model M1 is a Deep Q-Network composed of an input layer, a hidden layer, and an output layer. As described above, the first state information and the second state information for each mobile object 200 are input to the input layer. However, when the second state information is not obtained, dummy information is input as the second state information. The output of the deep reinforcement learning model M1 is a Q value of each action in a certain state (i.e., a state input to the deep reinforcement learning model M1). In this embodiment, the first state information includes the acceleration data of each moving body 200 acquired by the moving body information acquisition unit 101 and the position of each moving body 200 estimated by the position estimation unit 102, as well as the task currently assigned to each moving body 200 and the moving route currently assigned to each moving body 200 for the execution of the task. In this embodiment, the action associated with the Q value is a set of the task to be assigned to each moving body 200 and the moving route to be assigned to each moving body 200. That is, each node of the output layer of the deep reinforcement learning model M1 corresponds to the action of each moving body 200 (the task and moving route of each moving body 200). Therefore, the task and the moving route recommended for each moving body 200 are obtained from the output of the deep reinforcement learning model M1. The moving body control unit 103 changes the task and the moving route to be assigned to each moving body 200 according to the output result of the deep reinforcement learning model M1. Then, the moving body control unit 103 notifies each moving body 200 of the new assignment.

[0028] As shown in FIG. 2, the machine learning model M is a model composed of a deep reinforcement learning model M1 and a generative adversarial network M2. In particular, the output of the deep reinforcement learning model M1 is used as seed information input to the generative adversarial network M2. In addition, the output of the generative adversarial network M2 is used as the above-mentioned second state information input to the deep reinforcement learning model M1. That is, the second state information input to the deep reinforcement learning model M1 represents a future state of each moving object 200 predicted using an output obtained by inputting the seed information to the generative adversarial network M2. Here, the seed information includes the first state information (e.g., the position of each moving object 200) input to the deep reinforcement learning model M1 in the previous process in the repeated process and the information output from the deep reinforcement learning model M1 in the previous process. That is, the seed information is information including the state of each moving object 200 at a certain time point and the action recommended at the certain time point. Here, the state of each moving body 200 at a certain point in time used as seed information includes at least the position of each moving body 200, and may further include acceleration. In this way, the second state information input to the deep reinforcement learning model M1 in the n-th (n is an integer equal to or greater than 2) iteration of the repeated processing is information obtained by inputting the information output from the deep reinforcement learning model M1 in the n-1th processing to the generative adversarial network M2.

[0029] Based on this seed information, the generative adversarial network M2 outputs an image showing the position of each moving body 200 at a future time point. More specifically, the seed information is input to a generator of the generative adversarial network M2, and an image is output from the generator. The output image is a map of a predetermined area (i.e., an area in which each moving body 200 moves), and is an image of the map on which the positions of each moving body 200 are plotted. In this embodiment, in order to obtain acceleration data of each moving body 200 at a future time point, the generative adversarial network M2 outputs a plurality of images showing the positions of each moving body 200 at different future time points. In this way, if time-series data of the future positions of each moving body 200 is obtained, the speed data and acceleration data of each moving body 200 can be easily calculated. Therefore, the moving body control unit 103 may predict not only the positions of each moving body 200 but also the acceleration of each moving body 200 as a future state using the output of the generative adversarial network M2. Thus, in this embodiment, the second state information input to the deep reinforcement learning model M1 includes acceleration data at a future time point calculated from the positions of each moving body 200 at different times shown in the multiple images output by the generative adversarial network M2. Note that when the acceleration data of each moving body 200 at a future time point is not used as the second state information, in other words, when only the positions of each moving body 200 at a future time point is used as the second state information, the generative adversarial network M2 may generate a single image showing the positions of each moving body 200 at one future time point.

[0030] The above-mentioned machine learning model M is generated in advance by the model learning unit 105 and stored in the model storage unit 104. The model learning unit 105 performs a process of generating the machine learning model M. That is, the model learning unit 105 performs a process of the learning phase of the machine learning model M.

