Crop harvesting methods

Abstract

The present invention provides a crop harvesting apparatus and a crop harvesting method that can reduce the load on the collection unit. [Solution] A crop harvesting device 1 for collecting crops comprises an arm portion 10 including a collection unit for collecting crops, a main body portion 100 connected to the arm portion 10 and including a plurality of legs 120 that contact a walking surface 9, and a control unit that controls the legs 120 based on a walking learning model constructed using a virtual environment. The walking learning model is characterized by being constructed by reinforcement learning using virtual state information including virtual arm portion information corresponding to the characteristics of the arm portion 10, virtual leg portion control information corresponding to the control conditions of the legs 120, and walking rewards set for the results of executing the virtual leg portion control information based on the virtual state information.

Description

Technical Field

[0001] The present invention , agriculture relates to a method for harvesting crops.

Background Art

[0002] Conventionally, in order to reduce the workload of agricultural operations, for example, a crop harvesting system disclosed in Patent Document 1 has been proposed.

[0003] In Patent Document 1, a crop harvesting system is disclosed that includes an end effector for crop harvesting, a robotic arm, a vision device, a tilting processing unit, and a cutting processing unit. The tilting processing unit can execute a tilting process that tilts the end effector along the main stem or branch so that the passing part passes the harvesting object outward from the base end position. Then, the cutting processing unit can execute a cutting process that cuts the fruit stalk passed inside the passing part by operating the end effector to reduce the passing part.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Here, Patent Document 1 discloses a technology that drives and controls a robotic arm to move an end effector and execute a movement process to pass the harvesting target through a passage. However, in actual agricultural work environments, there are obstacles around the harvesting target, such as tree branches and leaves, and nets to prevent the harvesting target from falling. In particular, the positional relationship between the harvesting target and the obstacles is not always consistent. Therefore, when the collection unit such as the end effector is brought close to the harvesting target, there is a concern that the collection unit may be damaged by contact with the obstacles. Accordingly, there is a need for a means to reduce the load on the collection unit such as the end effector. In this regard, Patent Document 1 neither describes nor suggests the above-mentioned concerns or means to reduce the load on the end effector.

[0006] Therefore, the present invention was devised in view of the above-mentioned problems, and its purpose is to reduce the load on the collection unit. farm The objective is to provide a method for harvesting crops. [Means for solving the problem]

[0007] The crop harvesting method according to the first invention is a crop harvesting method using a crop harvesting device for collecting crops, comprising a first modification step and a second modification step, the crop harvestThe device comprises an arm section including a collection unit for collecting the crops, a main body section including a plurality of legs connected to the arm section, and a control unit that controls the legs based on a walking learning model constructed using a virtual environment, wherein the walking learning model is constructed by reinforcement learning using virtual state information including virtual arm section information corresponding to the characteristics of the arm section, virtual leg section control information corresponding to the control conditions of the legs, and walking rewards set for the results of executing the virtual leg section control information based on the virtual state information, and includes the arm section and the main body section. The system further includes a sensor unit that acquires center of gravity data indicating the center of gravity and collection distance data indicating the distance from the collection unit to the crop, the virtual state information includes virtual center of gravity data corresponding to the center of gravity data, the virtual arm unit information includes virtual collection distance data corresponding to the collection distance data, the control unit uses the center of gravity data and the collection distance data as explanatory variables for the walking learning model, acquires leg control information indicating the control conditions of the leg as an objective variable, controls the leg using the leg control information, and the control unit is constructed using a virtual environment Based on a harvesting learning model, crop information indicating the characteristics of the crop is used as an explanatory variable, arm control information indicating the control conditions of the arm is acquired as an objective variable, and the arm is controlled using the arm control information. The harvesting learning model is constructed by reinforcement learning using virtual crop information corresponding to the crop information, virtual arm control information corresponding to the arm control information, and a harvesting reward set for the result of executing the virtual arm control information based on the virtual crop information. The control unit controls either the leg or the arm based on the result of comparing a preset center of gravity range with the center of gravity data. The first modification step is to change the control of the arm to the control of the leg if the center of gravity data falls outside the center of gravity range due to the control of the arm based on the harvesting learning model. The second modification step is to change the control of the leg to the control of the arm if, after the first modification step, the center of gravity data is included in the safety range contained within the center of gravity range due to the control of multiple legs based on the walking learning model. [Effects of the Invention]

[0013] First shot Clearly According to the system, the control unit controls the legs based on a walking learning model constructed using a virtual environment. The walking learning model is a learning model constructed through reinforcement learning using virtual state information including virtual arm information, virtual leg control information, and walking rewards. As a result, optimal leg control conditions can be set based on the characteristics of the arm, and damage to the collection unit due to contact with obstacles, for example, can be suppressed. This makes it possible to reduce the load on the collection unit.

[0018] Also , the 1 According to the invention, the first modification step involves controlling the arm portion, and the center of gravity data If the center of gravity is outside the range, the control of the arm is switched to the control of the leg. Therefore, the center of gravity data By moving the center of gravity outside the range, the possibility of the crop harvesting device tipping over can be reduced. Yes, it is possible. This will reduce the risk of crop harvesting equipment tipping over during crop collection. This becomes possible.

[0019] Also, the 1 According to the invention, the second modification step, after the first modification step, is performed within a safe range. If center of gravity data is included, the control of the legs is changed to the control of the arms. Therefore, agricultural crops Once the harvesting equipment is back to a position where it is unlikely to tip over, resume collecting crops as soon as possible. This makes it possible to improve the efficiency of collecting agricultural products. [Brief explanation of the drawing]

[0020] [Figure 1] Figure 1 is a schematic perspective view showing an example of a crop harvesting apparatus in this embodiment. [Figure 2] Figures 2(a) and 2(b) are schematic diagrams showing an example of a harvesting section. [Figure 3] Figs. 3(a) and 3(b) are schematic diagrams showing an example of a cutter part and a gripper part. [Figure 4] Fig. 4 is a schematic perspective view showing an example of a main body part. [Figure 5] Fig. 5 is a schematic top view showing an example of a main body part. [Figure 6] Figs. 6(a) to 6(d) are schematic perspective views showing examples of legs. [Figure 7] Figs. 7(a) to 7(c) are schematic diagrams showing examples of the center of gravity of the main body part. [Figure 8] Fig. 8 is a flowchart showing an example of a method for harvesting agricultural crops in an embodiment

Mode for Carrying Out the Invention

[0021] Hereinafter, an example of an agricultural crop harvesting apparatus and an agricultural crop harvesting method according to the present invention will be described in detail.

