Platform Detection

Abstract

A computer-implemented method is provided that includes obtaining image data representative of a platform, processing the image data with an algorithm configured to determine a nature of the platform, obtaining data indicative of the nature of the platform from the algorithm, and determining a location to which the platform is to be transported based on the data indicative of the nature of the platform.

Description

[Technical Field]

[0001] The present invention relates to the automated recognition and sorting of platforms such as pallets, and in particular to determining locations for platforms based on the classification of the platforms. [Background technology]

[0002] Platforms such as pallets and containers are well known for transporting goods from one location to another. Such transports may occur, for example, between factories, ports, warehouses, and retailers. Figures 1a and 1b show images of a typical pallet 3. The pallet 3 has an upper surface 3a and a lower surface 3b (sometimes referred to as upper and lower decks). The upper surface 3a and the lower surface 3b each include wooden planks arranged adjacent to one another. The upper surface 3a is configured to support the goods to be transported, and the lower surface 3b is configured to support the pallet 3 on a surface such as a floor or conveyor. The upper surface 3a is typically connected to the lower surface 3b by a plurality of blocks 3c, also made of wood. Connecting plates or cradles may also be present between the blocks and the upper or lower surfaces 3a, 3b. The blocks 3c are typically secured to the upper and lower surfaces 3a, 3b by fastening elements such as nails or screws. That is, nails are threaded, for example, from the upper and lower surfaces 3a, 3b into the blocks 3c to secure the upper and lower surfaces 3a, 3b to the blocks 3c. In other instances, the plastic pallet may be integrally formed as a single unit.

[0003] When items are removed from a platform, the empty platforms are retrieved so that they can be reused. However, given that platforms arrive in many different shapes, sizes, and configurations, it is often necessary to sort the empty platforms into their respective groups. Furthermore, it is often necessary to sort the platforms into groups for onward movement to deliver the groups to different destinations. This can occur, for example, when the platforms must be returned to the platform owner.

[0004] The sorting referred to above typically occurs at a platform sorting facility, e.g., a service center, or may occur at a customer's site. The sorting facility may, for example, receive a number of unsorted pallets and manually sort the pallets into desired groups, e.g., sorting half pallet sizes into different groups from full pallet sizes. Pallets may also be sorted by owner for return to the owner or for onward delivery to a location designated by the owner.

[0005] It is an aim of some embodiments of the present invention to alleviate one or more problems associated with the prior art. Summary of the Invention

[0006] According to a first aspect of the present invention, there is provided a computer-implemented method comprising obtaining image data representative of a platform, processing the image data with an algorithm configured to determine a nature of the platform based on one or more characteristics of the platform, obtaining data indicative of the nature of the platform from the algorithm, and determining a location to which the platform is to be transported based on the data indicative of the nature of the platform.

[0007] The platform may include a pallet, dolly, or container.

[0008] Processing the image data with an algorithm may include executing the algorithm on one or more processors to determine properties of the platform, wherein the one or more properties of the platform may be inherent characteristics of the platform, such as color, material, and / or shape of the platform.

[0009] The location may be a location where the platforms are to be stacked, such as a stacker location in a platform sorting facility. The location may be one of a plurality of locations, each location associated with a particular property of the platform. In this way, the platforms may be sorted into groups having the same or similar properties.

[0010] The method may further include outputting data indicative of the determined location.

[0011] Outputting data indicative of the determined position may include outputting a control signal to an actuator configured to operate to transport the platform towards the determined position, and actuating the actuator based on the control signal to transport the platform towards the determined position.

[0012] The actuator may be part of a path selection mechanism. Actuation of the actuator may change the path taken by the platform. For example, actuation of the actuator may cause the platform to veer along a particular path that may differ from the original path taken by the platform.

[0013] The method may further include selecting an actuator from a plurality of actuators based on the determined position, for example, the plurality of actuators may be configured to respectively orient the platform to a plurality of positions, such that the selected actuator is then determined to be the actuator that orients the platform to the desired position.

