Group control system and method

Abstract

To provide a group control system and a group control method that allow machine learning by taking acceleration and deceleration data into account and group controlling based on correct position information.SOLUTION: A group control system controls a plurality of moving objects 10 that are capable of traveling autonomously within a predetermined area. The group control system includes a position information estimation unit 101 that estimates position information of each moving object, a route planning unit 102 that creates a route plan for each moving object based on the estimated position information, an acceleration data acquisition unit that acquires acceleration data acquired from an acceleration sensor 11, and a moving object position acquisition unit that acquires an actual position of each moving object using a position sensor. A deep reinforcement learning model is trained so as to correct the amount of deviation between the actual position of the moving body obtained using the position sensor and the estimated position of the moving body based on the obtained acceleration data.SELECTED DRAWING: Figure 4

Description

Technical Field

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

Background Art

[0002] In a group control system of a moving body that issues control commands to a plurality of moving bodies capable of autonomously traveling within a predetermined area, the position information of each moving body is estimated, and a path course for each moving body is created based on the estimated position information so that the moving bodies do not interfere with each other (Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the above prior art, the acceleration and deceleration of the moving body are not considered. Therefore, due to the acceleration and deceleration of the moving body, a positional deviation occurs between the estimated position of the moving body and the actual position of the moving body due to the motion behavior of the moving body, and group control based on correct position information cannot be performed. When introducing a plurality of AMRs (Autonomous Mobile Robots) in a narrow place where the influence of acceleration and deceleration is large, there is a risk of contact with each other, and it may be necessary to change the layout, which may result in an increase in cost.

[0005] The present disclosure has been made to solve such problems, and an object thereof is to provide a group control system and a group control method capable of performing machine learning considering acceleration and deceleration data and performing group control based on correct position information.

Means for Solving the Problems

[0006] A group control system according to an aspect of the present disclosure is In a swarm control system that controls multiple autonomously moving objects, At least one memory that stores instructions, At least one control unit configured to execute the aforementioned instruction, Acceleration sensors and position sensors attached to each moving object, Equipped with, The control unit, A position information estimation unit that estimates the position information of each moving object, A route planning unit that creates a route plan for each moving object based on the estimated location information, An acceleration data acquisition unit that acquires acceleration data from the acceleration sensor, A mobile body position acquisition unit that acquires the actual position of each mobile body using the position sensor, Equipped with, The deep reinforcement learning model is configured to train itself to correct the difference between the actual position of the moving body obtained using the position sensor and the estimated position of the moving body, based on the acquired acceleration data.

[0007] A group control method according to one aspect of this disclosure is: A group control method for controlling multiple autonomously moving objects, Estimating the positional information of each moving object, Based on the estimated location information, a path plan for each moving object is created, To acquire acceleration data from an acceleration sensor, Obtaining the actual position of each moving object using position sensors, Includes, The deep reinforcement learning model is trained to correct the difference between the actual position of the moving body obtained using the position sensor and the estimated position of the moving body, based on the acquired acceleration data. [Effects of the Invention]

[0008] According to this disclosure, it is possible to provide a group control system and a group control method that can perform group control based on correct position information, taking into account acceleration and deceleration data. [Brief explanation of the drawing]

[0009] [Figure 1] This diagram illustrates the AMR's actual test run. [Figure 2] This shows the AMR's location as recognized by the group control system and its actual location. [Figure 3] This graph shows the relationship between speed and time in several cases. [Figure 4] This is a schematic diagram showing the configuration of a group control system according to an embodiment. [Figure 5] This is a flowchart showing a group control method according to an embodiment. [Figure 6] This is a block diagram illustrating an example of the hardware configuration of a group control system. [Modes for carrying out the invention]

[0010] Specific embodiments of the present invention will be described in detail below with reference to the drawings. However, the present invention is not limited to the following embodiments. Also, for clarity of explanation, the following description and drawings have been simplified as appropriate.

