Unmanned forklift

Abstract

To detect a deviation amount between a position and an attitude of an object pallet with high accuracy by suppressing the reduction in working speed and the increase in cost without any limitation on the kind of a pallet to be handled.SOLUTION: An image acquisition unit acquires a photographed image from an imaging device S1. A pallet type identification unit has a learning model about a combination of images of plural types of pallets and the types of the pallets and identifies a type of an object pallet P1 by inputting the photographed image of the object pallet P1 acquired by the image acquisition unit in the learning model. A pallet position and shape acquisition unit acquires position and shape data of the object pallet P1 from a distance measurement device S2 for measuring a distance to the object pallet P1. A pallet deviation amount detection unit stores position and shape data of the pallet in advance and compares the stored position and shape data corresponding to the type of the identified object pallet P1 and position and shape data of the object pallet P1.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to an unmanned forklift that is self-propelled and performs cargo handling operations. [Background technology]

[0002] There is a technology for determining whether an unmanned forklift truck is in a state where it can insert forks into insertion openings (holes 20 in Patent Document 1) of a pallet (see, for example, the first embodiment of Patent Document 1). In this technology, an unmanned forklift truck 11 is equipped with a laser range finder 18 that can move up and down together with the forks 14. The laser range finder 18 scans an opening end face 19a of a pallet 19 placed at a loading position with a laser beam, and obtains image data of the edge line of the hole 20. A pattern controller 21 then compares the obtained image data of the line of the hole 20 with image data of the line of the hole 20 previously captured under multiple conditions where the distance and inclination of the end face of the pallet 19 relative to the unmanned forklift truck 11 are changed, thereby determining whether the forks 14 are in a state where they can insert into the holes 20 in the pallet 19.

[0003] There is also a technology for unmanned forklift trucks that identifies the type of pallet and adjusts the spacing between the forks to match the spacing of the insertion openings (holes 20 in Patent Document 1) of the identified pallet (see, for example, the second embodiment of Patent Document 1). In this technology, multiple types of pallets 19 with different spacing between the holes 20 are used, and the unmanned forklift truck 11 is equipped with a fork shifter that can adjust the spacing between the pair of forks 14. The type of pallet 19 is identified by a pattern controller 21 specifying the width of the pallet 19 from image data of the edge lines of the holes 20 captured by a laser range finder 18. Then, a control device 17 controls the fork shifter to adjust the spacing between the pair of forks 14 to match the spacing of the holes 20 that corresponds to the identified width of the pallet 19.

[0004] Furthermore, there is a technology for identifying the type of pallet in an unmanned forklift (see, for example, Patent Document 2). In this technology, a pattern 01 is provided on a pallet 05, and the unmanned forklift 5 is equipped with a pair of cameras 3 and 4. The cameras 3 and 4 photograph the pattern 01 on the pallet 05. Then, a position determination means determines the position of the pallet 05 from the photographed pattern, and a comparison means compares the photographed pattern with a reference pattern to identify the type of pallet 05. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2017-19596 A [Patent Document 2] Japanese Patent Application Publication No. 9-218014 Summary of the Invention [Problem to be solved by the invention]

[0006] The type of pallet in the unmanned forklift of Patent Document 1 is identified by detecting an image of the edge line of the hole on the open end face of the pallet using laser distance measurement. Therefore, for example, if there is a cargo or another pallet right next to the pallet, it is difficult to identify the type of pallet for fixing the cargo to the pallet. Stretch film If a part of the pallet blocks a hole in the pallet, or if a foreign object enters a hole in the pallet, there is a risk that the type of pallet cannot be identified or that the type of pallet will be identified incorrectly.

[0007] The identification of pallet types in the unmanned forklift of Patent Document 2 requires affixing patterns to all pallets that are to be handled by the unmanned forklift, which increases costs. Also, when handling many different types of pallets with different shapes, a common location for affixing patterns to all of those pallets is required, which limits the types of pallets that can be handled.

[0008] On the other hand, when detecting the amount of positional deviation of multiple types of pallets using only a distance detector such as a laser scanner, it is necessary to identify the type of corresponding pallet by comparing the measured shape data acquired by the distance detector with all types of stored position and shape data of multiple types of pallets.

[0009] Therefore, due to the characteristics of the distance detector, which measures using only distance data, if the number of types of pallets to be measured increases, there is an increased possibility that items other than pallets placed on the floor (such as cardboard boxes) will be mistakenly recognized as a specific type of pallet.

[0010] Furthermore, since the calculation load for comparing and matching with shape data of multiple types of pallets is large, the calculation takes time, slowing down work speed, and using an expensive computer with high processing power increases costs.

[0011] The present invention aims to provide an unmanned forklift that is not limited in the type of pallet it can handle, can suppress a decrease in work speed and an increase in costs, and can detect with high accuracy the amount of deviation in the position and posture of a target pallet from the normal position of the pallet, thereby enabling the required loading and unloading operations to be carried out reliably. [Means for solving the problem]

[0012] In order to solve the above problems, the unmanned forklift according to the present invention comprises: An unmanned forklift equipped with a pair of forks that automatically performs loading and unloading work using pallets, an image acquisition unit that acquires a photographed image from an image acquisition device that photographs a front side of the unmanned forklift; a pallet type identification unit that has a trained learning model that has undergone machine learning on the combination of each pallet image and the type of pallet for multiple types of pallets, and identifies the type of the target pallet by inputting the photographed image of the target pallet acquired by the image acquisition unit into the learning model; a pallet position / shape acquisition unit that acquires position / shape data of the target pallet from a distance measurement device that measures the distance to the target pallet; a pallet deviation amount detection unit that stores position and shape data of each of a plurality of types of pallets in advance, and detects deviation amounts of the position and posture of the target pallet from its normal position by comparing the stored position and shape data corresponding to the type of target pallet identified by the pallet type identification unit with the position and shape data of the target pallet acquired by the pallet position and shape acquisition unit; (A) an operation including inserting the forks into the insertion opening of the target pallet using the detected amount of deviation; (B) A stacking operation in which the pallet and cargo held by the forks inserted into the insertion openings are loaded onto the cargo of the target pallet in accordance with the amount of deviation of the target pallet; a travel control unit that controls travel to perform at least one of the above tasks; Equipped with.

[0013] With this configuration, even if the objects to be transported by the unmanned forklift are multiple types of pallets with different shapes, there is no limit to the types of pallets that can be handled, and by inputting the captured image into a trained learning model that has undergone machine learning for combinations of images of each pallet and the pallet type, the pallet type identification unit can identify the type of the target pallet with high accuracy.The pallet misalignment detection unit can then compare and match the stored position and shape data corresponding to the type of target pallet identified by the pallet type identification unit with the position and shape data of the target pallet obtained by the pallet position and shape acquisition unit.

