A voice-interactive-based item sorting method, control device, and system.

Abstract

This application relates to the field of intelligent sorting robot control technology, and in particular to a voice-interactive-based item sorting method, control device, and system. The method includes: upon receiving a voice sorting command, parsing multiple target types to be sorted; synchronously updating the electronic tags of each material box accordingly; and generating or refreshing a mapping relationship between item types and material boxes that the robot can directly read, i.e., an item mapping relationship. Then, the robot is driven to place the various types of items stacked on the rack into the corresponding material boxes. This solution achieves the linkage between voice triggering, tag changes, and sorting execution, reducing manual intervention and improving the real-time performance, continuity, and flexible adaptability in dynamic classification scenarios.

Description

Technical Field

[0001] This application relates to the field of intelligent sorting robot control, and in particular to a method, control device and system for sorting items based on voice interaction. Background Technology

[0002] With the continuous iteration and maturation of industrial robots and humanoid robots, robotic equipment has been widely used in various scenarios such as warehousing and logistics sorting, offline retail product classification, exhibition product organization and setup, and small item packaging. By relying on robots to replace manual labor in completing repetitive item sorting, classification and transfer operations, the intensity of repetitive manual labor can be effectively reduced and the efficiency of small item sorting operations can be improved.

[0003] In current automated sorting systems, robots can only perform sorting operations according to preset fixed programs, and the sorting mode is fixed and monotonous. When it is necessary to adjust the sorting target, it is often necessary for professionals to manually adjust the configuration and re-enter the corresponding rules through the control backend. The adjustment process often requires machine shutdown, and the operation threshold is high and the convenience is poor.

[0004] Therefore, there is an urgent need to provide a flexible item sorting method to adapt to the operational needs of dynamically adjusting sorting categories. Summary of the Invention

[0005] This application provides a voice-interactive-based item sorting method, control device, and system to solve the aforementioned technical problems. The solution revolves around a scenario of dynamically adjusting sorting rules, constructing a linkage mechanism between voice interaction information, electronic tags on material frames, and robot sorting execution. This enables the sorting system to quickly identify sorting targets, adjust sorting relationships, and switch sorting tasks based on changes in externally inputted classification requirements. This reduces manual intervention during classification rule changes, improves the convenience of changes, and enhances the real-time response and flexible adaptability of sorting operations.

[0006] In a first aspect, embodiments of this application provide a method for sorting items based on voice interaction, including:

[0007] The robot receives voice sorting instructions and determines multiple target types based on these instructions. Based on these target types, it updates the electronic tags of multiple material boxes and updates the item mapping relationship based on the electronic tags of each material box. The item mapping relationship is robot-readable and represents the mapping relationship between each material box and the item type. Based on the updated item mapping relationship, the robot is controlled to sort the various items stacked on the rack into the corresponding material boxes.

[0008] In one possible embodiment, multiple target types are determined based on voice sorting instructions, including:

[0009] The voice sorting instructions are input into the end-to-end voice processing model. The audio features of the voice sorting instructions are extracted by the end-to-end voice processing model, and multiple target types are determined based on the audio features.

[0010] In one possible embodiment, multiple target types are determined based on voice sorting instructions, including:

[0011] Convert voice sorting instructions into text instructions; extract keywords from the text instructions; and determine multiple target types based on the extracted keywords.

[0012] In one possible embodiment, extracting keywords from text instructions includes:

[0013] Based on a natural language processing model trained using a dataset from the warehousing and sorting vertical industry, keywords are extracted from text instructions.

[0014] In one possible implementation, based on the extracted keywords, multiple target types are determined, including:

[0015] Identify ambiguous words in the extracted keywords; disambiguate the ambiguous words based on a natural language processing model to obtain disambiguated keywords; and determine multiple target types based on the disambiguated keywords.

[0016] In one possible embodiment, the electronic tags of multiple material frames are updated based on multiple target types, and based on the electronic tags of each material frame, the following are included:

[0017] Based on the item type corresponding to each material box in the current item mapping relationship and multiple target types, determine the material box to be adjusted and its corresponding target type from multiple material boxes; update the electronic tag of the type to be adjusted to the electronic tag of the corresponding target type.