[0022] [[ID=]

[28] ](Embodiment: Agricultural Crop Harvesting Apparatus 1) Fig. 1 is a schematic perspective view showing an example of the agricultural crop harvesting apparatus 1 in the present embodiment.

[0023] The agricultural crop harvesting apparatus 1 is used for harvesting agricultural crops 5 and can reduce the work load of agricultural work. The agricultural crop harvesting apparatus 1 includes an arm part 10, a main body part 100, and a control part 30, and may include, for example, a sensor part.

[0024] The agricultural crops 5 include horticultural crops such as fruit trees and vegetables. The agricultural crops 5 indicate, for example, fruit trees and fruits such as oranges and apples that grow on the fruit trees.

[0025] <Arm part 10> The arm part 10 includes a plurality of joint parts 11 and a collection part 20. The plurality of joint parts 11 are each independently driven and can change the position and orientation of the tip part 12 of the arm part 10. The collection part 20 is provided at the tip part 12. As the arm part 10, for example, a configuration in which the collection part 20 is attached to a known robot arm may be used.

[0026] <Collection Section 20> The collection unit 20 is used to collect agricultural products 5, for example, as shown in Figures 2(a) and 2(b). The collection unit 20 includes, for example, a cutter unit 21 and a gripper unit 22. Alternatively, a known end effector for collecting agricultural products 5 may be used as the collection unit 20.

[0027] <Cutter section 21> The cutter section 21 is used to cut the supporting parts of the crop 5, such as fruit stalks and branches. The cutter section 21 includes a cutting section 21a, a cutting assist section 21b, and a rotating section 21c, as shown in Figures 3(a) and 3(b), for example.

[0028] The cutting portion 21a has a cutting edge for cutting the support portion. The cutting portion 21a may have any shape, for example, according to the characteristics of the crop 5.

[0029] The cutting assist portion 21b is used to suppress fluctuations in the support portion and to facilitate cutting of the support portion by the cutting portion 21a. The cutting assist portion 21b may have, for example, a slit shape that clamps the side surface of the support portion, or it may have any shape depending on the object.

[0030] The rotating part 21c is used in conjunction with the cutting part 21a to drive the cutting part 21a. For example, a shaft-shaped member may be used as the rotating part 21c, and the cutting part 21a is driven by the rotation of the rotating part 21c. The rotating part 21c may also be used in conjunction with, for example, the gripper part 22 to drive both the cutting part 21a and the gripper part 22.

[0031] <Gripper section 22> The gripper section 22 is used to hold the crops 5. The gripper section 22 includes, for example, a holding section 22a and a movable section 22b.

[0032] The holding portion 22a holds the crop 5 whose support portion has been cut by the cutter portion 21, for example. The holding portion 22a may have a flat plate shape, or it may have a shape that is easy to hold depending on the shape of the crop 5.

[0033] The movable part 22b is used to adjust the position of the holding part 22a. The movable part 22b is driven, for example, in conjunction with the rotation of the rotating part 21c, and changes the position of the holding part 22a. For example, when the rotating part 21c rotates to cut the support part, the movable part 22b changes the position of the holding part 22a in a direction that brings it closer to the crop 5. This makes it easier to hold the crop 5 with the holding part 22a when the support part is cut. In particular, when collecting fruits with easily deformable surfaces, such as oranges, deformation of the fruit surface can be suppressed by changing the position of the holding part 22a according to the size of the fruit. This makes it possible to suppress the deterioration of the quality of the crop 5.

[0034] <Control Unit 30> The control unit 30 controls the entire crop harvesting device 1, including the arm unit 10. The control unit 30 is housed in the main body unit 100, for example, as shown in Figure 1. The control unit 30 controls the arm unit 10, etc., by referring to a learning model constructed using, for example, a virtual environment. Details of the learning model will be described later.

[0035] The control unit 30 includes, for example, a central control unit, a storage unit, and a communication unit. The central control unit represents a processor such as a CPU (Central Processing Unit) and controls, for example, each component of the crop harvesting device 1. The storage unit represents, for example, volatile memory such as RAM (Random Access Memory) or non-volatile memory such as ROM (Read Only Memory), and stores various data and programs such as learning models. The communication unit represents known connection equipment for performing wireless communication such as Wi-Fi (registered trademark).

[0036] Furthermore, known control techniques can be used as means for the control unit 30 to control the arm unit 10, etc. For example, the central control unit can control the entire crop harvesting apparatus 1 by executing a program stored in non-volatile memory using volatile memory as a work area.

[0037] Furthermore, the control unit 30 may control each component of the arm unit 10, etc., using imaging information obtained from the sensor unit, depending on the application. In this case, the control unit 30 uses imaging information as an explanatory variable of the learning model and acquires control information indicating the control conditions of the arm unit 10, etc., as the objective variable. The control unit 30 uses the acquired control information to control the arm unit 10, etc.

[0038] For example, when the control unit 30 controls the arm unit 10, it stops the driving of the multiple joint units 11 before driving the cutter unit 21 and the gripper unit 22. In this case, when collecting crops 5 using the cutter unit 21 and the gripper unit 22, damage to the crops 5 caused by the driving of the joint units 11 can be prevented.

[0039] For example, if a movable device is used as the main body 100, the control unit 30 may perform efficient route selection based on sensor information obtained from the sensor unit. For example, route selection may involve A * Algorithms and known techniques such as Dijkstra's algorithm can be used. In this case, the control unit 30 can select a route using elevation data and topographic information (slope, presence or absence of obstacles, etc.) that have been stored in advance in the storage unit or the like.

[0040] <Imaging Unit 40> The sensor unit includes, for example, an imaging unit 40. The imaging unit 40 is provided at the tip 12. The imaging unit 40 images the crop 5 and generates imaging information included in the sensor information. The imaging unit 40 is mounted from the tip 12 toward the cutter 21, for example, as shown in Figures 2(a) and 2(b). In this case, it becomes easier to image the behavior of the crop 5 in response to the drive of the collection unit 20. As the imaging unit 40, for example, a known imaging device may be used, and for example, an imaging device capable of imaging in the infrared light region as well as the visible light region may be used.

[0041] The imaging unit 40, for example, images the crop 5 and acquires collection distance data, which indicates the distance from the collection unit 20 to the crop 5, as imaging information. The collection distance data may be calculated, for example, from an image in the visible light region using a known distance estimation technique, or it may be calculated from an image in the infrared region.

[0042] For example, if crop 5 refers to a fruit tree and the fruit growing on it, the imaging unit 40 will image the branches and leaves of the fruit tree, the fruit, and the environment surrounding the fruit tree, and generate imaging information. For example, if crop 5 refers to a root vegetable and its roots, the imaging unit 40 will image the root vegetable, its roots, and the soil, and generate imaging information.