[0014] This makes it possible to detect with high accuracy the amount of deviation in the position and posture of the target pallet relative to the pallet's normal position, and also prevents accidents such as the fork interfering with the pallet or the object due to an item (such as a cardboard box) being mistakenly recognized as a specific type of pallet. Also, the calculation time required for comparing and matching pallet shape data is shortened, improving work speed, eliminating the need for expensive computers with high processing power.

[0015] After detecting the amount of deviation of the target pallet with high accuracy, the unmanned forklift can perform the required loading and unloading operations, such as operation (A) of picking up the pallet and cargo on the floor and transporting it to another location, operation (B) of picking up the pallet and cargo on the second layer of the cargo on the first layer of the pallet and transporting it to another location, or operation (B) of stacking.

[0016] Here, the position and shape data is data on the height including the insertion opening of the pallet, In a preferred embodiment, the pallet deviation amount detection unit overlays and compares a line segment based on the stored position and shape data corresponding to the type of target pallet with a line segment based on the position and shape data of the target pallet acquired by the pallet position and shape acquisition unit.

[0017] With this configuration, the amount of deviation is determined by overlapping and comparing line segments, reducing calculation costs. Furthermore, because the height data can be obtained by simply scanning horizontally, there is no need to move the distance measuring device that measures the distance to the target pallet up and down, which shortens the time it takes to detect the amount of deviation in the position and orientation of the target pallet.

[0018] In a preferred embodiment, the apparatus further comprises a deviation error determining section that determines an error when the deviation detected by the pallet deviation detection section exceeds a predetermined threshold value.

[0019] According to this configuration, if the amount of deviation is too large to insert the forks into the target pallet, an error determination is made before the fork insertion operation, thereby preventing accidents caused by interference between the target pallet and the forks. For example, the threshold amount of deviation that serves as the basis for the error determination is set to the maximum value of the range in which the unmanned forklift can insert the forks into the target pallet while self-propelled from the predetermined stopping position at which the unmanned forklift is stopped when the distance measuring device measures the distance to the target pallet and the pallet position / shape acquisition unit acquires position / shape data of the target pallet.

[0020] Furthermore, in a preferred embodiment, the device further includes a pallet shape determination unit that compares the shape data of the target pallet acquired by the pallet position / shape acquisition unit with pre-stored shape data of a pallet of the same type as the target pallet, and if the difference between the two is greater than a predetermined threshold, determines that there is an error in the acquired shape data of the target pallet.

[0021] With this configuration, even if another item (such as a cardboard box) is placed right next to the target pallet and the size of the target pallet is incorrectly detected due to the influence of the other item, the error can be determined by comparing the shape of the target pallet with the stored pallet shape corresponding to the type of target pallet identified by the pallet type identification unit. This prevents accidents in which the forks interfere with the insertion opening of the target pallet and work is stopped.

[0022] Furthermore, a preferred embodiment further includes a pallet stacking determination unit that determines whether or not the target pallet is stacked by performing image processing on the photographed image of the target pallet acquired by the image acquisition unit and calculating the number of pallets within a predetermined specific range where the target pallet is present.

[0023] With this configuration, by calculating only the number of pallets within a specific range in front of the unmanned forklift, it is possible to reliably determine whether pallets are stacked without erroneously detecting pallets other than the target pallet. For example, there is no erroneous detection in which a non-target pallet placed to the side of the target pallet on the floor is detected and the number of pallets is calculated as two, so there is no erroneous determination that stacking is not possible when stacking is possible.

[0024] In a preferred embodiment, the pallet stacking determination unit is configured to perform machine learning in advance for pallets stacked in multiple layers, using pallet images as training data.

[0025] According to this configuration, when the surrounding images of a pallet differ depending on which tier the pallet is on, each different surrounding image is learned as training data, thereby improving the accuracy of pallet detection.

[0026] Furthermore, in a preferred embodiment, the device further includes a fork width change device that changes the spacing between the forks to match the spacing of the insertion openings of the target pallet depending on the type of the target pallet obtained from the pallet type identification unit.

[0027] With this configuration, there is no need to prepare an unmanned forklift for each target pallet with a different insertion hole spacing. By operating the fork width change device to adjust the fork spacing to correspond to the insertion hole spacing, multiple types of pallets can be transported with a single unmanned forklift.

[0028] Furthermore, a distance sensor disposed at the tip of the fork for detecting the presence or absence of an object ahead; a presence sensor that detects whether the forks are holding the pallet; Furthermore, Using the result of detecting the presence or absence of the object by the distance sensor and the result of detecting whether the pallet is being held by the inventory sensor, Stacking operations in which a second pallet and load are placed on top of the first pallet load, and / or In a preferred embodiment, a de-stack operation is performed in which the second tier pallet and its load on the first tier pallet are picked up and transported to another location.

[0029] With this configuration, when stacking, the distance sensor can detect the height of the load on the first tier of pallets, and the load presence sensor can detect the height at which the forks are withdrawn after the stacking operation. Also, when unstacking, the distance sensor can detect the height of the insertion opening of the second tier of pallets, and the load presence sensor can detect the state in which the forks are holding the pallet and load to be held during the unstacking operation. Therefore, the distance sensor and the load presence sensor enable reliable stacking and / or unstacking operations.

[0030] In addition, in a preferred embodiment, the photographing device photographs the target pallet at a first stop position and the pallet type identification unit identifies the type of the target pallet, and the distance measuring device measures the distance to the target pallet at a second stop position closer to the target pallet than the first stop position and the pallet position / shape acquisition unit acquires position / shape data of the target pallet.

[0031] With this configuration, the distance measuring device measures the distance to the target pallet at the second stop position, which is closer to the target pallet than the first stop position, and the pallet position and shape acquisition unit acquires position and shape data of the target pallet, so that the pallet deviation amount detection unit can detect the amount of deviation in the position and posture of the target pallet more accurately. [Effects of the Invention]

[0032] As described above, the unmanned forklift of the present invention has no restrictions on the types of pallets it can handle, can suppress a decrease in work speed and an increase in costs, and can detect with high accuracy the amount of deviation in the position and posture of the target pallet relative to the pallet's normal position, thereby ensuring the required loading and unloading operations. [Brief explanation of the drawings]