[0018] In one possible embodiment, based on the item type corresponding to each frame in the current item mapping relationship and multiple target types, the frame to be adjusted and its corresponding target type are determined from multiple frames, including:

[0019] Based on multiple target types and the item types corresponding to each material box in the current item mapping relationship, at least one material box to be adjusted is determined; based on the attributes of the material box to be adjusted and historical sorting data, the target type corresponding to the material box to be adjusted is determined.

[0020] In one possible embodiment, based on the updated item mapping relationship, the robot is controlled to sort multiple types of items stacked on the rack into corresponding storage boxes, including:

[0021] Control the robot to repeatedly perform the following steps until it receives another voice sorting instruction or the item in the bin is empty:

[0022] Based on the collected images of various items stacked on the rack, the type and spatial pose of the items to be sorted are determined; based on the updated item mapping relationship and the type of the items to be sorted, the target material box is determined; based on the spatial pose of the items to be sorted and the target material box, the robot's sorting planning path is generated; based on the sorting planning path, the robot is controlled to grab the items to be sorted and place the grabbed items into the target material box.

[0023] Secondly, embodiments of this application provide a control device, including: a memory and a processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, causing the processor to execute the method provided above to control a robot to sort items.

[0024] Thirdly, embodiments of this application provide a voice-interactive item sorting system, including: a robot, a rack, a sorting table, and a control unit; the rack is configured to stack multiple types of items; the sorting table is provided with multiple material boxes, each material box being configured to store one type of item; the control unit is used to execute the method provided above to control the robot to sort the multiple types of items stacked on the rack into multiple material boxes.

[0025] In one possible embodiment, the robot is a humanoid robot with dexterous hands at the ends of its arms.

[0026] In one possible embodiment, each material frame is equipped with a display screen to show the corresponding electronic tag.

[0027] This application provides a voice-interactive item sorting method, control device, and system, realizing a voice-driven dynamic sorting control strategy. First, multiple target types of items requiring sorting are determined through voice sorting commands. The electronic tags of multiple material boxes are updated based on these target types. Then, the robot-readable item mapping relationship is updated based on the electronic tags of each material box. This allows changes in classification requirements from the voice side to be directly transmitted to the material box identifiers and the robot sorting control process. The robot then sorts the various types of items stacked on the rack into their corresponding material boxes according to the updated item mapping relationship. This reduces manual intervention and response delays when classification rules change, improving the real-time linkage, sorting continuity, and flexible adaptability of the sorting system. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0029] Figure 1 A flowchart illustrating the voice-interaction-based item sorting method provided in this application;

[0030] Figure 2A schematic diagram of a voice-interactive item sorting system provided for this application;

[0031] Figure 3 This is a schematic diagram of a circular sorting table provided in this application.

[0032] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0034] Item sorting technology is commonly used in scenarios such as warehousing and logistics, retail stores, and exhibition setup, where multiple types of items need to be sorted quickly.

[0035] Automated sorting systems typically rely on robots to sort multiple types of items that are piled up haphazardly into the corresponding item bins.

[0036] Currently, robots in automated sorting systems can only sort items of fixed types and with fixed allocation relationships according to preset programs. The sorting mapping relationship between items and material boxes is fixed and stored in advance, and cannot be adjusted autonomously according to real-time sorting needs. If it is necessary to change the sorting category, professional personnel must manually modify the program configuration in the control panel, which is a cumbersome process that often requires machine downtime and is not very convenient to use. At the same time, the electronic tags attached to the material boxes only serve as static display terminals for operators, and are only used to display classification information.

[0037] To improve the ease of operation when changing sorting requirements, some automated sorting systems allow operators to input the latest sorting requirements via voice. The system recognizes the voice input and displays the result on an electronic tag attached to the material crate. The robot then performs the sorting according to the manually modified sorting procedure. However, in this approach, voice recognition, electronic tag updates, and robot control are often separate. Voice results mostly serve only as prompts or records, failing to directly form a sorting basis that the robot can read. Especially when sorting rules change temporarily or multiple target types are involved, while the electronic tag can change its display content, it cannot synchronously drive the robot to update the correspondence between items and material crates. Manual intervention to reconfigure the process is still required, failing to fundamentally solve the shortcomings of relying on manual operation for sorting rule switching and the system's lack of flexibility.