[0043] For example, there are many obstacles around the fruit, such as branches and leaves of the fruit tree and nets to prevent the fruit from falling. In this case, the imaging unit 40 images the crop 5 and generates imaging information. The control unit 30 then derives the positional relationship between the branches and leaves of the crop 5 and the fruit based on the captured imaging information (e.g., an image). In this way, by using an image, the arm unit 10 can be controlled without the collection unit 20 coming into contact with the obstacles, even when there are many obstacles such as branches and leaves. Furthermore, even if the imaging unit 40 images an obstacle such as a net to prevent the fruit from falling, the arm unit 10 can be controlled without the collection unit 20 coming into contact with the obstacle.

[0044] <Harvesting Learning Model HM> The harvesting learning model HM represents a learning model constructed using reinforcement learning in a virtual environment. In the harvesting learning model HM, crop information is used as the explanatory variable, and arm control information is used as the objective variable.

[0045] <<Agricultural Crop Information>> The crop information indicates information about the characteristics of crop 5, including, for example, the shape, type, and location of obstacles during collection of crop 5, and may also include information about the environment, such as sunny, rainy, or snowy conditions. The crop information may include, for example, location information indicating the positional relationship between the branches and leaves of a fruit tree and the fruit. The crop information may include, for example, sensor information acquired by the sensor unit. The crop information may be pre-set by, for example, an administrator.

[0046] <<Arm Control Information>> The arm control information indicates the control conditions for the arm 10 in relation to the crop information. The arm control information includes, for example, the degree of drive of the joint 11 for moving the tip 12 of the arm 10 (e.g., motor torque). Torque may include, for example, the magnitude of the force, as well as, for example, the amount of change in the magnitude of the force and the reaction force.

[0047] For example, the arm control information may include parameters for the collection unit 20 that is driven to collect crops 5. The arm control information includes, for example, driving conditions for bringing the tip 12 of the arm 10 close to the fruit, based on position information. The driving conditions include information for controlling a plurality of joints 11 so that, for example, the tip 12 of the arm 10 passes through a path that avoids contact with branches and leaves.

[0048] In the virtual environment where reinforcement learning is performed, a virtual crop modeled after crop 5 and a virtual arm modeled after arm 10 are set up. The virtual arm represents a virtual model with functions equivalent to arm 10 in the virtual environment. The virtual crop represents the object that the virtual arm collects in the virtual environment, and any object can be set depending on the application. The harvesting learning model HM is constructed using known electronic equipment capable of performing reinforcement learning. Alternatively, a virtual device with functions equivalent to crop harvesting device 1 may be used as the virtual model.

[0049] Reinforcement learning is described as a known learning method that uses a virtual environment to build a model, and is implemented using, for example, Isaac Lab or Isaac Sim (both from NVIDIA). Reinforcement learning derives the Q function and policy using, for example, pre-prepared virtual arm and virtual crop data. The Q function is derived using, for example, an offline reinforcement learning algorithm such as CQL (Conservative Q-Learning). The policy provides guidelines (e.g., a function) for the virtual arm to select the optimal action for the virtual crop.

[0050] In reinforcement learning used to construct the harvesting learning model HM, a policy, virtual crop information VH, virtual arm control information VM, and harvesting reward Ch are used. The learning method involves determining the virtual arm control information VM, which is suitable for the virtual crop information VH, using the policy described above. Then, the virtual arm is driven based on the virtual arm control information VM. A harvesting reward Ch is then set based on the result of the virtual arm driving, and the policy is updated based on the harvesting reward Ch. By repeating the above, the policy is optimized, and the harvesting learning model HM is constructed.

[0051] <<Virtual Crop Information VH>> Virtual crop information VH indicates information about the characteristics of a virtual crop and corresponds to crop information. Virtual crop information VH may include, for example, the shape, type, and location of obstacles during collection of the virtual crop, as well as information about the environment, such as sunny, rainy, or snowy conditions. Virtual crop information VH may also include virtual position information that shows the positional relationship between virtual branches and leaves that mimic the branches and leaves of a fruit tree and virtual fruits that mimic the fruits.

[0052] The virtual crop information VH may include spatial information such as the positional relationship between the virtual arm and the virtual crop. For example, the virtual crop information VH may include information indicating the orientation of the virtual crop relative to the virtual collection unit, as well as information indicating the straight-line distance from the virtual collection unit to the virtual crop.

[0053] <<Virtual Arm Control Information VM>> The virtual arm control information VM indicates information regarding the control conditions of the virtual arm relative to the virtual crop information VH, and corresponds to the arm control information. The virtual arm control information VM includes, for example, the degree of drive of the virtual joint for moving the tip of the virtual arm (e.g., motor torque). Torque may include, for example, the magnitude of the force, as well as, for example, the amount of change in the magnitude of the force and the reaction force.

[0054] For example, the virtual arm control information VM may include parameters for a virtual harvesting unit that is driven to collect virtual crops. The virtual arm control information VM includes, for example, driving conditions for bringing the tip of the virtual arm close to a virtual fruit, based on virtual position information. The driving conditions include information for controlling multiple virtual joints so that, for example, the tip of the virtual arm passes through a path that avoids contact with virtual branches and leaves.

[0055] <<Harvest Reward Ch>> The harvest reward Ch is set based on the result of the virtual arm unit executing the virtual arm unit control information VM. The virtual arm unit control information VM is identified, for example, based on the virtual crop information VH. For example, if the result is close to ideal, a high harvest reward Ch is set, and if the result deviates from ideal, a penalty is set as the harvest reward Ch.

[0056] The harvest reward Ch is used when updating the policy. The policy is updated, for example, based on multiple harvest reward Chs, and constructed as an optimal function.

[0057] For example, if the virtual arm control information VM includes the torque of the virtual joint, the harvesting reward Ch may be set based on a comparison between a preset threshold and the torque. For example, if the torque does not exceed the threshold, a high harvesting reward Ch is set, and if the torque exceeds the threshold, a penalty is set as the harvesting reward Ch.

[0058] For example, if the virtual arm control information VM includes driving conditions, the harvest reward Ch is set based on whether or not the position of the tip of the virtual arm enters the harvestable area of ​​the virtual crop. For example, if the position of the tip of the virtual arm enters the harvestable area, a high harvest reward Ch is set, and if the position of the tip of the virtual arm does not enter the harvestable area, a penalty is set as the harvest reward Ch.

[0059] In the constructed harvesting learning model HM, for example, imaging information generated from the imaging unit 40 and sensor information obtained from the main sensor unit 140 (described later) can be used as explanatory variables corresponding to the virtual crop information VH. In addition, control information of the arm unit 10, which corresponds to the virtual arm unit control information VM, can be obtained by inputting explanatory variables into the harvesting learning model HM.