[0033] [Figure 1] 1 is a perspective view of an unmanned forklift according to an embodiment of the present invention. [Figure 2] 1 is a left side view of the unmanned forklift truck holding a pallet and a load, showing a cross section of the pallet with the forks inserted into the insertion openings. FIG. [Figure 3] FIG. 2 is a perspective view showing a pallet and a load on a floor surface, and the unmanned forklift truck. [Figure 4] FIG. 2 is a block diagram showing an outline of the device configuration. [Figure 5A] FIG. 10 is an enlarged cross-sectional view of a main portion showing a sensor disposed at the tip of the right fork. [Figure 5B] FIG. 10 is an enlarged longitudinal cross-sectional view of a main portion showing a sensor disposed at the tip of the right fork. [Figure 6A] FIG. 10 is an enlarged view of the main part around the inventory sensor as seen from the left. [Figure 6B] This is an enlarged cross-sectional view of the main part around the inventory sensor. [Figure 7] 10 is a diagram showing an example of an image in which the palette type identification unit has identified the type of the target palette. FIG. [Figure 8A] FIG. 10 is a schematic plan view showing line segments based on stored pallet point cloud data corresponding to the type of target pallet. [Figure 8B] 8B is a schematic plan view showing a state in which the line segment in FIG. 8A is superimposed on a line segment based on point cloud data of the target pallet acquired by a pallet position / shape acquisition unit. FIG. [Figure 9] FIG. 2 is a front view of the fork width change device. [Figure 10] 10 is a schematic plan view illustrating the operation of narrowing the fork width using the fork width changing device. FIG. [Figure 11] FIG. 10 is a schematic plan view illustrating the operation of widening the fork width using the fork width changing device. [Figure 12A] FIG. 10 is an explanatory diagram of an example of stacking operation, showing a state in which an unmanned forklift approaches a target pallet and stops a predetermined distance before the target pallet. [Figure 12B] This is an explanatory diagram of an example of stacking operation, showing the state in which the forks of an unmanned forklift are raised and the height of the load on the target pallet is detected by a distance sensor, and then the forks are raised a further predetermined distance and stopped. [Figure 12C] FIG. 10 is an explanatory diagram of an example of stacking operation, showing a state in which the unmanned forklift has been advanced a predetermined distance to position a pallet and cargo to be stacked above the target pallet and cargo. [Figure 12D] This is an explanatory diagram of an example of stacking operation, showing the state at the moment when the forks are lowered and the underside of the pallet to be stacked comes into contact with the top surface of the load on the target pallet. [Figure 12E] FIG. 10 is an explanatory diagram of an example of stacking operation, showing a state in which the forks are further lowered and the load presence sensor detects that there is no load on the forks. DETAILED DESCRIPTION OF THE INVENTION

[0034] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0035] In the following embodiments, the direction from the base side to the tip side of the forks of the unmanned forklift is defined as the front, and the opposite direction is defined as the rear, and left and right are defined as looking forward.

[0036] For unmanned forklifts, the view facing the front where the forks are located is considered a front view. For the target pallet and load, the view facing the side where the fork insertion openings of the target pallet are located is considered a front view.

[0037] <Unmanned forklift> An unmanned forklift 1 according to an embodiment of the present invention, shown in the perspective view of Fig. 1, the view from the left in Fig. 2, the perspective view of Fig. 3, and the block diagram showing an overview of the equipment configuration in Fig. 4, has a self-position estimation function and performs cargo handling work by self-propelling according to destination instructions from a management device that wirelessly communicates with the unmanned forklift 1. The unmanned forklift 1 is equipped with a cargo handling device A that performs cargo handling work, a moving device B that performs traveling and turning operations, a control device C that controls the cargo handling device A and the moving device B, a fork width changing device D, a processing device 10, etc.

[0038] (Load handling equipment) The cargo handling device A has a mast M that moves up and down and tilts back and forth, forks 2 that carry a pallet P and cargo W, and a lift bracket 3 that moves up and down along the mast M and supports the forks 2. The forks 2 consist of a right fork 2R and a left fork 2L.

[0039] The fork 2 and lift bracket 3 move up and down along the inner mast 4 by means of a lift chain 7. The mast M consists of the inner mast 4, which supports the lift bracket 3 and moves up and down, and the outer mast 5, which guides the inner mast 4 so that it can move up and down. The outer mast 5 is tilted in the front-to-rear direction by a tilt cylinder 6.

[0040] (Mobile device) The transportation device B has a pair of left and right front wheels, rear wheels which are both drive wheels and steerable wheels, and a drive unit for the rear wheels.

[0041] (Control device) The control device C controls the lifting drive device of the cargo handling device A, the drive device of the moving device B, and the drive device of the tilt cylinder 6, and also has a communication device for communicating with a management device on the ground. The control device C also has an interface with the processing device 10 shown in Figure 4, as well as interfaces with the distance sensor S3 and the cargo presence sensor S4.

[0042] (Fork width change device) The fork width change device D changes the distance F between the right fork 2R and the left fork 2L.

[0043] (Processing device) 4 has an interface with the image capture device S1 and issues a command to the image capture device S1. The processing device 10 also has an interface with the distance measurement device S2 and issues a command to the distance measurement device S2.

[0044] <Palette> Pallets P, P1 shown in the left view of Figure 2 and the perspective view of Figure 3 are, for example, flat pallets and have insertion openings Q, Q for left and right forks 2L, 2R. A load W is placed on the pallets P, P1.

[0045] <Sensor group> As shown in the perspective view of Fig. 1 and the block diagram showing an outline of the equipment configuration of Fig. 4, the unmanned forklift 1 is equipped with an image capture device S1, a distance measurement device S2, a distance sensor S3, a load sensor S4, a self-position recognition sensor S5, and an obstacle detection sensor S6. The sensors S1 to S6 are arranged at positions to be described later. The control device C and processing device 10 shown in Fig. 4 are arranged inside the main body 1A of the unmanned forklift 1 as shown in Fig. 1.

[0046] (photography equipment, distance measuring equipment) As shown in Fig. 1, the camera device S1 and the distance measurement device S2 are held by a bracket J attached to a support plate I that protrudes forward from the center in the left-right direction of the front lower part of the main body 1A of the unmanned forklift 1. In other words, the camera device S1 and the distance measurement device S2 are not installed on movable members relative to the main body 1A, such as the lift bracket 3 or the forks 2, but on members that are immovable relative to the main body 1A. This holding structure for the camera device S1 and the distance measurement device S2 is also applied to the structure shown in other drawings, such as Fig. 9.

[0047] The image capturing device S1 is, for example, a monocular camera, which captures an image of the area in front of the unmanned forklift 1 at a predetermined stopping position where the unmanned forklift 1 will be stopped. The distance measuring device S2 is, for example, a 2D-LiDAR (Light Detection And Ranging) or TOF (Time Of Flight) camera, which measures the distance to a target pallet P1 (e.g., Figure 2) on the floor N at a predetermined stopping position where the unmanned forklift 1 will be stopped. If the distance measuring device S2 is a 2D-LiDAR, it emits pulsed laser light while changing its direction horizontally, detects the scattered light that is reflected and returned, and measures the distance and direction to the target object from the time it takes for the light to be reflected by the object and returned.

