Control method for a moving object, transport device, and work system

Abstract

A management device (100) performs, as a method for controlling a mobile object: a step for generating a machine learning model for generating an operation plan for a mobile object for transporting an object, on the basis of the position of a first feature portion in a simulation space; a step for identifying the position of a second feature portion in the real space; a step for acquiring the relationship between the position of the second feature portion in the real space and a position in the real space corresponding to the position of the first feature portion in the simulation space; a step for applying the position of the second feature portion in the real space and the relationship to the machine learning model, to thereby generate an operation plan for the mobile object; and a step for controlling the operation of the mobile object in accordance with the generated operation plan.

Description

Technical Field

[0001] The present invention relates to the control of the operation of a moving body for transporting an object.

Background Art

[0002] Conventionally, moving bodies such as robots have been used for transporting objects. Various studies have been made on the control of the operation of such moving bodies. For example, Japanese Patent No. 6457421 (Patent Document 1) discloses a mechanical system including a machine learning device for taking out a workpiece from a cage by a robot. In this mechanical system, the machine learning device performs training of a machine learning model using simulation by a simulator.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the control of the operation of a moving body, a given position such as a target position is used as a reference point to control the position of the moving body. In the simulation space, the control device can recognize the positional relationship between the moving body and the reference point. On the other hand, in the real space, in order for the control device to recognize the positional relationship between the moving body and the reference point, the position of a marker installed in the real space, the position of a given location of an existing device, a pattern, etc. are used as reference points. When a marker is installed, in the control of the operation of the moving body in the real space, the control device detects the marker and controls the operation of the moving body based on the position of the detected marker.

[0005] In real space, there may be obstacles to the detection of reference points used as a reference for position, such as the presence of objects other than the object and the moving object. If such objects are not detected in real space, a reference for position cannot be obtained, making it difficult to control the movement of the moving object.

[0006] This invention was conceived in view of the circumstances described herein, and its purpose is to provide a technology for reliably controlling the movement of a moving object using the results of simulations. [Means for solving the problem]

[0007] A method for controlling a moving body according to a certain aspect of the present disclosure comprises the steps of: generating a machine learning model for generating a motion plan for a moving body to transport an object based on the position of a first feature part in a simulation space; identifying the position of a second feature part in real space; obtaining a relationship between the position of the second feature part in real space and the position in real space corresponding to the position of the first feature part in simulation space; generating a motion plan for a moving body by applying the position and relationship of the second feature part in real space to the machine learning model; and controlling the motion of the moving body according to the generated motion plan.

[0008] A transport device according to a certain aspect of the present disclosure comprises a mobile body for transporting an object, a controller for generating a machine learning model for generating a motion plan for the mobile body based on the position of a first feature part in a simulation space, a data acquisition unit for acquiring data for identifying the position of a second feature part in real space, and a memory for storing the relationship between the position of the second feature part in real space and the position in real space corresponding to the position of the first feature part in a simulation space. The controller uses the data acquired by the data acquisition unit to identify the position of the second feature part in real space, applies the position and relationship of the second feature part in real space to the machine learning model to generate a motion plan for the mobile body, and operates the mobile body according to the generated motion plan. [Effects of the Invention]

[0009] According to certain aspects of this disclosure, the control of the movement of a moving object using the results of a simulation can be reliably achieved. [Brief explanation of the drawing]

[0010] [Figure 1] This is a diagram illustrating the overview of the work system in the embodiment. [Figure 2] This figure shows an example of the hardware configuration of the work system in the embodiment. [Figure 3] This is a flowchart of the processes performed by the control device 100. [Figure 4] This diagram schematically shows an example of a simulation space defined by the simulator unit 112. [Figure 5] This is a flowchart of the subroutine for the transport process in step S3. [Modes for carrying out the invention]

[0011] Embodiments of this disclosure will be described in detail below with reference to the drawings. Parts identical or corresponding to those shown in the drawings will be denoted by the same reference numerals, and their descriptions will not be repeated. The embodiments and modifications described below may be selectively combined as appropriate.