[0038] In view of this, how to reduce human intervention after voice triggering and enable the updating of classification tags to directly link with robot sorting has become an urgent technical problem to be solved. To solve the above problems, this application provides a voice-interactive item sorting method. After receiving a voice sorting instruction, multiple target types are determined, and then the electronic tags of multiple material boxes are updated based on the multiple target types. The robot-readable item mapping relationship is updated according to the electronic tags of each material box, thereby controlling the robot to sort the multiple types of items stacked on the rack into the corresponding material boxes, thereby improving the real-time linkage and flexible adaptability when classification rules change.

[0039] Figure 1 A flowchart illustrating the voice-interaction-based item sorting method provided in this application is shown below. Figure 1 As shown, this material turnover method includes:

[0040] S101: Receives voice sorting instructions and determines multiple target types based on the voice sorting instructions.

[0041] Voice sorting commands are used to convey the sorting requirements of items, such as adjustments made by the operator to the classification rules for the current batch of items. The target type indicates the type of items to be sorted in the current batch, as parsed from the voice sorting commands. This type is subsequently used as the basis for updating the electronic tags on the material boxes and controlling the robot sorting process. The target type can be a broad category or a subcategory, such as SKU (Stock Keeping Unit).

[0042] In this application, the executing entity can be a control program deployed in a sorting workstation, or a control device, such as a host computer, that communicates with a robot controller, an electronic tag control device, and a voice acquisition terminal. Specifically, in implementation, operators input voice sorting commands at the sorting site, and the control device receives and parses these commands, determining multiple target types based on the parsing results.

[0043] The control equipment can collect voice sorting instructions issued by authorized operators on site through the sound pickup device, and after preprocessing such as noise reduction and analog-to-digital conversion of the voice signal, it can analyze and obtain multiple target types based on the preprocessed voice signal.

[0044] For example, the voice sorting instruction could be "sort the headphones and power bank into one bin, and sort the apples into another bin", and the multiple target item types could include "digital accessories" and "fruit", or "digital accessories" and "apples".

[0045] To facilitate subsequent processing, the parsing results of voice sorting instructions can be standardized to generate multiple target types with a uniform format. Once multiple target types are identified, corresponding sorting task data can be generated as input data for subsequent electronic tag updates and item mapping relationship updates.

[0046] This step transforms the raw voice content into multiple structured target types, enabling changes in classification rules to enter the sorting control link in the form of machine-processable data, thus forming the initial foundation for the linkage between voice interaction and automatic sorting.

[0047] S102: Based on multiple target types, update the electronic tags of multiple material boxes, and based on the electronic tags of each material box, update the item mapping relationship; the item mapping relationship is robot-readable and is used to characterize the mapping relationship between each material box and the item type.

[0048] The material crates are target containers for receiving sorted items. Each crate is equipped with an electronic tag, which identifies the type of item currently contained in that crate in the current batch. The item mapping relationship is task data that the robot can directly read and parse, used to represent the correspondence between each crate and the item type.

[0049] Based on the multiple target types parsed by S101, the control program or control device updates the electronic tags of multiple material boxes, achieving a one-to-one correspondence between target types and electronic tags. After the electronic tags are updated, the control program or control device regenerates or updates the item mapping relationship based on the electronic tags of each material box, obtaining the updated item mapping relationship. This updated item mapping relationship reflects the correspondence between each material box and the item type and can be directly read and used by the robot. In this way, when sorting requirements change, the user can directly inform the control device of the latest sorting system through semantic means. The control device then updates the electronic tags based on the multiple target types obtained from parsing the voice sorting instructions, and the item mapping relationship read by the robot also changes synchronously when the electronic tags change. Through the aforementioned process, the voice sorting instructions representing the latest sorting requirements are directly transformed into data that the robot can execute.

[0050] Based on the above analysis, this step, by synchronously writing the target type into the electronic tag and converting it into a robot-readable item mapping relationship, makes the electronic tag no longer just for manual viewing, but a data intermediary carrier that drives the update of sorting strategies. It opens up the transmission link from voice sorting requirements to the robot execution end, and directly transforms the sorting change requirements input by voice into standardized sorting operation strategies for robots.