[0048] As shown in FIG. 4, an image captured by the image capturing device S1 is sent to the processing device 10, and data measured by the distance measuring device S2 is sent to the processing device 10.

[0049] (distance sensor) As shown in the perspective view of Fig. 1, the enlarged horizontal cross-sectional view of a main portion of Fig. 5A, and the enlarged vertical cross-sectional view of a main portion of Fig. 5B, the distance sensor S3 is disposed at the tip of the right fork 2R and fixed by a mounting plate 19. The distance sensor S3 may also be disposed at the tip of the left fork 2L. The distance sensor S3 is, for example, a distance setting type photoelectric sensor, and detects the presence or absence of an object ahead.

[0050] As shown in FIG. 4, a signal indicating the presence or absence of an object detected by the distance sensor S3 (for example, an ON / OFF signal) is sent to the control device C.

[0051] (obstacle detection sensor) As shown in the perspective view of Fig. 1 and the enlarged cross-sectional view of the main part of Fig. 5A, the obstacle detection sensors S6 are disposed on the left and right of the tips of the left and right forks 2L, 2R by mounting plates 20. The obstacle detection sensors S6 are, for example, photoelectric sensors, and detect obstacles approaching the tips of the forks 2.

[0052] A signal (for example, an ON / OFF signal) indicating the presence or absence of an obstacle detected by the obstacle detection sensor S6 is sent to the control device C. When the obstacle detection sensor S6 detects an obstacle, the control device C brings the unmanned forklift 1 to an emergency stop, for example.

[0053] (stock sensor) As shown in the view from the left in Fig. 2, the enlarged view of the main part from the left in Fig. 6A, and the enlarged cross-sectional view of the main part in Fig. 6B, the load presence sensor S4 is disposed at the base of the left and right forks 2L, 2R while being supported by a support plate 21. The load presence sensor S4 is, for example, a proximity switch, and detects whether the forks 2 are holding a pallet P.

[0054] As shown in FIG. 4, a signal (for example, an ON / OFF signal) indicating whether the forks 2 are holding the pallet P or not is sent to the control device C by the load sensor S4.

[0055] 6A and 6B, the structure around the presence sensor S4 at the base of the left fork 2L will be described. The seesaw plate 22, supported by the left-right support shaft R, is stopped by abutment plate 23 and is elastically biased by a tension coil spring 24. When the fork 2 is not holding a pallet P, the seesaw plate 22 remains stationary in the position shown by the solid line in FIG. 6A. When the fork 2 is holding a pallet P as shown in FIG. 2, the seesaw plate 22 swings so that the operating piece 22A descends and the detection piece 22B ascends, as shown by the imaginary line in FIG. 6A. As a result, the presence sensor S4 detects the detection piece 22B, making it possible to detect that the fork 2 is holding a pallet P.

[0056] (Self-position recognition sensor) As shown in the perspective view of FIG. 1, the self-location recognition sensor S5 is disposed facing upward on the upper surface of the frame K that protrudes upward from the main body 1A. In other words, the self-location recognition sensor S5 is installed on a member that is immovable relative to the main body 1A. The self-location recognition sensor S5 performs, for example, 3D imaging in all directions horizontally and with a vertical field of view of approximately 30°. The self-location recognition sensor S5 is, for example, a 3D-LiDAR, and is used for a laser SLAM (Simultaneously Localization And Mapping) type self-location estimation method. The self-location estimation of the unmanned forklift 1 may also be an image SLAM type self-location estimation method, etc.

[0057] In the unmanned forklift 1 controlled using the above-mentioned group of sensors S1 to S6, the signal cables and power supply cables of the sensors S3, S4, and S6 installed on movable members relative to the main body 1A are connected to the control device C and power supply within the main body 1A through the cable bear 8 shown in Figure 1.

[0058] <Image acquisition section> As shown in Fig. 4, the processing device 10 has an image acquisition unit 11. The image acquisition unit 11 acquires a photographed image from a photographing device S1. The photographing device S1 photographs, for example, a target pallet P1 shown in Fig. 2 or 3. The image acquisition unit 11 acquires the photographed image of the target pallet P1 from the photographing device S1.

[0059] <Pallet type identification unit> As shown in Fig. 4, the processing device 10 has a pallet type identification unit 12. The pallet type identification unit 12 has a trained learning model that has undergone machine learning on the combination of an image of each pallet P and the type of pallet P for multiple types of pallets P. The pallet type identification unit 12 identifies the type of the target pallet P1 by inputting the photographed image of the target pallet P1 acquired by the image acquisition unit 11 into the learning model.

[0060] Figure 7 shows an example of an image in which the pallet type identification unit 12 has identified the type of the target pallet. In this example, the pallet type identification unit 12 has identified the type of target pallet P1 on floor N as "small" and the type of target pallet P2 on the load of target pallet P1 as "small." The pallet type identification unit 12 can identify pallet types such as "small," "medium," and "large," and can also display the certainty (probability) of the identified pallet type on the image, such as (0.XX) and (0.△△) in Figure 7.

[0061] <Pallet position and shape acquisition section> As shown in Fig. 4, the processing device 10 has a pallet position / shape acquisition unit 13. The pallet position / shape acquisition unit 13 acquires position / shape data of the target pallet P1 from a distance measurement device S2 that measures the distance to the target pallet P1 when the unmanned forklift 1 shown in Fig. 2 or 3 is at a predetermined stopping position that is a predetermined distance E away from the target pallet P1. The position / shape data of the target pallet P1 is, for example, point cloud data of the height including the insertion opening Q of the target pallet P1.

[0062] <Pallet deviation detection unit> As shown in Fig. 4, the processing device 10 has a pallet misalignment detection unit 14. The pallet misalignment detection unit 14 stores in advance position and shape data for each of a plurality of types of pallets P. The position and shape data for each pallet P stored in advance in the pallet misalignment detection unit 14 is the position and shape data for each pallet P acquired by the pallet position and shape acquisition unit 13 when the distance measurement device S2 measures the distance to each pallet P at the predetermined stopping position of the unmanned forklift 1. The position and shape data for each pallet P is, for example, point cloud data of the height including the insertion opening Q of the pallet P.

[0063] The pallet deviation amount detection unit 14 detects the amount of deviation in the position and posture of the target pallet P1 from its normal position by comparing the stored position and shape data corresponding to the type of target pallet P1 identified by the pallet type identification unit 12, i.e., the position and shape data of the pallet P corresponding to the type of target pallet P1 among the individual pallets P previously stored as described above, with the position and shape data of the target pallet P1 acquired by the pallet position and shape acquisition unit 13.