[0014] The method may further include advancing the platform along a first path from an initial position using the transport means, wherein actuating the actuator advances the platform along a second path toward the determined position.

[0015] The transport means may include any suitable means for transporting the platform. For example, the transport means may include a conveyor belt advanced by the operation of one or more motors. Alternatively, the transport means may include a plurality of rollers arranged perpendicular to the direction of movement of the platform. The rollers may rotate freely (or may rotate by the operation of one or more motors) so that the platform can be pushed or pulled along the rollers to advance the platform. The transport means may include one or more tracks that can guide the platform along a path. The transport means may include a combination of different types of transport means, such as rollers and tracks.

[0016] The actuator may be configured to operate at a junction of a first path and a second path, or in other examples, the actuator may be configured to operate at a junction of three or more paths.

[0017] The method may further include capturing one or more images of the platform using one or more cameras, and obtaining image data representative of the platform from the one or more images of the platform.

[0018] For example, one or more images may be taken that include at least a portion of the platform. These images may be used to obtain image data, i.e., the image data may include one or more images.

[0019] The algorithm may include a machine learning model trained to classify one or more characteristics of the platform, and the data indicative of the nature of the platform may include data indicative of the classification of the platform.

[0020] Using a machine learning model, the model can learn one or more characteristics of the platform. In this way, the machine learning model can distinguish or recognize different platforms. For example, the machine learning model may be trained to classify the manufacturer and / or model of the platform. This classification can then be used to determine the location to which the platform will be sent, e.g., so that identical platforms are grouped together at different locations. The machine learning model may provide any suitable output that can be used for classification. For example, the machine learning model may output a score (e.g., 0 to 1) for each category that represents the likelihood that the platform represented in the image data belongs to that particular category.

[0021] The algorithm may include a neural network, for example a deep neural network such as a convolutional neural network.

[0022] Image data captured from one or more cameras may be used to train a machine learning model. Furthermore, the orientation of the cameras relative to the platform when generating the training data may be approximately the same as when images are taken during platform sorting (e.g., not during training). In this way, the image data used to train the algorithm is obtained from the same source and in the same orientation as image data used during a live setting, e.g., rather than a training setting, resulting in more robust classification.

[0023] According to a second aspect of the present invention, there is provided a system comprising one or more processors and a memory storing an algorithm that, when executed by the one or more processors, is configured to determine a nature of the platform based on one or more characteristics of the platform, the one or more processors are configured to obtain image data representative of the platform, process the image data with the algorithm, obtain data indicative of the nature of the platform from the algorithm, and determine a location to which the platform is to be transported based on the data indicative of the nature of the platform.

[0024] The one or more processors may include one or more CPUs and one or more GPUs.

[0025] The one or more processors may be further configured to output data indicative of the determined location.

[0026] The system may further include an actuator, wherein the output data indicative of the determined position is a control signal, the actuator configured to receive the control signal and further configured to operate based on the control signal to transport the platform towards the determined position.

[0027] The one or more processors may be further configured to select an actuator from the plurality of actuators based on the determined position.

[0028] The system may further comprise a transport means configured to advance the platform from an initial position along a first path, wherein actuation of the actuator advances the platform along a second path toward a determined position.

[0029] The system may further include one or more cameras, wherein the one or more cameras are configured to capture one or more images of the platform, and wherein the one or more processors are further configured to obtain image data representative of the platform from the one or more images of the platform.

[0030] According to a third aspect of the present invention, there is provided a non-transitory computer-readable medium comprising computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the above-described method.

[0031] It will be appreciated that features described in the context of one embodiment may be combined with other embodiments of the invention.