[0060] For example, the control unit 30 uses the shape and positional relationship of the crop 5 obtained from sensor information as explanatory variables for the harvesting learning model HM, and acquires control information for multiple joints 11 as the objective variable. Then, the control unit 30 uses the acquired control information to control the multiple joints 11 so that the arm 10 approaches the crop 5. As a result, the crop harvesting device 1 can bring the arm 10 close to the crop 5 while avoiding contact with branches and leaves, and can safely collect the crop 5.

[0061] <Main body 100> The main body 100 refers to any housing or the like on which the arm 10 can be fixed. The main body 100 may also refer to a four-legged or more robot that walks on a walking surface 9 including uneven terrain, as shown in Figure 4. The main body 100 can automatically walk to a destination set on uneven terrain such as mountainous areas or unpaved areas. The main body 100 can, for example, automatically move to crops 5 cultivated at the destination.

[0062] The main body 100 includes, for example, a torso 110 and a plurality of legs 120, as shown in Figures 4 and 5. The main body 100 represents, for example, a known quadruped walking robot using a torso 110 and four legs 120. The main body 100 may be provided with, for example, one or more main body sensor units 140 included in the sensor unit.

[0063] For example, the control unit 30 controls multiple leg units 120 based on a pre-constructed walking learning model WM. The control unit 30 is housed, for example, in the torso unit 110. The control unit 30 can control the main body unit 100 using known control techniques. For example, the central control unit can control the main body unit 100 by executing a program stored in non-volatile memory using volatile memory as a work area.

[0064] <Body section 110> The torso section 110 represents the core part of the main body section 100. The torso section 110 may have a shape such as a rectangular parallelepiped or a plate, and any shape can be used depending on the application. Steel can be used as the material for the torso section 110, and any material that can withstand outdoor use can be used. The torso section 110 is connected to the arm section 10.

[0065] The fuselage 110 is, for example, a housing that houses the control unit 30 and electronic equipment such as batteries for driving each structure. The fuselage 110 has dimensions such as approximately 500 mm to 2000 mm on each side, and can be arbitrarily set according to the application.

[0066] <Legs 120> Multiple leg sections 120 contact the walking surface 9 and support the torso section 110. Four or more leg sections 120 are provided, and any number can be provided depending on the application. Each leg section 120 has, for example, one or more joints. By using joints, it is possible to support the torso section 110 in a posture suitable for the characteristics of the walking surface 9.

[0067] The leg portion 120 includes, for example, a support portion 121, a motor portion 122, a protective portion 123, and a connecting portion 124, as shown in Figure 6(a). For example, the legs of a known walking robot may be used as the leg portion 120.

[0068] The support portion 121 includes a contact portion 121a that contacts the walking surface 9. The motor portion 122 is connected to the torso portion 110 and drives the support portion 121. The protective portion 123 is provided between the support portion 121 and the motor portion 122 and has, for example, a cavity. The connecting portion 124 is connected to the support portion 121 via a joint portion 125, extends into the cavity of the protective portion 123 and is connected to the motor portion 122. The connecting portion 124 drives the support portion 121 via the motor portion 122.

[0069] The motor unit 122 includes, for example, three motors 122a, 122b, and 122c. The first motor 122a can provide the support unit 121 and the protective unit 123 with a first rotation r around a first axis X along the walking direction, as shown in, for example, Figure 6(b). This makes it possible to change, for example, the height position and parallelism of the torso unit 110 with respect to the walking surface 9.

[0070] The second motor 122b can provide the support portion 121 and the protective portion 123 with a second rotation p1 around a second axis Y1 in a direction intersecting the walking direction, as shown in Figure 6(c), for example. This makes it possible to change, for example, the height position and degree of parallelism of the torso portion 110 with respect to the walking surface 9.

[0071] The third motor 122c can impart a third rotation p2 to the support portion 121, for example, as shown in Figure 6(d), around the third axis Y2 which passes through the joint portion 125 and is parallel to the second axis Y1. This allows, for example, the main body portion 100 to walk.

[0072] <Main unit sensor section 140> The main sensor unit 140 is provided, for example, on the torso unit 110, or in any location depending on the application, and the number of units provided is arbitrary. The main sensor unit 140 includes, for example, an imaging device such as a camera, as well as known sensors that can measure the state of the main unit 100, such as an inertial measurement unit (IMU), an obstacle detection sensor (LiDAR (Light Detection and Ranging), an ultrasonic sensor, etc.). The main sensor unit 140 acquires, for example, the collection distance data described above. In this case, for example, the main sensor unit 140 may acquire collection distance data from the collection unit 20 to the crop 5 based on the results of measuring the distance from the main sensor unit 140 to the crop 5 and the distance from the main sensor unit 140 to the collection unit 20. When acquiring collection distance data, the control unit 30 may perform some of the calculations. Also, when acquiring collection distance data, imaging information obtained by the imaging unit 40 may be used.

[0073] The main sensor unit 140 is used, for example, to measure the position of the crop harvesting device 1, and can utilize, for example, RTK (Real-Time Kinematic) GPS (Global Positioning System) or satellite systems (GLONASS (Global Navigation Satellite System), Galileo, etc.). The main sensor unit 140 may include, for example, at least one of a center of gravity measurement unit and an altitude measurement unit.

[0074] The control unit 30 may also control each component (for example, the motor unit 122 or the arm unit 10) using sensor information obtained from the main unit sensor unit 140, depending on the application. In this case, multiple control commands corresponding to the contents of multiple sensor information can be stored in the storage unit, and a control command can be selected according to the contents of the sensor information to control each component.

[0075] The control unit 30 may perform efficient route selection based on sensor information obtained from, for example, the main unit sensor unit 140. For example, route selection may involve A *Algorithms and known techniques such as Dijkstra's algorithm can be used. In this case, the control unit 30 can select a route using elevation data and topographic information (slope, presence or absence of obstacles, etc.) that have been stored in advance in the storage unit or the like.

[0076] <<Center of Gravity Measurement Unit>> The main sensor unit 140 includes, for example, a center of gravity measurement unit. The center of gravity measurement unit acquires center of gravity data indicating the center of gravity G of the crop harvesting device 1, including, for example, the arm unit 10 and the main body unit 100. The center of gravity measurement unit is provided, for example, in the torso unit 110. When the main sensor unit 140 includes a center of gravity measurement unit, the control unit 30 uses a walking learning model WM to derive, for example, the torque of the motor unit 122 according to the center of gravity data and controls the motor unit 122.