[0064] L0 in the schematic plan view of Fig. 8A is a line segment based on point cloud data of the height including the insertion port Q of the stored pallet P corresponding to the target pallet P1 in Fig. 2 or 3, for example, for a pallet on floor N. As described above, the pallet misalignment detection unit 14 stores data of the line segment L0 as shown in Fig. 8A for each of multiple types of pallets P.

[0065] As shown in the schematic plan view of Figure 8B, the pallet deviation amount detection unit 14 overlays and compares the line segment L0 with the line segment L1 based on the point cloud data of the height including the insertion port Q of the target pallet P1 acquired by the pallet position / shape acquisition unit 13.

[0066] This pallet misalignment detection unit 14 reduces calculation costs by determining the amount of misalignment by superimposing and comparing the line segments L0 and L1. Furthermore, because the height point cloud data can be obtained by simply scanning in the horizontal direction, there is no need to move the distance measuring device S2, which measures the distance to the target pallet P1, up and down, thereby reducing the time required to detect the amount of misalignment in the position and orientation of the target pallet P1.

[0067] <Shooting position of the shooting device, distance measurement position of the distance measuring device> The position where the unmanned forklift 1 stops when the photographing device S1 photographs the target pallet P1 and the pallet type identification unit 12 identifies the type of the target pallet P1 is defined as the first stop position, and the position where the unmanned forklift 1 stops when the distance measuring device S2 measures the distance to the target pallet P1 and the pallet position / shape acquisition unit 13 acquires the position / shape data of the target pallet P1 is defined as the second stop position.

[0068] 2 and 3, positions of the unmanned forklift 1 spaced a predetermined distance E from the target pallet P1 may be set as the first stop position and the second stop position, and the first stop position and the second stop position may be the same position, or the second stop position may be closer to the target pallet P1 than the first stop position. By setting the second stop position closer to the target pallet P1 than the first stop position, the pallet deviation amount detection unit 14 can detect the deviation amount of the position and posture of the target pallet P1 with higher accuracy.

[0069] <Error detection section> 4, the processing device 10 has a deviation amount error determination unit 15. The deviation amount error determination unit 15 determines that an error has occurred when the deviation amount of the position and posture of the target pallet P1 from the correct position detected by the pallet deviation amount detection unit 14 exceeds a predetermined threshold value.

[0070] According to this type of deviation amount error determination unit 15, if the deviation amount is too large to insert the forks 2 into the target pallet P1, an error determination is made before the fork insertion operation, thereby preventing an accident in which the target pallet P1 interferes with the forks 2. For example, the deviation amount threshold that serves as the basis for the error determination is set to the maximum value within the range in which the unmanned forklift 1 can insert the forks 2 into the target pallet P1 while self-propelled from the predetermined stopping position at which the unmanned forklift 1 is stopped when the distance measurement device S2 measures the distance to the target pallet P1 and the pallet position / shape acquisition unit 13 acquires the position / shape data of the target pallet P1.

[0071] <Pallet shape determination section> 4, the processing device 10 has a pallet shape determination unit 16. The pallet shape determination unit 16 compares the shape data of the target pallet P1 acquired by the pallet position / shape acquisition unit 13 with pre-stored shape data of a pallet of the same type as the target pallet P1, and if the difference between the two is greater than a predetermined threshold, determines that there is an error in the shape data of the acquired target pallet P1.

[0072] With this type of pallet shape determination unit 16, even if the size of the target pallet P1 is incorrectly detected due to the influence of another item (such as a cardboard box) placed right next to the target pallet P1, the error can be determined by comparing it with the stored pallet shape corresponding to the type of target pallet P1 identified by the pallet type identification unit 12. This prevents accidents in which the forks 2 interfere with the insertion opening Q of the target pallet P1, causing work to stop.

[0073] <Pallet stacking determination unit> As shown in Fig. 4, the processing device 10 has a pallet stacking determination unit 17. The pallet stacking determination unit 17 processes the photographed image of the target pallet (for example, P1 and P2 in Fig. 7) acquired by the image acquisition unit 11 as shown in Fig. 7, and determines whether the target pallets P1 and P2 are stacked by calculating the number of pallets within a predetermined specific range.

[0074] For example, if the number of pallets within the specific range is two, as shown in Figure 7, it can be determined that stacking, in which a second layer of pallets and cargo transported from another location is placed on top of the first layer of pallets and cargo, is not permitted, and if the number of pallets within the specific range is one, it can be determined that stacking is permitted. Furthermore, if the number of pallets within the specific range is one, it can be determined that unstacking, in which the second layer of pallets and cargo are picked up and transported to another location, is not permitted, and if the number of pallets within the specific range is two, it can be determined that unstacking is permitted. Furthermore, if it is preset that stacking is not permitted when the number of pallets within the specific range is two or more, it can be determined that stacking is not permitted when the number of pallets within the specific range is two or more.

[0075] Such a pallet stacking determination unit 17 can reliably determine whether pallets are stacked or not, without erroneously detecting pallets other than the target pallet, by calculating only the number of pallets within a specific range in front of the unmanned forklift 1. For example, there is no erroneous detection in which a non-target pallet placed to the side of the target pallet on floor N is detected and the number of pallets is calculated as two, so there is no erroneous determination that stacking is not possible when stacking is possible.

[0076] For example, in the example image of Figure 7, the image surrounding pallet P1 on the floor is different from the image surrounding pallet P2 on the upper level. Therefore, if a large number of images including pallets on the floor are prepared and a trained learning model is used that has undergone machine learning using the pallets on the floor as training data, it may not be able to identify pallet P2 on the upper level.

[0077] Therefore, in a preferred embodiment, the pallet stacking discrimination unit 17 is one that has been machine-learned in advance using pallet images of multiple tiers of pallets as training data. With this type of pallet stacking discrimination unit 17, when the surrounding images of a pallet differ depending on which tier the pallet is on, each of the different surrounding images is learned as training data, thereby improving the accuracy of pallet detection.

[0078] <Travel control unit> As shown in Fig. 4, the control device C has a travel control unit 18. The travel control unit 18 controls the travel of the unmanned forklift 1 to perform at least one of the following tasks (A) and (B). The travel control unit 18 may be provided in a management device on the ground.

[0079] (A) A task including inserting the fork 2 into the insertion opening of the target pallet using the amount of deviation detected by the pallet deviation amount detection unit 14. (B) A stacking operation in which a pallet P and a load W held by forks 2 inserted into insertion openings Q are loaded onto the load of the target pallet in accordance with the amount of deviation of the target pallet.

[0080] For example, as shown in Figure 3, the unmanned forklift 1 picks up a target pallet P1 and a load W on floor N and transports them to another location. Also, as shown in Figure 2, the unmanned forklift 1 stacks a pallet P brought from another location on top of the load W on the target pallet P1 on floor N. Furthermore, as shown in Figure 7, the unmanned forklift 1 picks up a target pallet P2 and a load W on the load W on the target pallet P1 on floor N and transports them to another location.