[0051] S103: Based on the updated item mapping relationship, control the robot to sort the various items piled on the rack into the corresponding material boxes.

[0052] The robot is an automated device that performs item sorting operations. A single sorting operation includes grasping, transporting, and placing actions. The rack is a stacking platform for multiple types of items to be sorted. It can be located at the feeding position. The multiple types of items refer to items in the current batch that need to be reclassified according to the updated item mapping relationship. This includes items of multiple target types, and each target type can have one or more items. The corresponding material frame is the material frame that establishes a correspondence with the target type to which the item belongs in the item mapping relationship.

[0053] In some embodiments, the control device can directly send the updated item mapping relationship to the robot, which then autonomously sorts multiple types of items based on its onboard chip and image acquisition device. Alternatively, the updated item mapping relationship can be sent to the robot controller. Upon receiving the updated mapping relationship, the robot controller determines the corresponding storage bins for different item types based on the mapping relationship and controls the robot to sort the various items stacked on the shelf into the corresponding bins. In this way, the robot uses the updated item mapping relationship as the basis for target decision-making during execution, rather than relying on static bin definitions in a fixed program.

[0054] In other embodiments, the control device can control the robot to perform sorting of items of each target type based on the updated item mapping relationship and the rack images collected on site, and sort the items of each target type into the corresponding rack.

[0055] Based on the above analysis, this step directly uses the item mapping relationship generated after the electronic tag is updated to make robot decisions, so that a continuous control process is formed from voice input, electronic tag update to sorting execution. When the sorting demand changes temporarily, there is no need to manually reprogram the robot's control logic. The item mapping relationship can be updated by issuing the corresponding voice command, thereby completing the transfer of multiple types of items in the new batch to the corresponding material box.

[0056] The voice-interactive item sorting method provided in this embodiment parses voice sorting commands into multiple target types and simultaneously applies these target types to update electronic tags and item mapping relationships. The robot then performs grasping and placement actions based on the updated item mapping relationships, enabling changes in classification requirements to be directly transmitted to the sorting execution end along the same data link. Thus, the voice results no longer remain at the prompt level but form a sorting basis that the robot can read. This allows for real-time switching and automatic linkage of continuous sorting tasks in scenarios involving multiple target types, multiple material boxes, and dynamically changing classification rules.

[0057] In one possible implementation, multiple target types are determined based on voice sorting instructions, including: inputting the voice sorting instructions into an end-to-end voice processing model, extracting audio features of the voice sorting instructions through the end-to-end voice processing model, and determining multiple target types based on the audio features.

[0058] Among them, the end-to-end speech processing model is a neural network model customized and trained for sorting scenarios. It takes speech sorting instructions as input and realizes one-stop output from the original speech signal, i.e., the speech sorting instructions, to the text indicating sorting requirements, i.e., multiple target types, without the need to convert the speech signal into text before inputting it into the model.

[0059] The end-to-end speech processing model can adopt a joint structure of a speech encoder and a type output layer based on a deep neural network. The speech encoder performs time-frequency transformation on the input speech sorting instructions and extracts frame-level representations. The type output layer maps the frame-level representations to multiple type identifiers, and the output results are multiple target types.

[0060] The speech encoder can perform frame division and Fourier transform on the input speech sorting instructions, extract multi-dimensional audio signals such as Mel spectrum and time-domain energy, and the type output layer determines multiple target types based on the multi-dimensional audio signals.

[0061] When training an end-to-end speech processing model, supervised learning can be performed based on historical voice commands and their corresponding target type labels, so that the model can learn the speech patterns of different item type names, quantifiers, and combined commands.

[0062] When a voice sorting command contains multiple classification objects, the end-to-end speech model performs joint recognition on each target type, and outputs a type set (including a plurality of target types) consistent with the update requirements of multiple material frames, and the type set can be directly used for subsequent electronic label update and item mapping relationship generation.

[0063] By directly inputting the voice sorting command into the end-to-end speech processing model and having the model output multiple target types, the intermediate step of first transcribing text and then performing rule matching is omitted, which saves computational overhead, eliminates the recognition deviation caused by the intermediate step, and improves the accuracy and efficiency of target type recognition.