[0012] [Overview of the work system] Figure 1 is a diagram illustrating the overview of the work system in the embodiment. The work system 1 in the embodiment includes a group of controlled devices 200 and a management device 100. The group of controlled devices 200 includes a transport robot 230, a work device 220, and a work device 240.

[0013] The transport robot 230 uses its arm 232 to transport the well plate 5 from the work device 220 to a target position Tp within the work device 240. The arm 232 is an example of a “moving body” in this disclosure. The arm 232 is configured as part of the transport robot 230 and moves with the movement of the transport robot 230. In this sense, the transport robot 230 itself can also be interpreted as an example of a “moving body”. The well plate 5 contains samples to be used by the work device 220 and the work device 240. The well plate 5 containing the samples is an example of an “object” in this disclosure. The work device 220 and the work device 240 are examples of devices for working with samples transported by the transport robot 230. The work device 220 is a rack for holding the well plate 5. The work device 240 is a dispensing device for dispensing the samples contained in the well plate 5.

[0014] In this embodiment, a system for culturing bacteria is used as an example of the work system 1, an experiment is used as an example of work on an object, and an experimental apparatus is used as the work apparatus. Note that the work on the object is not limited to experiments. Any type of work, such as metalworking, can be used as the work on the object. Therefore, any type of apparatus for performing work on an object, other than an experimental apparatus, can be used as the work apparatus.

[0015] The control device 100 controls the operation of the transport robot 230, the work device 220, and the work device 240. The control device 100 is an example of a controller. The transport device 300 is a device that includes the control device 100 and the transport robot 230.

[0016] The transfer robot 230 is a six-axis vertical articulated robot having one arm. More specifically, the transfer robot 230 includes a main body portion 231, an arm 232, a gripper 233, and an imaging unit 234. The main body portion 231 holds the arm 232. The arm 232 transfers the well plate 5 from the working device 220 to the target position Tp in the working device 240. The gripper 233 is provided at the tip of the arm 232 and grips the well plate 5. The arm 232 includes one or more movable parts 8. By moving the one or more movable parts 8, the arm 232 can move up and down, left and right, and back and forth. Thereby, the well plate 5 is transferred from the working device 220 to the target position Tp in the working device 240 by the arm 232.

[0017] In the example of FIG. 1, the controlled device group 200 includes a plurality of working devices, and the transfer robot 230 transfers an object (well plate) from one of the plurality of working devices to another working device. Note that the controlled device group 200 may be constituted by a single working device, and the transfer robot 230 may transfer the object from one position to another position within the single working device.

[0018] The imaging unit 234 is provided on the arm 232 and acquires an image of the imaging target. The operations of the arm 232, the gripper 233, and the imaging unit 234 are controlled by the management device 100.

[0019] In the working system 1, the mark Tm indicates the target position Tp. Further, marks Rm1, Rm2, and Rm3 are attached to the objects around the transfer robot 230. The mark Rm1 indicates the reference position Rp1, the mark Rm2 indicates the reference position Rp2, and the mark Rm3 indicates the reference position Rp3. Each of the reference positions Rp1, Rp2, and Rp3 is located at a place different from the target position Tp.

[0020] In the following description, when the marks Rm1, Rm2, and Rm3 are focused on for their common property (indicating a reference position) and do not need to be distinguished from each other, they are referred to as "mark Rm". The mark Rm may be any pattern that can identify the posture of the mark Rm. For example, it may be an AR marker, a QR code (registered trademark), or the like. In the present embodiment, "posture" includes position (the relative position of the marker with respect to the "imaging unit 234" described later) and orientation. The marks Rm1, Rm2, and Rm3 are distinguishable from each other. For example, when the marks Rm1, Rm2, and Rm3 are realized as QR codes, they are realized as QR codes having different information from each other.