[0032] Embodiments of the present invention will now be described, by way of example, with reference to the accompanying drawings, in which: [Brief explanation of the drawings]

[0033] [Figure 1a] FIG. [Figure 1b] FIG. [Figure 2] FIG. 1 shows a perspective view of a platform sorting facility. [Figure 3] A schematic diagram of a platform sorting facility is shown. [Figure 4a] FIG. 1 shows a perspective view of a platform recognition device. [Figure 4b] 1 shows an enlarged view of the platform recognition device. [Figure 4c] 1 shows a perspective view of a platform recognition device in use. [Figure 5] A schematic diagram of a machine learning model is shown. [Figure 6] 1 shows a flow chart of the method disclosed herein. DETAILED DESCRIPTION OF THE INVENTION

[0034] FIGS. 2 and 3 show diagrams of a platform sortation facility 1 that can be used to sort pallets 3 (or any platform) such as those shown in FIGS. 1a and 1b. While particular types of pallets are shown in FIGS. 1a and 1b, it should be understood that the disclosed methods may be applied to any type of pallet, such as plastic pallets, roll pallets (sometimes called dollies), and indeed any type of platform, such as containers. The platform sortation facility 1 is configured to sort pallets by detecting one or more characteristics of the pallet and determining, for each pallet, a location to which the pallet will be oriented. The platform sortation system 1 includes a transport means 2, a platform recognition device 4, and a path selection mechanism 7. While FIG. 2 shows four path selection mechanisms 7a-7d and FIG. 3 shows one path selection mechanism 7, it should be understood that any number of path selection mechanisms may be present.

[0035] The transport means 2 is configured to transport pallets 3 through the platform sortation facility 1. That is, unsorted pallets 17 may be individually placed, manually or automatically, on the transport means 2 at an initial position, such as position A, for movement through the platform sortation facility 1 to a desired position, such as position B, C, D, or E within the platform sortation facility 1. FIG. 2 shows exemplary positions B, C, D, and E, corresponding to particular stacker positions, where pallets 3 may be stacked by the stacker. FIG. 3 shows exemplary positions B and C located at the ends of paths P2 and P3. Of course, any number of positions may be used, depending on the requirements of the platform sortation facility 1. The transport means 2 may be any suitable mechanism for transporting pallets 3. For example, the transport means 2 may be one or more conveyor belts advanced by the operation of one or more motors. Alternatively, the transport means 2 may not be driven by a motor. For example, the transport means 2 may include a plurality of rollers arranged perpendicular to the direction of movement of the pallets 3. The rollers may rotate freely, allowing the pallets 3 to be pushed or pulled along the rollers to advance the pallets 3 through the platform sortation facility 1. In other examples, the rollers may be configured to rotate by the operation of a motor or other suitable torque source to advance the pallet 3. Of course, the transport means 2 may include a combination of both driven and non-driven conveyor sections. The transport means 2 may also include one or more autonomous agents, such as automated guided vehicles, configured to move the pallet from one location to another. In some cases, the transport means 2 may not be present. For example, the pallet may be moved manually between locations.

[0036] The platform recognition device 4 is configured to determine one or more characteristics of the pallet 3 (or platform). The platform recognition device 4 comprises one or more electromagnetic sensors 5a, 5b and a controller unit 6, where the controller unit 6 is coupled to the one or more electromagnetic sensors 5a, 5b (see FIG. 3). In one embodiment, the electromagnetic sensors 5a, 5b are cameras. The cameras 5a, 5b may be color area scan cameras, although other cameras may of course be used. The one or more cameras 5a, 5b are configured to image a portion of the pallet 3. In one example, the cameras 5a, 5b are configured to image a portion of the pallet 3 as it passes the platform recognition device 4 while moving along the path P1.

[0037] In the example shown in Figure 3, there are two cameras 5a, 5b. However, it will be appreciated that any suitable number of cameras may be present. The cameras 5a, 5b may be located on a structure 13 (shown in Figures 4a, 4b, and 4c) of the platform recognition device 4. For example, the structure 13 may extend above the vehicle 2 so that a pallet 3 may be seen passing through the platform recognition device 4. A diagram of the platform recognition device 4 is shown in Figure 4a without the pallet 3. Figure 4c shows the platform recognition device 4 with a pallet 3 passing through it (elements of the vehicle 2 have been omitted here for clarity).