[0064] On the basis of the foregoing embodiment, further, determining a plurality of target types based on a voice sorting command includes: converting the voice sorting command into a text command; extracting keywords in the text command; and determining the plurality of target types based on the extracted keywords.

[0065] The voice sorting command can be converted into a text command through ASR (Automatic Speech Recognition) technology.

[0066] When extracting keywords from the text command, stop word filtering is first performed on the text command to remove connecting function words, action auxiliary words and the like that have no actual category meaning, such as "ba", "fang dao", "dui ying", "de", "ge lei", "yi ge", etc.; then, word segmentation is performed on the remaining text, and entity words representing item categories or SKU specifications, such as earphones, power banks, apples, data cables, etc., can be matched and extracted from the remaining text by relying on a special keyword lexicon for warehouse sorting, so as to obtain a plurality of keywords.

[0067] Each extracted keyword can be directly mapped to one type to obtain a plurality of target types.

[0068] When extracting keywords, the extracted keywords may be grouped based on the semantic grouping relationship of the keywords in the text command to obtain a plurality of keyword groups, so that one target type is mapped for each keyword group based on the keywords contained in each keyword group.

[0069] The semantic grouping relationship is used to represent the sorting and storage associations of different item keywords in the text instruction. Based on the sorting and allocation limitation descriptions such as "same material box" and "another material box" in the representation text, keywords that need to be placed in the same material box and keywords that need to be placed in different material boxes can be distinguished, thereby dividing the keyword groups and ensuring that each group of keywords corresponds to only a single target type and a single storage material box.

[0070] By using speech-to-text conversion and keyword extraction, target type recognition is achieved, and intermediate data such as readable text and extracted keywords are fully retained, making the entire parsing process traceable. When new item types or SKUs are added, only the keyword mapping library needs to be updated, without retraining the model. Iterative maintenance is simple, and hardware computing power requirements are lower.

[0071] In one possible implementation, keywords are extracted from text instructions, including: extracting keywords from text instructions based on a natural language processing model trained using a dataset from the warehouse sorting vertical domain.

[0072] The warehousing and sorting vertical domain dataset can include warehousing operation instructions, material box identification information, item type names, synonym expression samples, and common colloquial instruction expressions in sorting scenarios. This dataset is used to enable natural language processing (NLP) models to learn specific semantic associations within the warehousing and sorting context. The NLP model can employ a deep neural network-based text semantic analysis model. Its input layer receives the segmented text instructions, the encoding layer represents the contextual relationships between terms, and the output layer generates keywords and their weights.

[0073] After obtaining voice sorting instructions and converting them into text instructions, the text instructions are input into the natural language processing model. Based on the domain semantic features formed during the training process, the model performs relevance judgment and screening on the terms in the text, and extracts information that can characterize the sorting object, classification intention, target material box or quantity limit as keywords.

[0074] During operation, the model trained on a vertical domain dataset for warehouse sorting can perform semantic recognition on text content such as "red stationery in the left-side material box," identifying terms directly related to the sorting task, such as "red stationery" and "left-side material box," while suppressing generalized terms irrelevant to the current scenario, making the output keywords more consistent with the business context of warehouse sorting. Since the model has learned the correspondence between category names, material box descriptions, and operation instructions in the vertical domain samples, its output can be directly used as the input basis for determining multiple target types.

[0075] By adopting the above method, keywords in text instructions can be extracted more accurately, and the correspondence between keywords and the semantics of warehouse sorting operations is more stable. This allows subsequent target type determination, electronic tag updates, and robot sorting control to be processed in a coordinated manner based on a unified semantic input. Because the keyword extraction process fully utilizes the semantic features of the vertical domain dataset, information related to the sorting task in the text instructions can be effectively preserved and output, thereby improving the consistency of recognition and execution matching of text instructions in the warehouse scenario throughout the entire sorting chain.

[0076] In some embodiments, a multimodal model can be used to determine multiple target types by taking voice sorting instructions and text instructions as inputs and combining the features of both voice and text modalities, thereby further improving the accuracy of target type recognition.