[0021] The reference position Rp is located at a place different from the target position Tp. In the following description, when the reference positions Rp1, Rp2, and Rp3 are focused on for their common property (being reference positions) and do not need to be distinguished from each other, they are referred to as "reference position Rp".

[0022] In the present embodiment, the target position Tp is an example of the first feature part, and the reference position Rp is an example of the second feature part.

[0023] [Hardware Configuration of the Work System] FIG. 2 is a diagram showing an example of the hardware configuration of the work system in the embodiment. The work system 1 includes a transfer device 300, a work device 220, and a work device 240. The transfer device 300 includes a transfer robot 230 and a management device 100.

[0024] The management device 100 is configured by, for example, a general-purpose computer. In the example of FIG. 2, the management device 100 includes a processor 101, a memory 102, a storage 103, an interface 104, a display 105, and an input device 106.

[0025] The processor 101 executes various programs for the management device 100 to perform various processes. The processor 101 is composed of hardware elements such as a CPU (Central Processing Unit) and an MPU (Micro-Processing Unit).

[0026] The processor 101 functions as a machine learning unit 111, a simulator unit 112, an information generation unit 113, and a transport control unit 114 by executing a given program.

[0027] The machine learning unit 111 performs processing for machine learning a model to generate a motion plan for the arm 232. In one implementation example, the motion plan for the arm 232 includes information defining the movement of the arm 232 (for example, the distance and / or velocity of movement in each of the three axes). The motion plan for the arm 232 may also include information defining the movements of elements other than the arm 232 (such as the gripper 233, the motor for rotating the wheels attached to the main body 231, etc.) that contribute to the movement of the arm 232 for transporting the object.

[0028] The simulator unit 112 simulates the movement of the arm 232 in the simulation space. In one implementation example, the simulator unit 112 generates time-series data of the position of the arm 232. The information generation unit 113 generates "positional relationship information," which will be described later. The transport control unit 114 controls the movement of the transport robot 230 for transporting the well plate 5.

[0029] Memory 102 functions as the main memory and is composed of volatile memory devices such as DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory).

[0030] The storage 103 stores programs executed by the processor 101 and various data necessary for program execution, and is composed of non-volatile storage devices such as an SSD (Solid State Drive) and / or flash memory.

[0031] Furthermore, the program may be provided not as a standalone program, but incorporated as part of any other program. In this case, the processing according to this embodiment is realized in cooperation with the other program. Even if a program does not include such a partial module, it does not deviate from the intent of the management device 100 according to this embodiment. In addition, some or all of the functions provided by the program may be implemented by dedicated hardware.

[0032] The data stored in storage 103 includes a machine learning model 131. The machine learning model 131 includes data that constitutes a machine learning model for generating a motion plan for arm 232 (for example, values ​​of one or more parameters that are the learning results of the machine learning model). The data stored in storage 103 further includes positional relationship information 132. The positional relationship information 132 indicates the positional relationship in real space between two or more marks.

[0033] Interface 104 relays communication between the management device 100 and external devices (for example, the transport robot 230, the work device 220, the work device 240, etc.).

[0034] The display 105 displays the results of the calculations performed by the processor 101, and the input device 106 (e.g., mouse, keyboard, touch sensor, etc.) accepts data input operations to the processor 101.

[0035] The transport robot 230 includes an imaging unit 234, an interface 235, a motor unit 236, and a driver unit 237. The interface 235 relays communication between the transport robot 230 and the control device 100. The motor unit 236 includes motors associated with one or more movable parts 8. The driver unit 237 includes drivers that drive each of the multiple motors included in the motor unit 236.

[0036] [Process Flow] Figure 3 is a flowchart of the process performed by the management device 100. In one implementation example, the process in Figure 3 is realized by the processor 101 executing a given program.

[0037] As shown in Figure 3, the control device 100 performs the learning process of the machine learning model 131 in step S1, the generation process of positional relationship information (generation of positional relationship information 132) in step S2, and the transport process to control the transport operation of the well plate 5 by the arm 232 in step S3. The contents of each step will be described in detail below.