[0011] In factories and other industrial settings, systems are being developed in which multiple Autonomous Mobile Robots (AMRs) autonomously navigate and transport goods to support and replace human work. Each AMR can basically estimate its own position individually and autonomously navigate within a predetermined area according to a path plan. However, as shown in Figure 1, when actual AMRs are tested, it has been found that when many AMRs are operating in a smaller area, they may come into contact with each other. Therefore, a group control system for multiple AMRs is being introduced to prevent collisions or interference between them.

[0012] Figure 2 shows the position (RP_10) of the AMR recognized by the group control system and the actual position (AP_10) of the AMR. The position of the AMR recognized by the group control system is the position where the AMR is estimated to be at a certain time when a destination (goal) is set for the AMR and the AMR is moved according to the route plan to the goal. On the other hand, the actual position of the AMR is the actual position of the AMR determined using a position sensor (for example, LiDAR).

[0013] As shown in Figure 2, the position (RP_10) of the AMR recognized by the group control system and the actual position (AP_10) of the AMR are slightly deviated. This may be considered because the influence of acceleration becomes significant when moving the AMR in a narrow area. Similar position deviations also occurred when conducting the same test on actual machines using various AMRs with different shapes, sizes, etc.

[0014] Figure 3 is a graph showing the relationship between speed and time in the traveling cases of several AMRs. Since acceleration is the rate of change of speed per unit time, Figure 3 also shows acceleration. In the case of A1, it shows the state where the moving object accelerates from a stopped state, then moves at a constant speed, and then decelerates and stops. In a relatively wide area, the moving object accelerates and decelerates as shown in A1. In the case of A2, it shows the state where the moving object accelerates from a stopped state and then decelerates and stops before reaching a constant speed state. The deviation between the position of the AMR recognized by the group control system and the actual position of the AMR described above is more significant in the case of A2 where acceleration and deceleration are required in a narrow area. Also, in the case of A1, position deviation may occur due to the influence of acceleration and deceleration when the constant speed state is relatively short. Generally, in a narrow layout, the position deviation due to the influence of acceleration and deceleration appears more significantly than in a wide layout.

[0015] Therefore, in this embodiment, a group control system and a group control method that can perform group control based on correct position information by considering the acceleration of the moving object are provided. More specifically, a deep reinforcement learning model is used to correct the position deviation of the AMR based on the obtained acceleration of the AMR.

[0016] Figure 4 is a schematic diagram showing the configuration of the group control system according to this embodiment. Each AMR managed by the group control system is equipped with an acceleration sensor (e.g., a gyroscope, an IMU (Inertial Measurement Unit)). Acceleration and deceleration data are acquired at each time point from the acceleration sensor on each AMR. The acquired acceleration and deceleration data is input to the input layer 201 of the deep reinforcement learning model 20. The input layer 201 is also called the acceleration data acquisition unit. In the example in Figure 3, for example, acceleration from a stationary state to acceleration, acceleration at a constant speed, and deceleration from deceleration to a stationary state may be input at predetermined intervals. The input layer 201 of the deep reinforcement learning model 20 also receives input for the position of the AMR recognized by the group control system and the actual position of the AMR. In this example, a Deep Q-Network is used as the deep reinforcement learning model, but it is not limited to this, and various neural network models that are understandable to those skilled in the art can be used.

[0017] The deep reinforcement learning model 20 learns to correct the discrepancy between the actual position of the AMR and the position of the AMR recognized by the swarm control system (i.e., to make the discrepancy zero). In addition, the deep reinforcement learning model learns, as a reward, to ensure that the actual positions of multiple AMRs do not overlap (i.e., that the AMRs do not come into contact with each other). In this way, the deep reinforcement learning model is trained based on a predetermined amount of input data, with several rewards provided, and a trained model is generated.