[0077] Based on the aforementioned embodiments, further, multiple target types are determined based on the extracted keywords, including: identifying ambiguous words in the extracted keywords; disambiguating the ambiguous words based on a natural language processing model to obtain disambiguated keywords; and determining multiple target types based on the disambiguated keywords.

[0078] Based on a polysemous vocabulary dictionary for sorting scenarios, ambiguous words in the extracted keywords, such as "notebook," can be identified. This step can be performed by a natural language processing model.

[0079] For ambiguous words in the extracted keywords, the natural language processing model can make a judgment based on the context of the ambiguous word in the text instruction, the scene features of the current classification scene, and the multiple definitions of the ambiguous word in the polysemous word dictionary, and complete the disambiguation.

[0080] For example, the ambiguous word "notebook" has two common interpretations: a paper notebook and a portable electronic notebook device. If the context includes electronic accessories such as data cables and cooling stands, the natural language processing model tends to classify it as an electronic notebook device; if the context includes stationery such as folders and sticky notes, it tends to classify it as a paper notebook. Simultaneously, scene tags are introduced for secondary constraints: in a stationery sorting scenario, paper notebooks are prioritized; in a digital item sorting scenario, electronic notebooks are prioritized. By combining the contextual entity association features and scene constraints, a unique interpretation is output, completing word meaning disambiguation and producing unambiguous keywords.

[0081] After disambiguating each ambiguity, multiple target types are determined based on each keyword. The extracted keywords can be grouped according to their semantic grouping relationships within the text instructions, resulting in multiple keyword groups. Then, based on the keywords contained within each keyword group, a target type is mapped to each keyword group.

[0082] By first identifying ambiguous words in the extracted keywords, and then using a natural language processing model combined with the context to complete semantic judgment and achieve disambiguation, multiple target types are determined based on the disambiguated keywords. This effectively reduces the classification bias caused by polysemous words, improves the accuracy of item classification, reduces the probability of misplacement during sorting, and enhances the automation and recognition stability of the entire voice sorting process.

[0083] Based on the aforementioned embodiments, further, based on multiple target types, the electronic tags of multiple material frames are updated, and based on the electronic tags of each material frame, the following steps are taken: based on the item type corresponding to each material frame in the current item mapping relationship and multiple target types, the material frame to be adjusted and its corresponding target type are determined from the multiple material frames; the electronic tag of the type to be adjusted is updated to the electronic tag of the corresponding target type.

[0084] The "Items to be Adjusted" category represents the items that require type or electronic tag changes in the current mapping relationship. The current item mapping relationship is the one before the update, used to characterize the mapping relationship between each item and its type.

[0085] After identifying multiple target types, the control equipment matches and compares the current item type of each material box with the target types to obtain a set of mismatched target types and a set of material boxes to be adjusted. When the item type corresponding to a certain material box is inconsistent with all target types, the material box is identified as a material box to be adjusted; when the target type is inconsistent with all the current item types of the material boxes, the target type is identified as a mismatched target type.

[0086] After determining each material frame to be adjusted and each mismatched target type, assign a target type to each material frame to be adjusted from among the multiple mismatched target types. This assignment can be random or according to certain rules.

[0087] For example, a target type can be assigned to each frame to be adjusted based on its position.

[0088] In some embodiments, there are multiple racks that can be deployed on both sides of the robot. When assigning target types to the boxes to be adjusted, the assignment can be made based on the location of the racks where the items of each target type are stacked and the location of each box to be adjusted, thereby shortening the robot's sorting path.

[0089] The electronic tag can be updated by rewriting the display content, replacing the pre-stored tag template, or adding target type codes. The updated electronic tag content is synchronously written into the item mapping relationship, so that the internal records of the system are consistent with the external display of the material box.

[0090] By updating the item mapping relationship and identifying multiple target types, the system accurately selects the material frames that need to be adjusted and updates the electronic tags of the corresponding material frames accordingly, without having to refresh all tags, thus reducing communication and computing overhead.

[0091] Based on the aforementioned embodiments, further, based on the item type corresponding to each material box in the current item mapping relationship and multiple target types, determine the material box to be adjusted and its corresponding target type from multiple material boxes, including: determining multiple material boxes to be adjusted based on multiple target types and the item type corresponding to each material box in the current item mapping relationship; and determining the target type corresponding to each material box to be adjusted based on the attributes and historical sorting data of each material box to be adjusted.