[0038] (1) Step S1: Learning process One example of the learning process is the use of a reinforcement learning algorithm. In this example, machine learning model 131 is trained according to a reinforcement learning algorithm.

[0039] In reinforcement learning, an agent selects various actions a under a given state s, and is given a reward for the action a it chooses. Through this, the machine learning model 131 is trained to learn the agent to make better action choices, i.e., to learn the correct value Q(s,a).

[0040] In this embodiment, arm 232 is used as the agent. As state s, the motion plan for arm 232 is adopted. An example of the data that makes up the motion plan includes the position where the transport robot 230 starts gripping the well plate 5, the movement path of arm 232, and / or the position where the transport robot 230 ends gripping the well plate 5 (the placement position of the well plate 5).

[0041] As a reward, the result of the transport of the well plate 5 (success / failure) is used. In one implementation example, if the position of the well plate 5 (placement position of the well plate 5) after the arm 232 has moved according to the motion plan is within a given range, the machine learning unit 111 identifies "success" as the result of the transport of the well plate 5. If the position of the well plate 5 (placement position of the well plate 5) after the arm 232 has moved according to the motion plan is outside the given range, the machine learning unit 111 identifies "failure" as the result of the transport of the well plate 5.

[0042] During the learning process, the machine learning unit 111 instructs the simulator unit 112 to simulate the movement of the arm 232 according to each of the multiple motion plans (states s).

[0043] Figure 4 is a schematic diagram showing an example of a simulation space defined by the simulator unit 112. The simulation space 900 includes elements 905, 920, 930, and 940.

[0044] Element 905 corresponds to well plate 5 in real space. Element 920 corresponds to work device 220 in real space. Element 930 corresponds to transport robot 230 in real space. Element 940 corresponds to work device 240 in real space. Element 930 includes elements 931, 932, and 933. Elements 931, 932, and 933 each correspond to the main body 231, arm 232, and gripper 233 in real space, respectively. Element Xp corresponds to the target position Tp in real space. In the above simulation, the motion plan causes element 930 to transport element 905 from element 920 to element 940 (more specifically, to the location specified by element Xp). Thus, in the simulation, element 930 transports element 905 from element 920 to element 940.

[0045] The machine learning unit 111 obtains the result (reward) of each action (simulation) according to multiple action plans. Then, the machine learning unit 111 uses the combination of state s and reward for each of the multiple action plans to perform the training process of the machine learning model 131.

[0046] The method for generating the machine learning model 131 is not limited to using the reinforcement learning algorithm described above. The machine learning model 131 may also be generated by searching for a trajectory according to a rule-based method (for example, searching for a trajectory connecting the above points) using information that identifies the movement path prepared by a given method (for example, a starting point for grasping, an ending point for grasping, and points to pass through).

[0047] (2) Step S2: Positional relationship information generation process In the positional relationship information generation process, information (positional relationship information) is generated that identifies the positional relationship between at least one of the reference positions Rp1, Rp2, and Rp3 and the target position Tp.

[0048] In one implementation example, the information generation unit 113 captures a mark Rm using the imaging unit 234, identifies the orientation of the mark Rm in the captured image, and determines the position of the reference position Rp in real space based on the orientation of the mark Rm. The information generation unit 113 also captures a mark Tm using the imaging unit 234, identifies the orientation of the mark Tm in the captured image, and determines the position of the target position Tp in real space based on the orientation of the mark Tm. The information generation unit 113 then generates positional relationship information as data representing the relationship between the position of the reference position Rp and the position of the target position Tp in real space. The generated positional relationship information is stored in the storage 103 as positional relationship information 132. An example of positional relationship information is the difference between the coordinates of the reference position Rp in real space and the coordinates of the target position Tp in real space.