[0038] The platform recognition device 4 may optionally include one or more lights 19 (not all lights are labeled in the figure for clarity). The lights 19 may be attached to the structure 13. The lights 19 may provide light so that clear images can be captured by the cameras 5a, 5b. The cameras 5a, 5b may be located in any suitable position on the structure 13 so that they can image the pallets. The cameras 5a, 5b may be configured to capture an image of one side of the pallet 3 as the pallet 3 passes through the platform recognition device 4, where each side of the pallet 3 corresponds to a side that is generally parallel to the direction of movement through the platform recognition device 4. However, it should be appreciated that the cameras 5a, 5b may be configured to capture any suitable angle of the pallet 3, such as the top, bottom, front, or back of the pallet. In the illustrated example, the cameras 5a, 5b are positioned at a similar height to the pallet 3 as the pallet 3 passes through the platform recognition device 4 to capture images of one or more sides of the pallet 3. The cameras 5a, 5b may be connected to the structure 13 using any suitable means. In the example shown in Figure 4a, the cameras 5a, 5b are fixed to mountings 22, which are in turn fixed to the structure 13. The example shown in Figure 4c has four cameras, two mounted on one side of the structure 13 (only three mountings 22 are visible).

[0039] Data output from cameras 5a, 5b is transmitted to controller 6 for processing. Controller 6 may include any suitable configuration for processing data output by cameras 5a, 5b. In one example, controller 6 includes an input module 8, a microprocessor such as a central processing unit (CPU) 9, a graphics processing unit (GPU) 18, non-volatile memory 10, and an output module 11, all connected via a bus. Although not shown, the GPU and CPU have access to volatile memory such as random access memory (RAM). Input module 8 is configured to receive output from sensors 5a-5l, such as data including one or more images. Memory 10 stores an algorithm that, in one embodiment, is a machine learning model M that can be executed by processor 9 or, in some embodiments, can be executed by graphics processing unit 18. Machine learning model M is configured to determine one or more characteristics of palette 3 based on image data representing the palette, where the image data may include output from one or more of cameras 5a, 5b. In one embodiment, image data representing palette 3 is processed by machine learning model M, and data indicating the classification of palette 3 is output by model M. The image data may include one or more images of the pallet 3 taken by the cameras 5a, 5b. Of course, the image data may be pre-processed before being input to the machine learning model M so that the image data is in a form suitable for input to the machine learning model M.

[0040] Based on the classification of the pallet 3 output by the machine learning model M, a location (such as locations B-E) to which the pallet will be transported may be determined. As noted above, locations B-E may correspond to the locations of particular stackers. The processor 9 or GPU 18 may perform the location determination, or may output the classification to any other suitable processor, such as a separate programmable logic controller (PLC) (not shown), that is capable of performing the location determination. Once determined, an output S may be output that includes data indicative of the location. In examples where the controller 6 performs the location determination, the data indicative of the location is output using the output module 11. Of course, if a PLC performs the location determination, the output S may be output from an output module of the PLC.

[0041] The output S may be sent to one or more routing mechanisms 7, or may be sent to a user interface 12, such as a monitor, or both. The output S may include a control signal that, for example, activates a particular routing mechanism 7 to change the path of the pallet 3 as the pallet 3 moves along the transport means 2. The controller 6 may determine a particular routing mechanism 7 to send the control signal S to, such that only that particular routing mechanism 7 changes the path of the pallet 3.

[0042] The path selection mechanisms 7 are configured to change the path taken by the pallets 3 while traveling along the transport means 2. The path selection mechanisms 7 include actuators that, when activated, cause the path of the pallets 3 to change. For example, the path selection mechanisms 7 may be located at positions along the paths where the paths branch off. In the example shown in FIG. 2, four path selection mechanisms 7a-7d are illustrated. Each of these path selection mechanisms is configured, when activated, to divert the pallets 3 from path P1 to a second path directed toward one of the stackers at positions B-E. For example, if the first path selection mechanism 7a receives the control signal S, it will operate to divert the pallets 3 from path P1 and move them along path P2 to position B. If the second path selection mechanism 7b instead receives the control signal, the first path selection mechanism 7a will not be activated (because no control signal is sent to the first path selection mechanism 7a). This allows the pallet 3 to continue on the path P1, pass the first path selection mechanism 7a, and arrive at the second path selection mechanism 7b, after which it can change course to position C.