[0092] The attributes of the adjustable crate can include its size, maximum load capacity, location, opening specifications, disabled status, and range of compatible types. Historical sorting data includes the type of the crate during each historical batch sorting.

[0093] When there are multiple boxes to be adjusted and multiple matching target types, in order to achieve accurate matching between the target type and multiple boxes to be adjusted, in this embodiment, it is necessary to combine the attributes of the boxes to be adjusted and historical sorting data to assign target types to the boxes to be adjusted.

[0094] There is no need to assign a target type to the adjustment box that is in a disabled state. You can directly set the electronic tag of the adjustment box to be empty, so that the items will be prohibited from being sorted into the box during the sorting of this batch of items.

[0095] When assigning target types to the boxes to be adjusted, multiple target types can be filtered based on the attributes of the boxes. For example, filtering can be done based on the range of compatible types for the boxes, eliminating target types outside that range, or based on the size and opening specifications of the boxes, eliminating target types with mismatched sizes. Subsequently, based on the historical sorting data of the boxes to be adjusted, the appropriate target type for each box can be determined from the remaining target types.

[0096] For example, if historical sorting data shows that a container has been used for the same target type in multiple consecutive sorting cycles, then that target type is assigned to that container.

[0097] Based on historical sorting data, the score of various combinations of boxes to be adjusted and target types can be determined, and allocation can be carried out based on the combination with the highest score.

[0098] Based on historical sorting data, the historical frequency of the target type to be assigned to the material box to be adjusted can be counted, so that the target type with higher historical frequency can be assigned to the material box to be adjusted first.

[0099] By combining attributes and historical sorting data to allocate material boxes to target types, it is possible to avoid items exceeding the carrying capacity of the material boxes in terms of size and weight, reduce the frequency of changing sorting correspondences, lower the frequency of electronic tag updates, optimize the sorting layout in the warehouse, and improve sorting stability and operational efficiency.

[0100] Based on the aforementioned embodiments, further, based on the updated item mapping relationship, the robot is controlled to sort multiple types of items stacked on the rack into corresponding material boxes, including: controlling the robot to repeatedly execute the following steps until a voice sorting command is received again or the items in the material box are empty: determining the type and spatial pose of the items to be sorted based on the collected images of the multiple types of items stacked on the rack; determining the target material box based on the updated item mapping relationship and the type of the items to be sorted; generating a sorting planning path for the robot based on the spatial pose of the items to be sorted and the target material box; controlling the robot to grab the items to be sorted based on the sorting planning path and placing the grabbed items into the target material box.

[0101] Images of the various items stacked on the shelf can be captured by a camera mounted on the robot, or by a camera fixed above the shelf.

[0102] After determining the items to be sorted and their types, the control equipment retrieves the target material box that matches the type based on the updated item mapping relationship, and uses the target material box's position parameters, opening orientation, and graspable space range as path generation inputs. Combined with the spatial pose of the items to be sorted, a sorting planning path is generated.

[0103] The sorting path can be solved by combining the robot's kinematic model with obstacle avoidance constraints. The path includes continuous motion segments approaching the gripping position, holding position, lifting position, and delivery position. The robot controls the robotic arm and end effector to complete gripping and releasing according to the path. The gripper can be a two-finger parallel gripper, an adsorption actuator, or a composite gripper.

[0104] After completing one grabbing and delivery, the robot continues to collect and identify images of the remaining items on the shelf. When it determines that there are still items to be sorted and no new voice sorting command is received, it continues to execute the aforementioned control process in a loop until the items on the shelf are empty or a new voice sorting command is received.

[0105] By linking identification, mapping query, path planning, and grasping and placing, the robot can directly complete continuous sorting based on the updated item mapping relationship. This method synchronizes the positioning of the target material box with the robot's movements, reduces human intervention, and makes the transfer of various types of items between racks and boxes continuous and adaptable.

[0106] This application also provides a control device, including: a memory and a processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, causing the processor to perform the method provided above to control a robot to sort items.