[0049] The information generation unit 113 may also identify the positions of the reference positions Rp1, Rp2, and Rp3 in real space, and generate positional relationship information, including data representing the relationship between the position of reference position Rp1 and the position of target position Tp, data representing the relationship between the position of reference position Rp2 and the position of target position Tp, and data representing the relationship between the position of reference position Rp3 and the position of target position Tp.

[0050] (3) Step S3: Conveying process Figure 5 is a flowchart of the subroutine for the transport process in step S3.

[0051] In step S31, the processor 101 commands the transport robot 230 to start searching for mark Rm. More specifically, the processor 101 commands the imaging unit 234 to start imaging and commands the drivers corresponding to one or more movable parts 8 to start driving the motors. As a result, imaging by the imaging unit 234 begins, and the arm 232 moves up and down, left and right, and forward and backward. The imaging unit 234 transmits the acquired images to the processor 101.

[0052] In step S32, the processor 101 determines whether or not the mark Rm has been found. The processor 101 determines that the mark Rm has been found if the image transmitted from the imaging unit 234 contains a pixel that indicates the mark Rm. If the mark Rm has been found (YES in step S32), the processor 101 proceeds to step S33.

[0053] In step S33, the processor 101 commands the transport robot 230 to end the search for mark Rm. More specifically, the processor 101 commands the imaging unit 234 to end imaging and commands the drivers corresponding to one or more movable parts 8 to stop driving the motors. As a result, the arm 232 stops in the posture (position and orientation) it had when mark Rm was found.

[0054] In step S34, the processor 101 commands the imaging unit 234 to take an image. The captured image includes pixels that represent the mark Rm. Note that in step S34, before commanding the imaging unit 234 to take an image, the processor 101 may bring the imaging unit 234 (arm 232) closer to the mark Rm. This may result in the captured image containing more detailed information about the mark Rm.

[0055] In step S35, the processor 101 uses the image captured in accordance with the command in step S34 to determine the position of the reference position Rp in real space. More specifically, the processor 101 identifies the orientation of the mark Rm in the image and, based on the identified orientation and the position of the imaging unit 234 in real space, determines the position of the reference position Rp in real space.

[0056] In step S36, the processor 101 reads the positional information 132 from the storage 103.

[0057] In step S37, the processor 101 derives the position of the target location in real space.

[0058] To derive the real-world position of the target location, the processor 101 utilizes positional relationship information and the real-world position of the reference position Rp (identified in step S35). In one example, the processor 101 derives the real-world coordinates of the target location by adding the coordinates stored as positional relationship information to the real-world coordinates of the reference position Rp, and then treats these derived coordinates as the real-world position of the target location. Theoretically, the derived position of the target location coincides with the target location Tp shown in Figure 1.

[0059] In step S38, the processor 101 generates a motion plan for the transport robot 230 using the position derived in step S37 (the position of the target location in real space). More specifically, the processor 101 generates the motion plan using the machine learning model 131. Even more specifically, the processor 101 inputs the position derived in step S37 (the position of the target location in real space) as the target location for transporting the well plate 5 into the machine learning model 131 to generate a motion plan for the transport robot 230.

[0060] In step S39, the processor 101 operates the transport robot 230 (arm 232) according to the motion plan generated in step S38. More specifically, the processor 101 commands the transport robot 230 to operate according to the motion plan. After that, the processor 101 returns control to the process shown in Figure 3.

[0061] In the embodiment described above, the position of the second feature part in real space (the position of the reference position Rm) is identified. Furthermore, the relationship (positional relationship information) between the position of the second feature part in real space and the position in real space corresponding to the position of the first feature part in simulation space (the position identified by element Xp) is obtained. The machine learning model then uses the position of the second feature part in real space and the above relationship to generate a motion plan for the moving object. That is, since the position in real space corresponding to the position of the first feature part in simulation space is indirectly identified when generating the motion plan, it is not necessary to directly detect the position of the first feature part in real space. Therefore, even if there is something that could hinder the detection of the position of the first feature part in real space, the situation in which it becomes difficult to control the motion of the moving object is avoided. Thus, the control of the motion of the moving object is reliably achieved.