[0081] <Example of fork width change device structure> As shown in the front view of Figure 9, the right fork 2R and the left fork 2L are guided by linear guides 25, 26 so that they can move left and right. The actuators of the fork width change device D, which changes the gap F between the right fork 2R and the left fork 2L, are, for example, electric cylinders 27, 28 with position sensors. That is, by driving the electric cylinder 27 to advance and retract the piston 27A, the right fork 2R moves left and right, and by driving the electric cylinder 28 to advance and retract the piston 28A, the left fork 2L moves left and right. The electric cylinders 27, 28 with position sensors may also be hydraulic cylinders with position sensors, etc.

[0082] As shown in the schematic plan view of Figure 10, if the fork spacing F is larger than the width G between the insertion openings Q, Q of the target pallet P1, the fork width change device D operates to narrow the fork spacing F, as shown by the arrows at the tips of the forks. As shown in the schematic plan view of Figure 11, if the fork spacing F is smaller than the width G between the insertion openings Q, Q of the target pallet P1, the fork width change device D operates to widen the fork spacing F, as shown by the arrows at the tips of the forks.

[0083] Depending on the type of target pallet P1 obtained from the pallet type identification unit 12, the fork width change device D can change the spacing F between the left and right forks 2L, 2R to match the spacing G between the insertion openings Q, Q of that target pallet P1. Therefore, there is no need to prepare an unmanned forklift 1 for each target pallet with a different spacing G between the insertion openings Q, Q. By operating the fork width change device D to set the spacing F between the forks 2L, 2R to correspond to the spacing G between the insertion openings Q, Q, a single unmanned forklift 1 can transport multiple types of pallets.

[0084] <Work using the detection results of distance sensors and inventory sensors> The unmanned forklift 1 performs the following operations, for example, using the detection results of the distance sensor S3, which detects whether or not an object is present ahead, and the detection results of the load sensor S4, which detects whether or not the forks 2 are holding a pallet P.

[0085] That is, the unmanned forklift 1 performs a stacking operation in which a second-tier pallet P and a load W are loaded onto the load W of the first-tier target pallet P1, as shown in Fig. 2. The unmanned forklift 1 also performs an unstacking operation in which a second-tier pallet P2 and a load are picked up from the load of the first-tier pallet P1, as shown in Fig. 7, and transported to another location.

[0086] When performing the stacking operation, the distance sensor S3 can detect the height of the load W on the first pallet P1, and the load presence sensor S4 can detect the height at which the forks 2 are pulled out after the stacking operation. When performing the unstacking operation, the distance sensor S3 can detect the height of the insertion opening of the second pallet P2, and the load presence sensor S4 can detect the state in which the forks 2 are holding the pallet P2 and load to be held during the unstacking operation. Therefore, the distance sensor S3 and the load presence sensor S4 make it possible to perform reliable stacking and / or unstacking operations.

[0087] (Example of stacking operation) The unmanned forklift 1 is positioned a predetermined distance E away from the target pallet P1 shown in Fig. 2, with the photographing device S1 (Fig. 1) photographing the area ahead, and the image acquisition unit 11 acquires the photographed image from the photographing device S1. The photographed image of the target pallet P1 acquired by the image acquisition unit 11 is input to the learning model of the pallet type identification unit 12, and the pallet type identification unit 12 identifies the type of the target pallet P1.

[0088] A distance measuring device S2 (FIG. 1) measures the distance to the target pallet P1, and a pallet position and shape acquisition unit 13 acquires the position and shape data of the target pallet P1. A pallet deviation amount detection unit 14 compares the stored position and shape data corresponding to the type of target pallet P1 identified by the pallet type identification unit 12 with the position and shape data of the target pallet P1 acquired by the pallet position and shape acquisition unit 13, thereby detecting the amount of deviation in the position and posture of the target pallet P1 from its normal position.

[0089] Furthermore, the pallet stacking determination unit 17 determines whether or not the target pallet P1 is stacked by calculating the number of pallets within a predetermined specific range when the target pallet P1 is present, through image processing of the photographed image of the target pallet P1 acquired by the image acquisition unit 11. That is, since the number of pallets within the specific range is one as shown in Figure 2, the pallet stacking determination unit 17 determines that stacking is possible.

[0090] The height of the forks 2 is lowered and the unmanned forklift 1 is moved forward closer to the target pallet P1, and the unmanned forklift 1 is stopped at a position where the target pallet P1 and the pallet P are separated by, for example, a predetermined distance U, as shown in Figure 12A. At this position, the distance sensor S3 (Figure 1) that detects the presence or absence of an object detects the target pallet P1, as indicated by arrow V in Figure 12A.

[0091] Next, the forks 2 are raised as shown by arrow T1 in Fig. 12A, and the position at which the distance sensor S3 at the tip of the forks 2 no longer detects the load W on the target pallet P1 is the height H of the load W on the target pallet P1 shown in Fig. 12B. From the position at which the distance sensor S3 no longer detects the load W, the forks 2 are raised a further predetermined distance as shown in Fig. 12B.

[0092] Next, the unmanned forklift 1 is moved forward as indicated by arrow T2 in Figure 12B, and the unmanned forklift 1 is stopped at the position in Figure 12C where pallet P is overlapped directly on top of the target pallet P1, in accordance with the amount of deviation in the position and posture of the target pallet P1 from the correct position detected by the pallet deviation amount detection unit 14.

[0093] Next, the forks 2 are lowered as indicated by arrow T3 in Figure 12C, and the pallet P and the load W are stacked on top of the load W on the target pallet P1 as indicated by arrow T3 in Figure 12D. At the position in Figure 12E where the forks 2 are further lowered as indicated by arrow T4 in Figure 12D, the forks 2 are no longer holding the pallet P, i.e., the load presence sensor S4 has detected that the pallet P is not being held, and so the lowering of the forks 2 is stopped.

[0094] Next, the unmanned forklift 1 is moved backward from the state of FIG. 12E as indicated by arrow T5 in FIG. 12E, and the forks 2 are pulled out from the pallet P, thereby completing the stacking operation.

[0095] (Example of unstacking work) For example, an example of unpacking work will be described in which a second pallet P2 and its load on a first pallet P1 shown in FIG. 7 are picked up and transported to another location.

[0096] When the second pallet P2 is stacked in accordance with the first pallet P1, the second pallet P2 can be picked up based on the position and posture of the first pallet P1.