[0077] For example, as shown in Figure 7(a), the control unit 30 derives a support region R1 enclosed by the positions where each of the multiple contact parts 121a contacts the walking surface 9, and identifies the position of the center of gravity G relative to the support region R1. The support region R1 may be identified, for example, by the degree of torque acting on each motor part 122, or based on the results of imaging the contact positions between the walking surface 9 and the contact parts 121a using an imaging device or the like. The position of the center of gravity G indicates, for example, the position obtained by projecting the center of gravity G onto the walking surface 9. For example, a range R2 contained within the support region R1 may be set in advance, and it may be identified whether the position of the center of gravity G falls within range R2. The control unit 30 may evaluate the position of the center of gravity G based on two-dimensional coordinates or based on three-dimensional coordinates.

[0078] For example, as shown in Figure 7(b), if the position of the center of gravity G does not fall within the range R2, the control unit 30 controls the motor unit 122 so that the position of the center of gravity G falls within the range R2 (for example, Figure 7(c)). This makes it possible to control the leg unit 120 while taking the position of the center of gravity G into consideration.

[0079] For example, the control unit 30 may evaluate the positional relationship between a preset center of gravity range and the center of gravity data that fluctuates in accordance with the control of the arm unit 10. In this case, the control unit 30 may control either the leg unit 120 or the arm unit 10 based on the result of comparing the center of gravity range and the center of gravity data.

[0080] For example, if the support area R1 described above is set as the center of gravity range, the control unit 30 evaluates the positional relationship between the center of gravity data acquired by the center of gravity measurement unit and the support area R1. For example, if the position of the center of gravity G is outside the support area R1, the control unit 30 controls the leg portion 120 so that the position of the center of gravity G is included within the support area R1. Subsequently, if the position of the center of gravity G changes from a position outside the support area R1 to a position within the safe range contained within the support area R1 (for example, the range R2 described above), the control unit 30 controls the arm portion 10 instead of the leg portion 120. This process makes it possible to suppress the risk of the crop harvesting device 1 tipping over due to the control of the arm portion 10 and to improve work efficiency.

[0081] <<Altitude Measurement Unit>> The main sensor unit 140 includes, for example, an altitude measurement unit. The altitude measurement unit is provided on the torso unit 110 and measures the height position of the torso unit 110 with respect to the walking surface 9. When the main sensor unit 140 includes an altitude measurement unit, the control unit 30 uses a walking learning model WM to derive the torque of the motor unit 122 according to the height position and controls the motor unit 122.

[0082] For example, the control unit 30 compares a preset threshold with the height position. If the height position exceeds the threshold, the control unit 30 controls the motor unit 122 so that the height position becomes below the threshold. This enables optimal control of the leg unit 120 for the state of the torso unit 110.

[0083] <Walking Learning Model WM> The gait learning model WM represents a learning model constructed using reinforcement learning in a virtual environment. In the gait learning model WM, state information is used as the explanatory variable, and leg control information is used as the objective variable.

[0084] <<Status Information>> The status information indicates information that may affect the control of the leg portion 120, and includes information about the crop harvesting device 1 as well as information about the external environment. The status information includes arm portion information regarding the characteristics of the arm portion 10, and may include, for example, environmental information. The status information includes, for example, center of gravity data. The center of gravity data may be used in conjunction with, for example, arm portion information.

[0085] The arm information includes, for example, the control results of the arm 10 and its characteristics in a stationary state, such as the state after controlling multiple joints 11, etc., with specific parameters. The arm information also includes, for example, the center of gravity of the arm 10, as well as information such as the orientation of the tip 12 of the arm 10 and the degree of extension of the arm 10.

[0086] The arm information may include, for example, collection distance data indicating the distance between the crop 5 and the collection unit 20. The collection distance data indicates, for example, the shortest distance from the tip of the collection unit 20 to the crop 5.

[0087] The arm information may include, for example, information about the operation of the arm 10, such as angular velocity and movement speed. The information about the operation of the arm may be data equivalent to sensor information such as angular velocity (e.g., roll ratio, pitch ratio, etc.) and tilt acquired using an inertial measurement unit.

[0088] Environmental information includes surface conditions such as the slope and altitude of the walking surface 9, and information about topographic features such as slope and elevation differences. Environmental information may also include information about obstacles such as stones, wood, and puddles, and information about the environment such as sunny, rainy, and snowy conditions.

[0089] <<Leg Control Information>> The leg control information indicates information regarding the control conditions of multiple legs 120 in relation to the state information. The leg control information includes, for example, the torque of the motor unit 122 for driving each leg 120. The torque may include, for example, the magnitude of the force, as well as, for example, the amount of change in the magnitude of the force and the reaction force.

[0090] The leg control information includes, for example, a target walking position to which the crop harvesting device 1 should move when the leg 120 is driven. The target walking position indicates, for example, the difference between the position of the crop harvesting device 1 before the leg 120 is driven and the position of the crop harvesting device 1 after the leg 120 is driven. In addition to the above, the target walking position may also indicate the difference between a target trajectory to a pre-set destination and the current position. As the target walking position, data is used that is expected to be calculated based on sensor information obtained from, for example, GPS, an inertial measurement unit, an obstacle detection sensor, etc., and is represented, for example, as a vector.

[0091] The leg control information may include, for example, the positional relationship between the center of gravity range and the center of gravity position, and may also include control conditions for the leg 120 to achieve a preset positional relationship. The center of gravity range may indicate, for example, the support area R1 described above, or the range R2 described above. The center of gravity position indicates the center of gravity G of the crop harvesting device 1 and corresponds to the center of gravity data measured by the center of gravity measurement unit.

[0092] The leg control information may include, for example, the ground contact pattern of the leg 120 (e.g., a pattern where the front leg is lifted and the rear leg is in contact with the ground). As the ground contact pattern, data may be used that is intended to be identified by, for example, the degree of torque controlling the motor section 122 of each leg 120, or data may be used that is intended to be identified based on the results of imaging the contact position between the walking surface 9 and the contact section 121a using an imaging device or the like.

[0093] In the virtual environment, a virtual device modeled after the crop harvesting device 1 is set up, and a virtual walking surface modeled after the walking surface 9 may also be set up. The virtual device represents a virtual model with functions equivalent to the crop harvesting device 1 set up in the virtual environment, and may represent a virtual model set up under the same conditions as the virtual model used when constructing the harvesting learning model HM.