[0031] The model learning unit 105 first uses only the first state information as an input to learn the deep reinforcement learning model M1 (hereinafter referred to as a first learning process). Specifically, the model learning unit 105 performs deep reinforcement learning of the deep reinforcement learning model M1 using an environment in which a reward is set so that a reward is given based on the difference between the position of each moving body 200 estimated by the position estimation unit 102 and the actual position of each moving body 200. That is, the model learning unit 105 updates the parameters of the deep reinforcement learning model M1 by repeating the following process. This process is a process of inputting the first state information obtained when each moving body 200 is made to act in an environment in which the reward is set as described above to the deep reinforcement learning model M1, and making each moving body 200 perform an action (i.e., movement according to an assigned task and movement route) selected based on the output obtained from the deep reinforcement learning model M1. As a result, the deep reinforcement learning model M1 is trained so as to reduce the deviation between the position of each moving body 200 estimated by the position estimation unit 102 and the actual position of each moving body 200. Note that, in setting the reward, for example, the smaller the difference between the position of each moving body 200 estimated by the position estimation unit 102 and the actual position of each moving body 200, the larger the reward that can be obtained. Note that, as the difference referred to here, for example, the average of the difference in the position for each moving body 200 may be used, or the sum of the difference in the position for each moving body 200 may be used.

[0032] Next, the model learning unit 105 trains the generative adversarial network M2 so as to generate the above-mentioned second state information (hereinafter, referred to as the second learning process). Specifically, the model learning unit 105 trains the generative adversarial network M2 by comparing the image generated by the generator with an image as a real sample by a discriminator. Here, in this embodiment, as the real sample, time series data of the position of each moving body 200 actually observed, that is, a set of images showing the position of each moving body 200 at each time, is used. For example, the model learning unit 105 may perform the second learning process by using the time series data of the position of each moving body 200 observed during the first learning process as the real sample. In addition, the model learning unit 105 may use the time series data of the seed information obtained during the first learning process in order to generate an image to be output to the discriminator by the generator. In this way, the generative adversarial network M2 that generates an image showing the future position of each moving body 200 is trained.

[0033] After the generative adversarial network M2 has been trained, the model training unit 105 retrains the deep reinforcement learning model M1 that has been subjected to the first training process (hereinafter referred to as the third training process). That is, after it becomes possible to generate the second state information, the model training unit 105 performs the third training process. In this training process, the model training unit 105 trains the deep reinforcement learning model M1 using the first state information and the second state information as inputs. This training process is the same as the first training process, except that the second state information is added as an input to the deep reinforcement learning model M1. Note that, in order to generate the second state information, the output of the generative adversarial network M2 that has completed the second training process is used. In the third training process, the second state information input to the deep reinforcement learning model M1 in the n-th repeated process is information obtained by inputting the information output from the deep reinforcement learning model M1 in the n-1th process to the generative adversarial network M2. When the second state information is not obtained (first processing), dummy information is input as the second state information. This causes the deep reinforcement learning model M1 to learn so that the deviation between the position of each moving body 200 estimated by the position estimation unit 102 and the actual position of each moving body 200 becomes even smaller. This completes the learning of the machine learning model M. The model learning unit 105 stores the learned machine learning model M in the model storage unit 104.

[0034] Next, the flow of operations of the above-mentioned information processing device 100 will be described with reference to the flowchart shown in FIG. 3. First, in step S10, the deep reinforcement learning model M1 is trained (first learning process) without using a future state (second state information). Next, in step S11, the generative adversarial network M2 is trained (second learning process). Next, in step S12, the deep reinforcement learning model M1 is trained (third learning process) using the future state as well. Through the above processes, the learning of the machine learning model M is completed. After that, in step S13, the group control system 10 is operated using the trained machine learning model M.

[0035] Next, an example of a hardware configuration of the information processing device 100 will be described with reference to Fig. 4. The information processing device 100 includes a network interface 150, a memory 151, and a processor 152.

[0036] The network interface 150 is used to communicate with other devices, such as the mobile unit 200. The network interface 150 may include, for example, a network interface card (NIC).

[0037] The memory 151 is configured, for example, by a combination of a volatile memory and a non-volatile memory. The memory 151 is used to store programs executed by the processor 152, data used in processing, and the like. The memory 151 may include multiple memories. In addition, the above-mentioned model storage unit 104 may be realized by a storage device such as the memory 151.