[0097] In this case, the unmanned forklift 1 is stopped in front of the stacked pallets P1, P2 and the load shown in Figure 7, and then the forks 2 are raised. The lifting of the forks 2 is stopped when the distance sensor S3 at the tip of the forks 2 changes from detecting an object in front to no longer detecting the object in front. In this state, the left and right forks 2L, 2R are located in front of the insertion openings Q, Q of the second pallet P2, so the unmanned forklift 1 is moved forward a predetermined distance, and then the forks 2 are raised a predetermined distance. With the load presence sensor S4 detecting the pallet P2, the forks 2 are holding the load, so the lifting of the forks 2 is stopped after the predetermined distance, and the unmanned forklift 1 is moved backward.

[0098] When performing the above-described unpacking operation, for example, in the example of the image in Figure 7, if the load on pallet P1 is not covered with stretch film, the left-right position of the gap between the left and right cardboard boxes that make up the load may be the same as the left-right positions of the insertion openings Q, Q of pallets P1, P2. In such a case, distance sensor S3 at the tip of fork 2 may not detect an object in front of the gap between the left and right cardboard boxes. When distance sensor S3 at the tip of fork 2 no longer detects an object in front, the fork 2 stops rising, and the left and right forks 2L, 2R may erroneously determine that they are located in front of the insertion openings Q, Q of the second pallet P2. This causes a problem in which the unpacking operation cannot be performed properly.

[0099] One possible solution to this problem is to estimate the approximate height of the underside of the second-tier pallet P2 based on image data when the pallet type identification unit 12 identifies the types of the target pallets P1 and P2 at the first stop position. For example, in an image (e.g., Figure 7) in which the pallet type identification unit 12 identifies the types of the target pallets P1 and P2, the ratio of the number of pixels in the width of the first-tier pallet P1 to the number of pixels in the vertical space between the first-tier pallet P1 and the second-tier pallet P2 is calculated. Furthermore, the actual width and height dimensions of pallet P1 are obtained based on pre-stored shape data of pallets of the same type as the target pallet P1. The approximate height of the underside of the second-tier pallet P2 can be estimated from this ratio and these actual dimensions.

[0100] Then, the unstack operation is performed after recognizing that the insertion openings Q, Q of the second pallet P2 are located above the estimated approximate height of the underside of the second pallet P2. That is, after the unmanned forklift 1 is stopped in front of the stacked pallets P1, P2 and the cargo, the forks 2 are raised, and when the distance sensor S3 no longer detects an object ahead, the position at which the forks 2 are stopped from rising is set to a position above the estimated approximate height of the underside of the second pallet P2. By operating in this manner, the above-mentioned problem can be resolved.

[0101] (Another example of unstacking work) A 2D-LiDAR, which is the distance measurement device S2, is installed on the lift bracket 3, which rises and falls together with the forks 2, and detects the position, shape data, and height of the second-stage pallet P2 while the forks 2 are rising and falling. The 2D-LiDAR is installed on the lift bracket 3 at a location midway between the bases of the forks 2L and 2R. Alternatively, as shown in Figure 1, the distance measurement device S2 installed on the main body 1A of the unmanned forklift 1 can be changed to a 3D-LiDAR, so that the position, shape data, and height of the second-stage pallet P2 can also be detected. In these cases, the distance sensor S3 at the tip of the forks 2 is used to check for interference with pallets P1 and P2.

[0102] For example, for stacked pallets P1 and P2 shown in Figure 7, the unmanned forklift 1 detects the position, shape data and height of the second pallet P2, and then, with the forks 2 at the height of insertion opening Q of pallet P2, the unmanned forklift 1 moves forward a predetermined distance so that the left and right forks 2L and 2R are aligned with the insertion openings Q and Q of the second pallet P2. Next, the forks 2 are raised, and since the forks 2 are holding a load with the load presence sensor S4 detecting the pallet P2, the forks 2 are stopped from raising after being raised a predetermined distance, and the unmanned forklift 1 moves backward.

[0103] In the above embodiment of the present invention, the case where the target pallet is on floor N has been described, but the target pallet may be on the bed of a truck. An example of an operation when the target pallet is on the bed of a truck will be described below.

[0104] (1) Devanning one pallet and cargo from the truck bed The first layer of pallets and their cargo on the truck bed are devanned in the same manner as the second layer of pallet P2 and its cargo are devanned in the "Another example of operation for devanning work" above.

[0105] (2) Devanning the upper pallet and cargo of the two-tiered pallets and cargo on the truck bed (tiered disassembly) The upper pallet and cargo of the two upper and lower tiers of pallets and cargo on the truck bed are devanned (devanned) using the same operation as the devanning of the second tier pallet P2 and its cargo in the "Another example of devanning operation" above.

[0106] If the upper pallet is stacked to fit the lower pallet on the truck bed, the upper pallet and cargo on the truck bed may be unloaded as follows:

[0107] That is, the position, shape data, and height of the lower pallet on the bed are detected using the same operation as for unloading the second tier of pallet P2 and its cargo in the "Another Operational Example of Unloading Work." Then, after stopping the unmanned forklift 1 in front of the two tiers of pallets and cargo, when raising the forks 2, the lifting of the forks 2 is stopped when the distance sensor S3 at the tip of the forks 2 changes from detecting an object in front to no longer detecting the object in front. The unmanned forklift 1 moves forward a predetermined distance, the forks 2 rise a predetermined distance, the load sensor S4 stops the lifting of the forks 2, and the unmanned forklift 1 moves backward, thereby unloading the upper pallet and cargo from the bed of the truck.

[0108] (3) Vanning (stacking) of pallets and cargo onto a single layer of pallets and cargo on the bed of a truck. A 2D-LiDAR distance measuring device S2 is installed on the lift bracket 3 that rises and falls together with the forks 2, and detects the position, shape data and height of a single layer of pallets on the truck bed while the forks 2 are rising and falling. Alternatively, by changing the distance measuring device S2 installed on the main body 1A of the unmanned forklift 1 as shown in Figure 1 to a 3D-LiDAR, it becomes possible to detect the position, shape data and height of a single layer of pallets on the truck bed.

[0109] Then, in the "Example of stacking operation," the target pallet P1 is replaced with a single layer of pallets on the truck bed, and the pallet and cargo are vanned (stacking) onto the single layer of pallets and cargo on the truck bed.

[0110] In the work examples (1) to (3) above, the learning model used to identify the type of pallet is a trained model that has undergone machine learning using images of the first pallet (lower pallet) or the second pallet (upper pallet) on the truck bed as training data.