[0018] An autonomous driving control device 100 is provided for each mobile body 10. The autonomous driving control device 100 consists of a computer equipped with a processor such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) and memory. The autonomous driving control device 100 controls the autonomous driving of the mobile body 10 to a designated destination. Each autonomous driving control device 100 includes a self-position estimation unit 101 and a route planning unit 102. The self-position estimation unit 101 estimates the position of the mobile body at each time as it moves to the destination according to the route plan. The route planning unit 102 sets a destination for each mobile body and creates a route plan from the current position to the destination. In Figure 4, two autonomous driving control devices 100 are provided, corresponding to two mobile bodies, but there are also cases where three or more autonomous driving control devices 100 are provided, corresponding to three or more mobile bodies.

[0019] Following the autonomous driving control device 100 is a group control device 300 that controls the operation plan so that each mobile unit does not collide with or interfere with one another. The group control device 300 may be, for example, a server computer equipped with a processor such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) and memory. The group control device 300 includes an optimization unit 301, a task management assignment unit 302, and an operation management unit 303. The optimization unit 301 optimizes the route plan and destination from each autonomous driving control device 100 using the deep reinforcement learning model that takes acceleration data into consideration, as described above. The task management assignment unit 302 receives the output from the deep reinforcement learning model 20 described above and assigns the optimized route plan and destination, and other tasks, to each mobile unit. The operation management unit 303 manages the operation so that each mobile unit does not collide with another.

[0020] The task management assignment unit 302 assigns task and path information to each mobile unit (i.e., AMR) to prevent collisions between AMRs. Each AMR moves according to the assigned task and path information, and during this process, it acquires acceleration and deceleration values ​​from each acceleration sensor. In this way, the deep reinforcement learning model is retrained using the acquired acceleration and deceleration values. If environmental changes occur (for example, changes in layout, an increase in the number of AMRs, the appearance of obstacles, etc.), AMRs may collide with each other. Therefore, such retraining can be effective because it can self-correct without human intervention. Instructions are given to each AMR based on the results of the retraining.

[0021] According to the embodiment described above, task and path information that takes into account the positional deviation caused by the actual AMR due to acceleration and deceleration and the AMR recognized by the group control system can be instructed to each AMR. This makes it possible to utilize multiple AMRs in a confined space and avoids increased costs due to layout changes.

[0022] Figure 5 is a flowchart showing a group control method according to an embodiment. The acceleration data acquisition unit acquires acceleration and deceleration data from acceleration sensors 11 (e.g., gyro sensors) mounted on each AMR (step S11). The acquired acceleration and deceleration data is input to the deep reinforcement learning model (step S12). The deep reinforcement learning model is trained to correct the discrepancy between the actual position of the AMR acquired by the mobile position acquisition unit using a position sensor (e.g., LiDAR) and the position of the AMR recognized by the swarm control system (position information estimation unit) (to prevent collisions between AMRs) (step S13). The swarm control unit assigns task and path information to each AMR to prevent collisions between AMRs (step S14). The acceleration and deceleration values ​​of each AMR moving with the assigned task and path information are acquired, and the system is retrained from the acquired acceleration and deceleration values ​​(step S15). Instructions are given to each AMR based on the results of the retraining (step S16).

[0023] Figure 6 is a block diagram showing an example configuration of a swarm control system consisting of an autonomous driving control device 100 and a swarm control device 300. Referring to Figure 6, the swarm control system includes a network interface 1201, a processor 1202, and a memory 1203. The network interface 1201 is used to communicate with other network node devices that constitute the communication system. The network interface 1201 may also be used for wireless communication. For example, the network interface 1201 may be used for wireless LAN communication as defined in the IEEE 802.11 series, or for mobile communication as defined in 3GPP® (3rd Generation Partnership Project). Alternatively, the network interface 1201 may include, for example, a network interface card (NIC) compliant with the IEEE 802.3 series. The aforementioned acceleration sensor 11 and position sensor 12 are connected to the network interface 1201.

[0024] The processor 1202 reads and executes software (computer programs) from the memory 1203 to perform the processing of the group control system described using a flowchart or sequence in the above embodiment. The processor 1202 may be, for example, a microprocessor, an MPU (Micro Processing Unit), a CPU (Central Processing Unit), or a GPU (Graphics Processing Unit). The processor 1202 may include multiple processors.