[0043] FIG. 3 shows another arrangement of the path selection mechanism 7, in which path P1 splits into path P2 and path P3. Path P2 leads to location B, and path P3 leads to location C. Locations B and C may lead to areas where similar pallets can be grouped or stacked, or may lead to further locations where the paths split. Of course, if desired, additional platform recognition devices and / or path selection mechanisms may be located at either location B or C. Of course, in instances where there is no path selection mechanism, the user may manually place a pallet on either path P2 or path P3 based on output from user interface 12. For example, if user interface 12 outputs that a particular pallet 3 is to be sent to location B, the user may manually place the pallet on path P2 or path P3, or may use, for example, a forklift truck, to place the pallet at location B.

[0044] In some examples where output S from controller 6 is sent to path selection mechanism 7, path selection mechanism 7 may select a path using output S. For example, referring to FIG. 2, pallet 3 may have been classified as a CHEP pallet by machine learning model M. Location B may be a location where CHEP pallets are stacked. Controller 6 (or another PLC) may output a control signal S directly to path selection mechanism 7a to move pallet 3 along path P2 toward location B.

[0045] The path selection mechanism 7 may include any suitable hardware for selecting or changing a path. For example, the path selection mechanism 7 may include one or more processors, memory, I / O interfaces along with any actuator or combination of actuators that directs the platform along a particular path (such as P2 or P3 in FIG. 3 ). Such actuators are well known and examples include mechanical arms or movable surfaces or guides that move under the action of motors, hydraulic and / or pneumatic systems, etc. to apply forces to the pallets to change their orientation. The path selection mechanism 7 may be integrated into the transport means 2 in some implementations.

[0046] As described above, the decision regarding where to send the pallet 3 is based on the results of the machine learning model M. In one embodiment, the machine learning model M comprises a neural network. In one embodiment, the machine learning model M comprises a convolutional neural network. The machine learning model M is configured to determine a property of the pallet 3. In one embodiment, the machine learning model M is trained to classify the pallet 3 (e.g., the property is classification). For example, the machine learning model M may classify the pallet as being of a particular shape, type, size, or color. In other words, the machine learning model M classifies an image of the pallet as belonging to one of a plurality of trained categories. The machine learning model M receives as input image data representing the pallet 3. In one embodiment, the image data includes one or more images of the pallet 3 captured by the cameras 5a, 5b. It will of course be understood that the image data may include one or more images of the pallet captured by only one of the cameras 5a, 5b. While other machine learning models may be used, the inventors have found that a convolutional neural network is particularly well suited for this task.

[0047] An example of a machine learning model M is shown in FIG. 5. An input image 14 containing image data representing palette 3 is input into the machine learning model M. In this particular example, the input image shows block 3c of palette 3. In the illustrated example, model M includes a feature extractor portion 15 having multiple convolutional layers 23 and pooling layers 24, and a classifier portion 16 including several fully connected layers (dense layers 25) and a softmax output layer 26. In the illustrated example, each pooling layer 24 is preceded by a pair of convolutional layers 23. Of course, it will be understood that this is not the only case. For example, a single convolutional layer 23 (or three or more convolutional layers 23) may be used before each pooling layer 24. Furthermore, it will be understood that the structure of model M (e.g., the number of layers) may vary depending on the specific requirements of model M. In the illustrated example, two dense layers 25 are present, but any number of dense layers, such as one, may be used. Multiple dense layers may be used to reduce the size of the data output from the final pooling layer.