[0062] [Differentiation] In step S37, the reference position Rp used together with the positional relationship information may be singular or plural. That is, the marker indicating the reference position Rp may consist of one element (for example, one of marks Rm1, Rm2, or Rm3) or multiple elements (for example, two or more of marks Rm1, Rm2, or Rm3).

[0063] The following describes in detail the processing flow when three reference positions Rp1, Rp2, and Rp3 are used along with positional relationship information in step S37.

[0064] In this case, steps S31 to S35 are performed for each of the marks Rm1, Rm2, and Rm3. As a result, in step S34, the processor 101 is controlled to capture an image containing the pixel indicating mark Rm1, an image containing the pixel indicating mark Rm2, and an image containing the pixel indicating mark Rm3. In step S35, the processor 101 is controlled to determine the real-world positions of each of the reference positions Rp1, Rp2, and Rp3. Then, in step S37, the processor 101 derives a first provisional target position using the positional relationship information and the real-world position of reference position Rp1, a second provisional target position using the positional relationship information and the real-world position of reference position Rp2, and a third provisional target position using the positional relationship information and the real-world position of reference position Rp3. Then, the processor 101 derives the final position of the target location in real space as the average value of the first to third provisional target locations (for example, the average value of the coordinates).

[0065] In this embodiment, the mark Rm indicating the reference position Rp is affixed to the work devices 220 and 240. That is, the mark Rm is configured separately from the work devices 220 and 240. The processor 101 recognizes pixels corresponding to the image information registered as the mark Rm from the image captured by the imaging unit 234. The mark Rm may also be composed of a part of the work device 220 and / or the work device 240 (for example, the manufacturer's logo portion of the work device affixed to the work device). In this case, the storage 103 may store image information for identifying a part of the work device 220 and / or the work device 240 as the mark Rm. The processor 101 may use the image captured by the imaging unit 234 and the image information stored in the storage 103 to determine the orientation of the part and then determine the position of the part in real space.

[0066] Furthermore, in this embodiment, the image captured by the imaging unit 234 is used as a method for determining the position of the second feature point in real space. In this sense, the imaging unit 234 is an example of a data acquisition unit that acquires data for determining the position of the second feature point in real space. Note that the method for determining the position of the second feature point in real space may be a method other than using the captured image, such as using a beacon. For example, a beacon receiver may be installed on the transport robot 230. The second feature point may be composed of multiple beacons. The processor 101 may determine the position of the second feature point in real space based on the received signal strength from each of the multiple (three or more) beacons. In this case, a receiver that receives signals from beacons constitutes an example of a data acquisition unit that acquires data for determining the position of the second feature point in real space.

[0067] [Pattern] Those skilled in the art will understand that the above-described exemplary embodiments are specific examples of the following embodiments.

[0068] (Section 1) A method for controlling a moving body according to one embodiment may include the steps of: generating a machine learning model for generating a motion plan for a moving body to transport an object based on the position of a first feature part in a simulation space; identifying the position of a second feature part in real space; obtaining a relationship between the position of the second feature part in real space and the position in real space corresponding to the position of the first feature part in a simulation space; generating a motion plan for the moving body by applying the position of the second feature part in real space and the positional relationship to the machine learning model; and controlling the motion of the moving body according to the generated motion plan.

[0069] According to the control method for the mobile body described in paragraph 1, the control of the mobile body's movement using the results of the simulation can be reliably achieved.

[0070] (Clause 2) In the method for controlling a moving body described in paragraph 1, the second feature part may include a part of a work device for performing work using the object.

[0071] According to the control method for a mobile body described in paragraph 2, the number of types of components used to implement the control method is minimized.

[0072] (Clause 3) In the method for controlling a moving object described in paragraph 1, the second feature part may include a marker configured separately from a work device for working with the object.

[0073] According to the control method for the moving body described in paragraph 3, an element having a structure and form suitable for use as a second feature part can be used as a second feature part.