[0111] <Operational effects of the unmanned forklift according to the embodiment of the present invention> Even if the objects to be transported by the unmanned forklift 1 are multiple types of pallets with different shapes, there is no limit to the types of pallets that can be handled. The photographed image of the target pallet acquired by the image acquisition unit 11 is input to the learning model of the pallet type identification unit 12, which has been trained by machine learning on combinations of images of each pallet and the type of pallet, and the pallet type identification unit 12 identifies the type of the target pallet. By inputting the photographed image, the pallet type identification unit 12 identifies the type of the target pallet with high accuracy. The pallet misalignment detection unit 14 can then compare and match the stored position and shape data corresponding to the type of the target pallet identified by the pallet type identification unit 12 with the position and shape data of the target pallet acquired by the pallet position and shape acquisition unit 13.

[0112] This makes it possible to detect with high accuracy the amount of deviation in the position and posture of the target pallet relative to the pallet's normal position, and also prevents accidents such as items other than pallets placed on the floor (such as cardboard boxes) being mistakenly recognized as a specific type of pallet, causing interference between the forks 2 and the pallet or item.In addition, the calculation time for comparing and matching pallet shape data is shortened, improving work speed, eliminating the need for expensive computers with high processing power.

[0113] After detecting the amount of misalignment of the target pallet with high accuracy, the unmanned forklift 1 can perform, for example, an operation of picking up a pallet and cargo on the floor and transporting them to another location, or an operation of picking up a pallet and cargo on a second level above the load on the first level pallet and transporting them to another location, which includes inserting the forks 2 into the insertion openings of the target pallet, using the amount of misalignment detected by the pallet misalignment detection unit 14. The unmanned forklift 1 can also perform a stacking operation in which the pallet and cargo held by inserting the forks 2 into the insertion openings are loaded onto the load on the target pallet in accordance with the amount of misalignment of the target pallet.

[0114] The above description of the embodiments is given by way of example only and is not intended to be limiting, and various improvements and modifications can be made without departing from the scope of the present invention. [Explanation of symbols]

[0115] 1 Unmanned forklift 1A main body 2 Fork 2L Left Fork 2R Right Fork 3 Lift Bracket 4 Inner Mast 5 Outer Mast 6 Tilt cylinder 7 Lift chain 8 Cableveyor 10 Processing device 11 Image acquisition unit 12 Pallet type identification unit 13 Pallet position / shape acquisition unit 14 Pallet deviation amount detection unit 15. Deviation amount error determination unit 16. Pallet shape determination unit 17 Pallet stacking determination unit 18 Travel control unit 19, 20 Mounting plate 21 Support plate 22 Seesaw board 22A Operating piece 22B Detector piece 23 Stop plate 24 Extension coil spring 25,26 Linear guide 27,28 Electric cylinder with position sensor 27A,28A Piston A. Load handling equipment B. Moving equipment C. Control device D. Fork width change device E Predetermined distance F Fork spacing G Width between slots H Load height I Support plate J Bracket K Frame L0 Line segment based on stored point cloud data corresponding to the type of target pallet L1 Line segment based on point cloud data of the target pallet acquired by the pallet position and shape acquisition unit M Mast N Floor P Pallet P1,P2 Target pallet Q Socket R Support shaft S1 Camera S2 Distance measuring device S3 Distance sensor S4 Location sensor S5 Self-position recognition sensor S6 Obstacle detection sensor T1~T5 Operation U Specified distance V Object detection by distance sensor W Cargo

Claims

1. An unmanned forklift equipped with a pair of forks that automatically performs loading and unloading work using pallets, an image acquisition unit that acquires a photographed image from an image acquisition device that photographs a front side of the unmanned forklift; a pallet type identification unit that has a trained learning model that has undergone machine learning on the combination of each pallet image and the type of pallet for multiple types of pallets, and identifies the type of the target pallet by inputting the photographed image of the target pallet acquired by the image acquisition unit into the learning model; a pallet position and shape acquisition unit that acquires position and shape data of the target pallet from a distance measurement device that measures the distance to the target pallet; a pallet deviation amount detection unit that stores position and shape data of each of the plurality of types of pallets in advance, and detects the amount of deviation of the position and posture of the target pallet from its normal position by comparing the stored position and shape data corresponding to the type of target pallet identified by the pallet type identification unit with the position and shape data of the target pallet acquired by the pallet position and shape acquisition unit; (A) an operation including inserting the forks into the insertion opening of the target pallet using the detected amount of deviation; (B) a stacking operation in which the pallet and cargo held by the forks inserted into the insertion openings are loaded onto the cargo of the target pallet in accordance with the amount of deviation of the target pallet; a travel control unit that controls travel to perform at least one of the above tasks; Equipped with Unmanned forklift.

2. The position and shape data is data on the height of the pallet including its insertion opening, The pallet deviation amount detection unit superimposes and compares a line segment based on the stored position and shape data corresponding to the type of the target pallet with a line segment based on the position and shape data of the target pallet acquired by the pallet position and shape acquisition unit. The unmanned forklift according to claim 1.

3. a deviation error determination unit that determines an error when the deviation detected by the pallet deviation detection unit exceeds a predetermined threshold value; 3. The unmanned forklift according to claim 1 or 2.

4. The system further includes a pallet shape determination unit that compares the shape data of the target pallet acquired by the pallet position / shape acquisition unit with pre-stored shape data of a pallet of the same type as the target pallet, and determines that there is an error in the acquired shape data of the target pallet if the difference between the shape data and the pre-stored shape data is greater than a predetermined threshold value. The unmanned forklift according to any one of claims 1 to 3.

5. The system further includes a pallet stacking determination unit that performs image processing on the photographed image of the target pallet acquired by the image acquisition unit to calculate the number of pallets within a specific range that is set in advance to include the target pallet, thereby determining whether or not the target pallet is stacked. The unmanned forklift according to any one of claims 1 to 4.

6. The pallet stacking determination unit has previously performed machine learning on pallet images of multiple tiers of pallets, using training data. The unmanned forklift according to claim 5.

7. The system further includes a fork width changer that changes the spacing between the forks to match the width between the insertion openings of the target pallet based on the type of the target pallet obtained from the pallet type identification unit. The unmanned forklift according to any one of claims 1 to 6.

8. a distance sensor disposed at the tip of the fork to detect the presence or absence of an object ahead; a presence sensor that detects whether the forks are holding the pallet; Furthermore, Using the result of detecting the presence or absence of the object by the distance sensor and the result of detecting whether the pallet is being held by the inventory sensor, Stacking operations in which a second pallet and load are loaded onto the load of the first pallet, and / or A pallet unpacking operation is performed to pick up the second pallet and cargo on the first pallet and transport it to another location. The unmanned forklift according to any one of claims 1 to 7.

9. At a first stop position, the photographing device photographs the target pallet and the pallet type identification unit identifies the type of the target pallet, and at a second stop position closer to the target pallet than the first stop position, the distance measuring device measures the distance to the target pallet and the pallet position / shape acquisition unit acquires position / shape data of the target pallet. The unmanned forklift according to any one of claims 1 to 8.

Images