[0048] The output 21 of the model M may be a score (e.g., 0 to 1) for each category that represents the likelihood that the platform represented in the image data belongs to that particular category. In the example shown in FIG. 5, the categories for the output 21 include palette P1, palette P2, palette P3, ..., palette PN. Each category may correspond to a palette manufacturer; for example, the output may be: palette P1 may be a CHEP palette, and palettes P2, ..., palette PN may be palettes not manufactured by CHEP. In another example, palette P1 may correspond to a particular model of a CHEP palette, such as B4840A, palette P2 may correspond to a different model of a CHEP palette, such as B1210A, and palettes P3 through palette PN correspond to different models of other palettes manufactured by other manufacturers. Of course, it will be understood that other outputs can be used for different classification tasks, such as color, size, material, etc. In one embodiment, the machine learning model M is configured to run on a GPU 18. The inventors have found that running the model M on a GPU performs significantly faster than running it on a CPU.

[0049] The machine learning model M may be trained in any suitable manner. As an example, the machine learning model M is trained with training data including labeled platform images, such as labeled pallet images. The labeled pallet images include images of pallets along with labels that specify the category of the pallet image. For example, the pallet images may be images of pallets owned by a CHEP, and the labels may indicate that the images are of pallets owned by the CHEP. In this manner, the machine learning model M learns the characteristics of different pallets to identify different pallets. Of course, if the platforms to be recognized and sorted are dollies or containers, the training data may include labeled images of dollies or containers. Training the machine learning model M may include minimizing a cost function using backpropagation in conjunction with a steepest descent method. For example, training images may be input to the model M, which generates an output. A cost can be calculated based on the given output, and backpropagation can be used to adjust the network weights to minimize the cost function.

[0050] The training data may be constructed by capturing images of many different pallets (or portions of pallets) as they pass through the platform recognition device 4 (or a device having a similar arrangement of cameras 5a, 5b as the platform recognition device). In this way, the shape of the training data (e.g., camera orientation relative to the pallets, lighting, etc.) may generally match real data obtained in use (e.g., when sorting pallets in the platform sortation facility 1), resulting in more accurate classification. In a specific example, the training data is captured by one or both of the two cameras 5a, 5b of the platform recognition device 4.

[0051] Referring to FIG. 6, a method according to the present disclosure is described.

[0052] In step S1, a platform 3 (such as the pallet described above) is loaded onto the transport means 2, for example at position A, and advanced along path P1. The platform 3 may be loaded onto the transport means 2 using any known method.

[0053] In step S2, image data representing the platform 3 is obtained. For example, the platform 3 moves forward through the platform recognition device 4, and one or more cameras 5a, 5b capture images of the platform 3. The images may be color images. The cameras 5a, 5b may be triggered to begin capturing when the platform 3 reaches a particular point along the path P1. For example, the trigger may be configured to cause the cameras 5a, 5b to begin capturing images or video of the platform such that a particular portion of the platform 3 is captured. In one example, the trigger may be configured to capture a leading edge of the platform 3, where the leading edge is relative to the direction of movement along the path P1. However, it should be understood that such a trigger is not required in all embodiments, or the trigger point may be different. The cameras 5a, 5b may capture images in any suitable configuration. For example, the cameras may capture a set number of frames of video, each frame containing an image. The cameras 5a, 5b may be configured to capture only a portion of the platform, or may be configured to capture an entire side view of the platform, for example.

[0054] In step S3, the image data is processed using an algorithm, such as the machine learning model M, to determine a nature of the platform based on one or more characteristics of the platform. The nature of the platform may be a classification of the platform. The one or more characteristics of the platform may be learned characteristics of the platform, such as the color, shape, size, material, etc.

[0055] Batch processing may be used. For example, image data processed by machine learning model M may include data for a batch of images. For example, image data including multiple images may be processed by machine learning model M, and a platform classification may be determined based on the accumulated classifications. An example of this is if the image data consists of data for 25 images, and 22 of the images are classified as palette P1 and 3 of the images are classified as palette P3, the determination is that palette 3 may be classified as palette P1. In this case, the majority output was used to determine the classification as palette P1. Alternatively, a threshold may be used to determine the final classification. The threshold may be any suitable threshold, such as a certain percentage of images being classified as a particular palette type. In other examples, only one image needs to be processed. In other examples, batch processing need not be used. For example, image data including a single image may be processed to determine the nature of platform 3.