[0025] Memory 1203 is composed of a combination of volatile and non-volatile memory. Memory 1203 may also include storage located away from the processor 1202. In this case, the processor 1202 may access memory 1203 via an I / O interface not shown.

[0026] In the example shown in Figure 6, memory 1203 is used to store a group of software modules. The processor 1202 can read these software modules from memory 1203 and execute them, thereby performing the group control system processing described in the above embodiment.

[0027] As explained with reference to Figure 5, each processor in the group control system executes one or more programs containing a set of instructions for causing the computer to perform the algorithm described in the diagram.

[0028] In the examples described above, the program includes a set of instructions (or software code) that, when loaded into a computer, cause the computer to perform one or more of the functions described in the embodiments. The program may be stored on a non-temporary computer-readable medium or a physical storage medium. Examples, but not limited to, include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technologies, CD-ROM, digital versatile disc (DVD), Blu-ray® disc or other optical disc storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices. The program may be transmitted over a temporary computer-readable medium or a communication medium. Examples, but not limited to, include temporary computer-readable medium or a communication medium that includes electrically, optically, acoustically or otherwise propagating signals.

[0029] It should be noted that the present invention is not limited to the embodiments described above, and can be modified as appropriate without departing from the spirit of the invention. For example, although an AMR was described as a mobile body in the above embodiments, it can also be applied to mobile bodies that can move in the vertical direction, such as drones and airplanes. [Explanation of symbols]

[0030] 10 Mobile Units 11. Accelerometer 12 Position Sensors 20 Deep Reinforcement Learning Models 100 Autonomous Driving Control System 101 Self-position estimation part 102 Route Planning Department 201 Input Layer 202 Hidden Layer 203 Output Layer 300 Group Control Device 301 Optimization Department 302 Task Management and Assignment Section 303 Operation Management Department

Claims

1. In a swarm control system that controls multiple autonomously moving objects, At least one memory that stores instructions, At least one control unit configured to execute the aforementioned instruction, Acceleration sensors and position sensors attached to each moving object, Equipped with, The control unit, A position information estimation unit that estimates the position information of each moving object, A route planning unit that creates a route plan for each moving object based on the estimated location information, An acceleration data acquisition unit that acquires acceleration data from the acceleration sensor, A mobile body position acquisition unit that acquires the actual position of each mobile body using the position sensor, Equipped with, A swarm control system configured to optimize the route planning and destination from an autonomous driving control device using a deep reinforcement learning model that has been trained to correct the amount of discrepancy based on the acquired acceleration data so that the amount of discrepancy between the actual position of the moving body obtained using the position sensor and the estimated position of the moving body becomes zero.

2. The group control system according to claim 1, configured to optimize the route plan and destination from the autonomous driving control device using the deep reinforcement learning model, which has been trained to prevent interference between each moving object based on the estimated position information and the amount of displacement.

3. A group control system according to claim 1, configured to assign a path plan, planned using a trained deep reinforcement learning model, to each moving object, to move each moving object according to the path plan, to acquire acceleration data obtained from the acceleration sensor during the movement, to correct the amount of discrepancy between the actual position of the moving object obtained using the position sensor and the estimated position of the moving object based on the acquired acceleration data, and to retrain the deep reinforcement learning model so that the moving objects do not interfere with each other.

4. The group control system according to claim 1, configured to assign a path plan to each mobile object and move each mobile object based on the learned deep reinforcement learning model.

5. A group control method for controlling multiple autonomously moving objects, Estimating the positional information of each moving object, Based on the estimated location information, a path plan for each moving object is created, To acquire acceleration data from an acceleration sensor, Obtaining the actual position of each moving object using position sensors, Includes, A group control method that optimizes the route plan and destination from an autonomous driving control device using a deep reinforcement learning model that has been trained to correct the amount of discrepancy based on the acquired acceleration data so that the amount of discrepancy between the actual position of the moving body obtained using the position sensor and the estimated position of the moving body becomes zero.

Images