[0094] The virtual device represents a virtual model in the virtual environment that has functions equivalent to the main unit 100, and may also represent a virtual model that has functions equivalent to both the main unit 100 and the arm unit 10. The virtual walking surface corresponds to the walking surface 9 on which the crop harvesting device 1 walks in the real environment, and any virtual uneven terrain can be set according to the application. The walking learning model WM is constructed using known electronic equipment capable of performing reinforcement learning.

[0095] Reinforcement learning employs the same learning methods as when constructing the HM harvesting model described above, and is implemented using, for example, Isaac Lab or Isaac Sim (both from NVIDIA). Reinforcement learning derives the Q function and policy using, for example, data from pre-prepared virtual devices. The Q function is derived using, for example, an offline reinforcement learning algorithm such as CQL (Conservative Q-Learning). The policy provides guidelines (functions) for the virtual device to select the optimal action for the virtual environment.

[0096] In reinforcement learning used to construct a walking learning model (WM), a policy, virtual state information (VA), virtual leg control information (VB), and walking reward (Cw) are employed. The learning method involves using the aforementioned policy to determine the virtual leg control information (VB) that is suitable for the virtual state information (VA). Then, based on the virtual leg control information (VB), the virtual device executes the movement of the virtual legs. A walking reward (Cw) is then set based on the result of the virtual leg movement, and the policy is updated based on the walking reward (Cw). By repeating the above process, the policy is optimized, and the walking learning model (WM) is constructed.

[0097] <<Virtual State Information VA>> Virtual state information VA indicates information that may affect the control of the virtual leg, and includes information about the virtual device as well as information about the external environment. Virtual state information VA includes virtual arm information corresponding to arm information, and may also include virtual environment information corresponding to environment information, for example.

[0098] The virtual state information VA includes, for example, virtual center of gravity data, which corresponds to the center of gravity data. The virtual center of gravity data may be used in conjunction with, for example, virtual arm information.

[0099] Virtual arm information includes, for example, the control results of the virtual arm and its characteristics in a stationary state, such as the state after controlling multiple virtual joints with specific parameters. Virtual arm information also includes, for example, the center of gravity of the virtual arm, as well as information such as the orientation of the tip of the virtual arm and the degree of extension of the virtual arm.

[0100] The virtual arm information may include, for example, the results of controlling the virtual arm using the virtual arm control information VM used when constructing the harvesting learning model HM. In this case, when constructing the walking learning model WM, the learning data of the harvesting learning model HM can be used, thus reducing the amount of learning data to be prepared and lowering costs. Furthermore, the control results of the arm 10 using the harvesting learning model HM can be easily used to control the leg 120 using the walking learning model WM. As a result, there is no need to acquire unnecessary information when controlling the leg 120, which improves processing efficiency and accuracy.

[0101] The virtual arm information may include, for example, virtual collection distance data corresponding to the distance between the crop 5 and the collection unit 20. The virtual collection distance data indicates, for example, the shortest distance from the tip of the virtual collection unit to the virtual crop.

[0102] The virtual arm information may include, for example, information about the operation of the virtual arm, such as angular velocity and movement speed. The information about the operation of the virtual arm may be data equivalent to sensor information such as angular velocity (e.g., roll ratio, pitch ratio, etc.) and tilt acquired using an inertial measurement unit.

[0103] Virtual environment information includes information about surface conditions such as the slope and altitude of the virtual walking surface, and information about topographic features such as slope and elevation differences. Virtual environment information may also include information about obstacles such as stones, wood, and puddles, and information about the environment such as sunny, rainy, and snowy conditions.

[0104] <<Virtual Leg Control Information VB>> The virtual leg control information VB represents information about the control conditions of multiple virtual legs relative to the virtual state information VA, and corresponds to leg control information. The virtual leg control information VB includes, for example, the torque of the virtual motor for driving each virtual leg. The torque may include, for example, the magnitude of the force, as well as, for example, the change in the magnitude of the force and the reaction force.

[0105] The virtual leg control information VB includes, for example, the assumed walking position where the virtual device is expected to move due to the driving of the virtual leg. The assumed walking position indicates, for example, the difference between the position of the virtual device before the driving of the virtual leg and the position of the virtual device after the driving of the virtual leg. In addition to the above, the assumed walking position may also indicate the difference between the target trajectory for moving toward a pre-set destination and the current position. As the assumed walking position, data is used that is expected to be calculated based on sensor information obtained from, for example, GPS, inertial measurement units, obstacle detection sensors, etc., and is represented, for example, as a vector.

[0106] The virtual leg control information VB may include, for example, the positional relationship between the virtual center of gravity range and the virtual center of gravity position, and may also include control conditions for the virtual leg to realize a pre-set positional relationship. The virtual center of gravity range may, for example, indicate the support area R1 area described above, or the range R2 described above. The virtual center of gravity position indicates the center of gravity of the virtual device and corresponds to the target value of the center of gravity data measured by the center of gravity measurement unit.

[0107] The virtual leg control information VB may include, for example, the ground contact pattern of the virtual leg (e.g., a pattern where the front leg is lifted and the rear leg is in contact with the ground). As the ground contact pattern, data may be used that is intended to be identified by, for example, the degree of torque controlling the motor section 122 of each leg 120, or data may be used that is intended to be identified based on the results of imaging the contact position between the walking surface 9 and the contact section 121a using an imaging device or the like.

[0108] <<Walking Reward Cw>> The walking reward Cw is set based on the result of the virtual device executing the virtual leg control information VB. The virtual leg control information VB is identified, for example, based on the virtual state information VA. For example, if the result is close to ideal, a high walking reward Cw is set, and if the result deviates from ideal, a penalty is set as the walking reward Cw.

[0109] The walking reward Cw is used when updating the policy. The policy is updated, for example, based on multiple walking reward Cws and constructed as an optimal function.

[0110] For example, if the virtual leg control information VB includes the torque of a virtual motor, a walking reward Cw may be set based on a comparison between a pre-set threshold and the torque. For example, if the torque does not exceed the threshold, a high walking reward Cw is set, and if the torque exceeds the threshold, a penalty is set as the walking reward Cw.

[0111] For example, if the virtual leg control information VB includes the assumed walking position, the walking reward Cw is set based on the result of calculating the dot product of the position vector and the target vector. In addition to the above, the walking reward Cw may also be set based on the result of calculating the difference between the target trajectory and the current position. For example, if the calculated difference is less than or equal to a preset threshold, a high walking reward Cw is set to achieve emphasis on the amount of progress (e.g., increased torque), and if the calculated difference exceeds the threshold, a penalty is set as the walking reward Cw.

[0112] For example, if the virtual leg control information VB includes positional relationships, the walking reward Cw is set based on whether or not the virtual center of gravity position falls within the virtual support area (or a pre-set virtual range included within the virtual support area). For example, if the virtual center of gravity position falls within the virtual support area, a high walking reward Cw is set, and if the virtual center of gravity position does not fall within the virtual support area, a penalty is set as the walking reward Cw.