[0056] In step S4, data indicative of a nature of the platform is obtained from an algorithm, such as machine learning model M. The nature may be a classification of the platform. That is, model M may output data indicative of a classification of platform 3. For example, the classification may be that the platform is a platform manufactured by CHEP or another manufacturer, or may be a classification of a particular model of the platform.

[0057] In step S5, a determination is made of the location to which the platform 3 will be transported. For example, the controller 6 (or another controller such as a PLC) may determine the location based on the platform classification output by the model M. For example, a stacker will typically One or more types ofThe machine learning model M may be configured to stack specific platforms. When platform 3 is recognized as being of a specific type, this information is used to determine a location to send platform 3 to. This location may be the location of a stacker for the recognized platform type. This location may be determined using any suitable method. For example, the output provided by the machine learning model M may be compared to a database listing one or more classifications of platforms and their respective destinations or the specific routing mechanisms or actuators that need to be activated to guide the platform to its destination. In another example, the data output by model M may itself indicate a location.

[0058] In step S6, data indicating a position, such as position B, C, D, or E, is output. The data indicating a position may take any suitable form. For example, the data indicating a position may include the control signal described above. A control signal S may be sent to a path selection mechanism 7, where the path selection mechanism 7 is configured, upon receiving the control signal S, to automatically change the path of the platform 3 to guide the platform toward the position. A specific path selection mechanism 7 may be identified, and a control signal S is sent to the identified path selection mechanism 7. For example, the platform 3 may be classified as a CHEP platform by the machine learning model M. Position B may be a position where CHEP platforms are stacked. The controller 6 (or another PLC) may output a control signal S directly to the path selection mechanism 7a (operating at the junction of path 1 and path 2) to activate an actuator of the path selection mechanism 7a, thereby moving the platform 3 along path P2 toward position B. In this manner, the platforms 3 may be sorted into their respective groups.

[0059] Alternatively or additionally, the output S may be sent to the user interface 12, for example to be displayed on a display device. In some cases, the user may change the path of the platform 3 in the user interface 12 based on the output S. For example, the user may move the platform 3 to the position shown on the display device of the user interface, or may move the platform to the position of the vehicle 2, thereby directing the platform 3 to the position shown.

[0060] While various embodiments have been described herein, it is to be understood that this description is to be considered in all respects as illustrative and not restrictive, and that various modifications thereof which do not depart from the spirit and scope of the invention will become apparent to those skilled in the art.

[0061] The controller unit 6 may take any suitable form. For example, while only one processor, input module, output module, GPU, and memory are described, the controller unit 6 may of course have multiple such components (e.g., multiple processors) or may not have some components, such as a GPU. Furthermore, while separate inputs and outputs have been described, they may be combined together if desired. The controller unit 6 or components of the controller unit 6 may be geographically distributed away from other components of the platform sorting facility 1. That is, the control unit 6 may be located on a remote computer, such as a remote server in the cloud. Portions of the method may be performed by one or more edge devices or IoT devices. In some implementations, a smart camera may be used, in which the controller 6 is embedded within the smart camera. The smart camera may be capable of capturing images of the platform 3 and running the machine learning model M in the camera's processor. The output from the camera may then include the output from the machine learning model M or even data indicative of the location to which the platform is to be sent. In some cases, the smart camera may output control signals to a routing mechanism.

[0062] User interface 12 may include any suitable user interface, such as a PC, laptop, tablet, mobile phone, a monitor connected to speakers, etc.

[0063] Of course, the machine learning model M may be stored in any suitable location. For example, although the machine learning model M has been described as being stored in the memory 10 of the control device 6, the machine learning model M may be stored elsewhere, such as in the cloud. In such a case, the platform awareness device 4 may have an interface to obtain the machine learning model, such as a network controller.

[0064] Although the machine learning model M has been described as a convolutional neural network, which has been shown to provide robust platform recognition, other models may be used. For example, algorithms such as support vector machines, decision trees, or random forests may be used, where image features such as color information are used to classify platforms. In some cases, the algorithm may be a non-machine learning model. For example, if the platforms are sufficiently distinctive such that certain markers (such as color information) are sufficient to distinguish between platforms, different platforms may be recognized without the need to train a model using image processing algorithms.