[0038] The processor 152 reads out a program from the memory 151 and executes it to perform processing of the information processing device 100. The processing of each component shown in Fig. 2 is performed, for example, by the processor 152 reading out a program from the memory 151 and executing it. The processor 152 may be, for example, a microprocessor, an MPU (Micro Processor Unit), a CPU (Central Processing Unit), or a GPU (Graphics Processing Unit). The processor 152 may include multiple processors. In this way, the information processing device 100 has the functions of a computer.

[0039] Each moving object 200 may also have the configuration shown in Fig. 4. That is, the moving object 200 may also have a function as a computer. In this case, the processing of the moving object 200 may be performed by a processor reading a program from a memory and executing the program.

[0040] The embodiment has been described above. According to the group control system 10, the behavior of each moving body 200 is controlled based on the first state information including the current acceleration data and position of each moving body 200 and the second state information representing the predicted future state of each moving body. Therefore, since the behavior is controlled taking into account various state information of each moving body 200, the behavior of the moving body 200 can be controlled more appropriately than when these are not taken into account. Therefore, the behavior of the multiple moving bodies 200 can be appropriately controlled. In particular, the behavior of each moving body 200 is controlled using a model in which deep reinforcement learning is performed so that the difference between the position grasped by the device performing group control and the actual position of each moving body 200 is small. By making each moving body 200 behave according to such a model, it is possible to suppress the deviation between the position grasped by the device performing group control (information processing device 100) and the actual position of each moving body 200, and the positions of the multiple moving bodies 200 can be appropriately managed. As a result, the moving body 200 can be appropriately operated even in a small area.

[0041] The present invention is not limited to the above-described embodiment, and can be modified as appropriate without departing from the spirit and scope of the present invention. For example, at least a part of the configuration of the information processing device 100 (e.g., the model learning unit 105, etc.) may be mounted on a device other than the information processing device 100. [Explanation of symbols]

[0042] 10 Group Control System 100 Information processing device 101 Mobile information acquisition unit 102 Position estimation part 103 Mobile control unit 104 Model Memory Unit 105 Model Learning Department 150 Network Interface 151 Memory 152 processors 200 Mobile 201 IMU 202 LiDAR Sensor 203 Position detection unit 204 Driving control unit 205 Communication processing unit M Machine Learning Model M1 deep reinforcement learning model M2 Generative Adversarial Network

Claims

1. A group control system that controls a plurality of moving objects capable of autonomously traveling within a predetermined area, an acceleration data acquisition unit that acquires, for each of the moving objects, acceleration data detected by a sensor provided in the moving object; a position estimation unit that estimates a current position of each of the moving objects; a mobile object control unit that inputs first state information, including the acceleration data and the estimated position of each of the mobile objects, and second state information representing a future state of each of the mobile objects, into a deep reinforcement learning model, and repeatedly executes a process of controlling the behavior of each of the mobile objects based on an output from the deep reinforcement learning model; having The second state information input to the deep reinforcement learning model represents a future state of each of the moving objects predicted using an output obtained by inputting seed information to a generative adversarial network; The seed information includes information output from the deep reinforcement learning model in the previous process. Swarm control system.

2. The deep reinforcement learning model is a model in which deep reinforcement learning is performed in an environment set so that a reward is given based on the difference between the position of each moving object estimated by the position estimation unit and the actual position of each moving object. The group control system of claim 1 .

3. The second state information includes a position and acceleration data of each of the moving objects at a future time point. A group control system according to claim 1 or 2.

4. The generative adversarial network outputs a plurality of images showing the position of each of the moving objects at different times in the future; The second state information includes the acceleration data at a future time point calculated from the positions of the moving objects at different times shown in the plurality of images. The group control system of claim 3.

5. A group control method for controlling a plurality of moving objects capable of autonomously traveling within a predetermined area, comprising the steps of: acquiring, for each of the moving objects, acceleration data detected by a sensor provided in the moving object; Estimating a current location of each of the moving objects; inputting first state information including the acceleration data and the estimated position of each of the moving bodies and second state information representing a future state of each of the moving bodies into a deep reinforcement learning model, and repeatedly executing a process of controlling the behavior of each of the moving bodies based on an output from the deep reinforcement learning model; The second state information input to the deep reinforcement learning model represents a future state of each of the moving objects predicted using an output obtained by inputting seed information to a generative adversarial network; The seed information includes information output from the deep reinforcement learning model in the previous process. Group control methods.

Images