[0113] For example, if the virtual leg control information VB includes state information, the walking reward Cw may be set based on the result of comparing the state information with a pre-set threshold. For example, if the state information does not exceed the threshold, a high walking reward Cw is set, and if the state information exceeds the threshold, a penalty is set as the walking reward Cw.

[0114] For example, if the virtual leg control information VB includes a ground contact pattern, a walking reward Cw may be set based on the result of comparing the ground contact pattern with a pre-set reference ground contact pattern. For example, if the ground contact pattern matches or is similar to the reference ground contact pattern, a high walking reward Cw is set, and if the ground contact pattern is different from the reference ground contact pattern, a penalty is set as a walking reward Cw for an improper contact state.

[0115] (Embodiment: Method for harvesting agricultural products) Next, an example of a crop harvesting method in this embodiment will be described. Figure 8 is a flowchart of an example of a crop harvesting method. The crop harvesting method can be carried out using the crop harvesting device 1 described above. The crop harvesting method comprises a first modification step S110 and a second modification step S120. The crop harvesting method may also include, for example, a walking step S101 and a collection step S102 before the first modification step S110.

[0116] <Walking step S101> The walking step S101 controls the legs 120 to walk on the walking surface 9. For example, the control unit 30 acquires data regarding the location of the crops 5 to be collected. The control unit 30 then controls the legs 120 to reach the location of the crops 5 from the current position. In this case, for example, the control unit 30 controls the legs 120 using a walking learning model WM. The control unit 30 refers to the walking learning model WM, uses state information as an explanatory variable, outputs (acquires) leg control information as an objective variable, and controls the legs 120 using the acquired leg control information.

[0117] The walking step S101 may acquire, for example, crop information regarding the characteristics of crop 5 as state information. Crop information may be identified based on imaging information generated by imaging by the imaging unit 40, or based on information generated by the distance measuring device and imaging device included in the main sensor unit 140. For example, the walking step S101 may be set to walk toward a predetermined specific position, or to walk until a specific crop 5 is identified based on imaging information, etc.

[0118] Furthermore, if the walking surface 9 is level, for example, the walking step S101 can make the crop harvesting device 1 walk using known robotic walking technology. On the other hand, if the walking surface 9 is uneven, for example, the walking step S101 can reduce the risk of falling by controlling it using leg control information based on the arm state information described above.

[0119] <Collection Step S102> The collection step S102 controls the arm unit 10 to collect the crop 5. The collection step S102 is performed after confirming the crop 5 using imaging information acquired in the walking step S101, for example. For example, the control unit 30 brings the arm unit 10 closer to the crop 5. Along with the control of the arm unit 10, the control unit 30 acquires center of gravity data and collection distance data via the main unit sensor unit 140. The collection distance data may also be acquired via the imaging unit 40, for example.

[0120] For example, the control unit 30 identifies control conditions for the arm unit 10 that are suitable for the characteristics of the crop 5, using a harvest learning model HM. In this case, the control unit 30 uses crop information as an explanatory variable and acquires arm unit control information as an objective variable based on the harvest learning model HM, and controls the arm unit 10 using the acquired arm unit control information.

[0121] For example, the control unit 30 compares a preset center of gravity range (for example, the support area R1 described above) with the center of gravity data acquired via the center of gravity measurement unit. If the center of gravity data falls within the center of gravity range, the control unit 30 continues to control the arm unit 10.

[0122] <First modification step S110> In the first modification step S110, if the center of gravity data falls outside the center of gravity range due to the control of the arm unit 10, the control of the arm unit 10 is changed to the control of multiple legs 120. For example, the control unit 30 uses the center of gravity data and the collected distance data as explanatory variables for the walking learning model WM, acquires leg control information as the objective variable, and controls the legs 120 using the acquired leg control information. In addition, in conjunction with the control of the legs 120, the main unit sensor unit 140 acquires the center of gravity data and the collected distance data, and the control unit 30 uses the acquired center of gravity data and the collected distance data to acquire leg control information again and control the legs 120.

[0123] For example, the control unit 30 controls the legs 120 so that the center of gravity data is included in the safety range. The safety range indicates, for example, a range in which the possibility of the crop harvesting device 1 tipping over can be reduced. For example, in addition to the support area R1 described above, a range R2 or the like may be used as the safety range. In particular, by setting the safety range to a range that is included within the center of gravity range, it is easier to reduce the possibility of the crop harvesting device 1 tipping over. If, for example, the center of gravity data is outside the safety range, the control unit 30 continues to control the multiple legs 120.

[0124] For example, the control unit 30 may control the legs 120 such that the center of gravity data is included in the safety range and that fluctuations in the collected distance data do not exceed a preset threshold. In this case, the possibility of the crop harvesting device 1 tipping over can be suppressed while preventing it from moving away from the crops 5. This makes it possible to suppress the risk of the crop harvesting device 1 tipping over without reducing the efficiency of harvesting crops.

[0125] In the above case, the construction of a walking learning model (WM) can be achieved by setting a high walking reward Cw when the conditions are met based on a pre-set threshold, and setting a penalty as the walking reward Cw when the variation in the collected distance data exceeds the threshold.

[0126] <Second modification step S120> The second modification step S120, after the first modification step S110, changes the control of the legs 120 to the control of the arm 10 if the center of gravity data is included within the safety range due to the control of the multiple legs 120. After changing to the control of the arm 10, for example, the collection step S102 is performed. In this case, the control of the arm 10 can be performed with a more stable center of gravity G of the crop harvesting device 1 compared to before performing the first modification step S110 and the second modification step S120.

[0127] After performing each of the steps described above, the crop harvesting method is complete. Note that each of the steps S101, S102, S110, and S120 may be repeated multiple times.

[0128] According to this embodiment, the control unit 30 controls the leg portion 120 based on a walking learning model WM constructed using a virtual environment. The walking learning model WM represents a learning model constructed by reinforcement learning using virtual state information VA, which includes virtual arm portion information corresponding to the characteristics of the arm portion 10, virtual leg portion control information VB, and walking reward Cw. As a result, optimal control conditions for the leg portion 120 can be set based on the characteristics of the arm portion 10, and damage to the collection unit 20 due to contact with obstacles, for example, can be suppressed. This makes it possible to reduce the load on the collection unit 20.