[0065] Although the pallet 3 has been described as passing through the platform recognition device 4, it will be appreciated that the pallet need not "pass through" the platform recognition device; that is, the pallet 3 may simply pass by one or more cameras 5a, 5b, said cameras being mounted on any suitable structure such that the cameras 5a, 5b can image the image pallet 3.

[0066] Although certain aspects have been described with respect to a pallet, it will be appreciated that these aspects also apply to any platform such as a dolly or container.

Claims

1. A method of transporting a platform through a platform sorting facility, comprising: receiving a first platform on the vehicle at an initial position; a) obtaining image data representative of said platform; b) processing the image data with an algorithm configured to determine a nature of the platform based on one or more characteristics of the platform, the one or more characteristics of the platform including color, shape, size, or material of the platform; c) obtaining data indicative of properties of the platform from the algorithm; and d) determining a location to which the platform will be transported based on data indicative of the platform's characteristics; e) using said transport means to transport said platform to a location where said platform will be transported; on the first platform; receiving a second platform on the vehicle at the initial position; executing the steps a) to e) on the second platform; The method, wherein the first platform and the second platform each comprise a pallet or a dolly.

2. The method of claim 1, wherein the conveying of step e) further includes outputting data indicating the determined location.

3. outputting data indicative of the determined position outputting a control signal to the actuator; and activating the actuator based on the control signal to transport the platform toward the determined position. The method of claim 2 , comprising:

4. The method of claim 3 , further comprising selecting the actuator from a plurality of actuators based on the determined position.

5. The conveying of claim 5 comprises advancing the platform along a first path from the initial position using the conveying means, and advancing the platform along a second path toward the determined position by operating the actuator.

5. The method of claim 3 or 4, further comprising:

6. capturing one or more images of each of said platforms using one or more cameras; obtaining image data representative of each platform from one or more images of the respective platform; The method of claim 1 , further comprising:

7. the algorithm comprises a machine learning model trained to classify one or more characteristics of each of the platforms; The method of claim 1 , wherein the data indicative of the nature of each platform comprises data indicative of a classification of each platform.

8. The method of claim 7 when dependent on claim 6, wherein the machine learning model is trained using image data captured from the one or more cameras.

9. The method of claim 1 , wherein the algorithm comprises a neural network.

10. A method according to any one of claims 1 to 9, wherein the second platform is transported to a different location from the first platform.

11. A system for transporting a platform through a platform sorting facility, comprising: a vehicle; one or more processors; and a memory storing an algorithm, the algorithm, when executed by the one or more processors, configured to determine a nature of the platform based on one or more characteristics of the platform, the one or more characteristics of the platform including a color, a shape, a size, or a material of the platform; A system comprising: The system comprises: automatically receiving the first platform onto the vehicle at an initial position; a) obtaining image data representative of said platform; b) processing the image data using the algorithm; c) obtaining data from the algorithm indicative of the properties of the platform; and d) determining a location to which the platform will be transported based on data indicative of the platform's characteristics; e) using said transport means to transport said platform to the location where said platform is to be transported; With respect to the first platform, receiving a second platform on the vehicle at the initial position; Executing the steps a) to e) on the second platform. It is structured as follows: The system wherein the first platform and the second platform each comprise a pallet or a dolly.

12. The system of claim 11, configured to transport each of the platforms by outputting data indicating the determined position.

13. an actuator, wherein output data indicative of the determined position is a control signal, the actuator configured to receive the control signal and further configured to operate based on the control signal to transport the respective platform toward the determined position. The system of claim 12 further comprising:

14. The system described in claim 13, wherein the transport means is configured to advance each of the platforms from the initial position along a first path, and by activating the actuator, advances each of the platforms along a second path toward the determined position.

15. 15. The system of claim 11, further comprising one or more cameras, the one or more cameras configured to capture one or more images of the respective platform, and the one or more processors further configured to derive image data representative of the platform from the one or more images of the platform.

Images