[0074] (Clause 4) In the method for controlling a moving body described in any one of paragraphs 1 to 3, the second feature part may include a plurality of elements that are identifiable from one another.

[0075] According to the control method for the moving object described in Section 4, the position of the second feature part in real space can be determined using the positions of each of the multiple elements, thereby improving the accuracy of determining the position of the second feature part in real space.

[0076] (Clause 5) In the method for controlling a moving object described in any one of Clauses 1 to 4, a reinforcement learning algorithm may be used to generate the machine learning model.

[0077] According to the control method for moving objects described in Section 5, the training process of a machine learning model for generating complex motion plans can be easily carried out.

[0078] (Clause 6) In the method for controlling a moving object described in any one of Clauses 1 to 5, applying the position of the second feature portion in real space and the positional relationship to the machine learning model may include using the positional relationship to derive a position in real space corresponding to the position of the first feature portion in simulation space from the position of the second feature portion in real space.

[0079] According to the control method for the moving body described in paragraph 6, a specific method for utilizing the position of the second feature part in real space and the above-mentioned relationship can be presented.

[0080] (Clause 7) A transport device according to one embodiment comprises a mobile body for transporting an object, a controller for generating a machine learning model for generating an action plan for the mobile body based on the position of a first feature part in a simulation space, a data acquisition unit for acquiring data for identifying the position of a second feature part in real space, and a memory for storing the relationship between the position of the second feature part in real space and the position in real space corresponding to the position of the first feature part in a simulation space, wherein the controller may use the data acquired by the data acquisition unit to identify the position of the second feature part in real space, apply the position of the second feature part in real space and the positional relationship to the machine learning model to generate an action plan for the mobile body, and operate the mobile body according to the generated action plan.

[0081] According to the transport device described in Section 7, the control of the movement of the moving object using the results of the simulation can be reliably achieved.

[0082] (Clause 8) In the transport device described in Clause 7, the controller may recognize a part of a work device for working with the object as the second feature part.

[0083] According to the conveying device described in paragraph 8, the number of types of components used for conveying the moving object is minimized.

[0084] (Clause 9) In the transport device described in Clause 7, the controller may recognize a marker separately configured for a work device for working with the object, as the second feature component.

[0085] According to the conveying device described in paragraph 9, elements having a structure and form suitable for use as a second feature can be used as a second feature.

[0086] (Clause 10) In the transport device described in any one of paragraphs 7 to 9, the controller may recognize each of a plurality of elements that are identifiable from one another as the second feature part.

[0087] According to the transport device described in paragraph 10, the position of the second feature portion in real space can be determined using the positions of each of the multiple elements, thereby improving the accuracy of determining the position of the second feature portion in real space.

[0088] (Clause 11) In the transport device described in any one of Clauses 7 to 10, a reinforcement learning algorithm may be used to generate the machine learning model.

[0089] According to the transport device described in paragraph 11, the training process of a machine learning model for generating complex motion plans can be easily carried out.

[0090] (Clause 12) In the transport device described in any one of Clauses 7 to 11, applying the position of the second feature portion in real space and the positional relationship to the machine learning model may include using the positional relationship to derive a position in real space corresponding to the position of the first feature portion in simulation space from the position of the second feature portion in real space.

[0091] According to the transport device described in paragraph 12, the position of the second feature in real space and the method of utilizing the above-mentioned relationship can be specifically presented.

[0092] (Paragraph 13) A work system according to one embodiment may include a transport device described in any one of paragraphs 7 to 12, and a work device for performing work using an object transported by the transport device.

[0093] According to the work system described in Section 13, the control of the movement of the mobile object using the simulation results can be reliably achieved.

[0094] (Clause 14) The work system described in paragraph 13 may further comprise one or more elements constituting the second feature, which are configured separately from the work apparatus.