[0129] Furthermore, according to this embodiment, the control unit 30 uses center of gravity data and collection distance data as explanatory variables for the walking learning model WM, acquires leg control information as the objective variable, and controls the leg 120 using the leg control information. As a result, the balance of the crop harvesting device 1 can be maintained during the crop collection process 5, and the position of the center of gravity G, which fluctuates with control to bring the arm 10 closer to the crop 5, can be adjusted by controlling the leg 120. This makes it possible to suppress the risk of the crop harvesting device 1 tipping over.

[0130] Furthermore, according to this embodiment, the control unit 30 uses crop information as an explanatory variable and arm control information as an objective variable based on the harvesting learning model HM, and controls the arm 10 using the arm control information. In other words, the control of the arm 10 is performed using a different learning model than that of the leg 120. Therefore, compared to the case where the arm 10 and leg 120 are controlled based on a single learning model, the processing time when controlling the arm 10 or leg 120 can be shortened. This makes it possible to improve work efficiency.

[0131] Furthermore, according to this embodiment, the control unit 30 controls either the leg portion 120 or the arm portion 10 based on the result of comparing a preset center of gravity range with the center of gravity data. Therefore, even if the center of gravity G becomes too biased due to excessive movement of the arm portion 10, it can be adjusted through the control of the leg portion 120. This makes it possible to further suppress the risk of the crop harvesting device 1 tipping over. In addition, since it is not necessary to use the walking learning model WM and the harvesting learning model HM when determining the control of either the leg portion 120 or the arm portion 10, the cost of constructing each learning model can be reduced. In addition to the above, the processing time when determining the control of either the leg portion 120 or the arm portion 10 can be shortened, making it possible to improve processing capacity.

[0132] Furthermore, according to this embodiment, the movable part 22b is driven in conjunction with the rotation of the rotating part 21c, causing the position of the holding part 22a to change. Therefore, when the support portion of the crop 5 is cut, the crop 5 can be easily held by the holding part 22a. This makes it possible to suppress deterioration of the quality of the crop 5.

[0133] Furthermore, according to this embodiment, in the first modification step S110, if the center of gravity data falls outside the center of gravity range due to the control of the arm 10, the control of the arm 10 is changed to the control of the leg 120. Therefore, the possibility of the crop harvesting device 1 tipping over due to the center of gravity data falling outside the center of gravity range can be suppressed. This makes it possible to suppress the risk of the crop harvesting device 1 tipping over when collecting crops 5.

[0134] Furthermore, according to this embodiment, the second modification step S120, after the first modification step S110, changes the control of the leg portion 120 to the control of the arm portion 10 if the center of gravity data is included in the safe range. As a result, as soon as the crop harvesting device 1 returns to a state where there is little possibility of it tipping over, the collection of crops 5 can be quickly resumed. This makes it possible to improve the efficiency of collecting crops 5.

[0135] For example, the virtual arm control information VM may include driving conditions for bringing the tip of the virtual arm closer to the virtual fruit based on virtual position information. In this case, when controlling the arm 10 using the harvesting learning model HM, it is possible to approach the fruit while avoiding contact with the branches and leaves of the fruit tree. This makes it possible to reduce the time spent collecting the crop 5.

[0136] For example, the control unit 30 may derive the positional relationship between the branches and leaves of the crop 5, such as a fruit tree, and the fruit, based on the image captured by the imaging unit 40, and control the arm unit 10 to approach the fruit based on the positional relationship. In this case, by using images, it is possible to suppress the possibility of the arm unit 10 coming into contact with obstacles such as branches and leaves, even when there are many obstacles.

[0137] For example, when the control unit 30 controls the arm unit 10, it may stop the driving of the multiple joint units 11 before driving the cutter unit 21 and the gripper unit 22. In this case, when collecting crops 5 using the cutter unit 21 and the gripper unit 22, damage to the crops 5 caused by the driving of the joint units 11 can be prevented. This makes it possible to suppress a deterioration in the quality of the collected crops 5.

[0138] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. [Explanation of Symbols]

[0139] 1: Crop harvesting device 10: Arm section 11: Joints 12:Tip 20: Collection Department 21: Cutter section 21a: Cutting section 21b: Cutting auxiliary part 21c: Rotating part 22: Gripper section 22a: Holding part 22b: Moving part 30: Control Unit 40: Imaging Unit 100: Main body 110: Torso 120: Legs 121: Support part 121a: Contact part 122: Motor section 123:Protection Department 124: Connection part 125: Joint 140: Main unit sensor section 5: Crops 9: Walking surface G: Center of gravity HM: Harvesting Learning Model WM: Learning model for walking S110: First modification step S120: Second modification step

Claims

[Claim 1] A method for harvesting crops using a crop harvesting device for collecting crops, The system comprises a first modification step and a second modification step, The aforementioned crop harvesting apparatus, An arm section including a collection unit for collecting the aforementioned agricultural products, A main body portion including multiple leg portions and connected to the arm portion, A control unit that controls the leg based on a walking learning model constructed using a virtual environment, Equipped with, The aforementioned walking learning model is Virtual state information including virtual arm information corresponding to the characteristics of the arm portion, Virtual leg control information corresponding to the control conditions of the leg, Based on the virtual state information, a walking reward is set for the result of executing the virtual leg control information, It was constructed using reinforcement learning, The sensor unit further comprises a sensor unit that acquires center of gravity data indicating the center of gravity including the arm portion and the main body portion, and collection distance data indicating the distance from the collection unit to the crop. The virtual state information includes virtual centroid data corresponding to the centroid data, The virtual arm information includes virtual collection distance data corresponding to the collection distance data, The control unit uses the center of gravity data and the collected distance data as explanatory variables for the walking learning model, acquires leg control information indicating the control conditions of the leg as the objective variable, and controls the leg using the leg control information. The control unit, based on a harvesting learning model constructed using a virtual environment, uses crop information indicating the characteristics of the crop as an explanatory variable, acquires arm control information indicating the control conditions of the arm as an objective variable, and controls the arm using the arm control information. The aforementioned learning model for harvesting, Virtual crop information corresponding to the aforementioned crop information, Virtual arm control information corresponding to the aforementioned arm control information, Based on the virtual crop information, the harvesting reward is set for the result of executing the virtual arm control information, It was constructed using reinforcement learning, The control unit controls either the leg portion or the arm portion based on the result of comparing a preset center of gravity range with the center of gravity data. The first modification step involves changing the control of the arm portion to the control of the leg portion if, in conjunction with the control of the arm portion based on the harvesting learning model, the center of gravity data falls outside the center of gravity range. The second modification step, after the first modification step, involves changing the control of the legs to the control of the arms if, in conjunction with the control of the multiple legs based on the walking learning model, the center of gravity data falls within the safety range encompassed by the center of gravity range. A crop harvesting method characterized by the following.

Images