[0095] According to the work system described in paragraph 14, the work system can be constructed more reliably. The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of this disclosure is indicated by the claims rather than by the description of the embodiments above, and all modifications within the meaning and scope of the claims are intended to be included. Furthermore, each technique in the embodiments is intended to be practiced individually or, as far as possible, in combination with other techniques in the embodiments. [Explanation of symbols]

[0096] 1 Work system, 5 Well plate, 8 Movable parts, 100 Management device, 101 Processor, 102 Memory, 103 Storage, 104, 235 Interface, 105 Display, 106 Input device, 111 Machine learning unit, 112 Simulator unit, 113 Information generation unit, 114 Transport control unit, 131 Machine learning model, 132 Positional relationship information, 200 Controlled group, 220, 240 Work device, 230 Transport robot, 231 Main body, 232 Arm, 233 Gripper, 234 Imaging unit, 236 Motor unit, 237 Driver unit, 300 Transport device, 900 Simulation space, 905, 920, 930, 931, 932, 933, 940, Xp element.

Claims

1. A step of generating a machine learning model for generating a motion plan for a moving body to transport an object, based on a first virtual position in a simulation space that corresponds to a first reference position in real space where a first feature part is located, The steps include identifying a second reference position in the real space where the second feature portion is located, A step of obtaining the positional relationship between the first reference position and the second reference position, The steps include: applying the second reference position and the positional relationship to the machine learning model to generate a motion plan for the moving object; A method for controlling a mobile body, comprising the steps of controlling the movement of the mobile body in accordance with the generated movement plan.

2. The method for controlling a moving body according to claim 1, wherein the second feature part includes a part of a work device for performing work using the object.

3. The method for controlling a moving object according to claim 1, wherein the second feature includes a marker configured separately from a work device for performing work using the object.

4. The method for controlling a moving body according to claim 1, wherein the second feature portion includes a plurality of elements that are identifiable from one another.

5. The method for controlling a moving object according to claim 1, wherein a reinforcement learning algorithm is used to generate the machine learning model.

6. The method for controlling a moving object according to claim 1, wherein applying the second reference position and the positional relationship to the machine learning model includes using the positional relationship to derive the first reference position from the position of the second feature portion in real space.

7. A method for controlling a moving body according to claim 1, wherein a motion plan for the moving body is generated by applying the second reference position and the positional relationship to the machine learning model without requiring the first reference position.

8. A mobile vehicle for transporting objects, A controller that generates a machine learning model for generating a motion plan for the moving object based on a first virtual position in the simulation space that corresponds to a first reference position in the real space where a first feature part is located, A data acquisition unit that acquires data for identifying a second reference position in the real space where the second feature portion is located, The system includes a memory for storing the positional relationship between the first reference position and the second reference position, The aforementioned controller, Using the data acquired by the data acquisition unit, the second reference position is identified. By applying the second reference position and the positional relationship to the machine learning model, a motion plan for the moving object is generated. A transport device that operates the moving body according to the generated motion plan.

9. The transport device according to claim 8, wherein the controller, as the second feature, recognizes a part of a work device for performing work using the object.

10. The transport device according to claim 8, wherein the controller, as a second feature, recognizes a marker separately configured for a work device for working with the object.

11. The transport device according to claim 8, wherein the controller recognizes each of a plurality of elements that are identifiable from one another as the second feature unit.

12. The transport device according to claim 8, wherein a reinforcement learning algorithm is used to generate the machine learning model.

13. The conveying device according to claim 8, wherein applying the second reference position and the positional relationship to the machine learning model includes using the positional relationship to derive a position in real space corresponding to the first reference position from the position in real space of the second feature portion.

14. The transport device according to claim 8, wherein the controller generates a motion plan for the moving body by applying the second reference position and the positional relationship to the machine learning model without requiring the first reference position.

15. A work system comprising a conveying device according to any one of claims 8 to 14, and a work device for performing work using an object conveyed by the conveying device.

16. The work system according to claim 15, further comprising one or more elements constituting the second characteristic part, which are configured separately from the work device.

Images