[0107] By pre-storing computer execution instructions in memory and having them executed by the processor, the control device can directly implement the aforementioned sorting control method. This integrates the determination of target type after voice triggering, electronic tag updates, and robot-readable mapping updates into a single control chain, enabling the robot to continuously sort multiple types of items according to the latest classification rules throughout multiple work cycles. This reduces the system disconnect between voice recognition, tag changes, and robot execution, thereby reducing the frequency of manual reconfiguration. It also allows for rapid task switching even when classification rules change temporarily, thus improving response speed, operational stability, and sorting accuracy in continuous sorting processes.

[0108] In the specific implementation process, the processor executes computer execution instructions stored in the memory, causing the processor to perform the method provided in the above embodiments.

[0109] The specific implementation process of the processor can be found in the above method embodiments, and its implementation principle and technical effect are similar, so it will not be repeated here.

[0110] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0111] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0112] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0113] Figure 2 A schematic diagram of a voice-interactive item sorting system provided in this application is shown below. Figure 2 As shown, the item sorting system includes a robot, a rack, a sorting table, and a control unit.

[0114] Figure 2 Taking a material rack as an example, in some embodiments there can be multiple material racks, which can be distributed circumferentially around the robot, such as on the left and right sides of the robot.

[0115] The rack is configured to hold multiple types of items, and the sorting table is equipped with multiple material boxes, each of which is configured to hold one type of item. The control unit is used to execute the voice-triggered sorting control method to control the robot to sort the multiple types of items stacked on the rack into multiple material boxes.

[0116] A robot is an automated work subject that performs grasping, handling and placing actions under the scheduling of a control unit. It is used to pick up and place items on the rack and transfer them to the corresponding material boxes on the sorting table.

[0117] Robots are typically positioned between the racks and the sorting table or on one side of the racks to form a work path suitable for back-and-forth transport. They are connected to the control unit via wired communication interfaces, wireless communication interfaces, or fieldbuses to receive sorting tasks, motion commands, and status feedback.

[0118] In one possible embodiment, the robot may take the form of a six-axis robotic arm, a collaborative robot body, or a gantry-type handling mechanism. Its end effector may be equipped with an electromagnetic suction mechanism, a vacuum suction cup gripper, or a parallel gripper to adapt to items of different shapes and materials. The robot body may be constructed of a metal frame, a cast aluminum shell, or a high-strength composite material. The length of the working arm, the turning radius, and the opening and closing stroke of the end effector may be matched according to the width of the rack, the distribution range of the rack, and the size of the items. The exemplary gripping space should cover the accessible range between the rack stacking area and the sorting table delivery area.

[0119] A rack is a temporary storage and supply carrier for stacking various types of items to be sorted. Its function is to provide material support for robots to pick up materials by category or batch, and to make different types of items form identifiable gripping positions in space.

[0120] The material rack can be set on one side of the robot and correspond to the robot's working space, enabling the robot to complete the grasping action according to the preset trajectory, while the control unit is set on the other side of the robot.

[0121] In one possible embodiment, the rack can take the form of multi-layer shelves, side-by-side shelves, or modular pallet units. It can also be a vertical frame, stacked, or drawer-type structure. The material can be metal profiles, engineering plastics, or aluminum alloy sheets to balance structural strength, lightweight, and corrosion resistance. Different layers or areas can correspond to the stacking positions of different types of items. Its overall height, layer spacing, and single-layer bearing area can be adapted to the robot arm span, gripper opening size, and maximum external dimensions of the items. The exemplary interlayer clearance can be set according to the height of the items and combined with a safety margin to ensure that the robot end effector has sufficient space to enter and exit when gripping.

[0122] A sorting station is a sorting platform equipped with multiple material frames for receiving items dropped by robots. Its function is to provide classified storage locations for sorted items, allowing different categories of items to fall into their corresponding material frames, thus forming a sorting result by category. The sorting station is usually set within the robot's reach and is located on the other side or in front of the material rack so that the robot can directly turn and drop the items after completing the grabbing.

[0123] In one possible embodiment, the sorting table can be configured as a linear, ring, or matrix arrangement. Multiple material frames can be arranged along the length of the sorting table, circumferentially, or in a grid pattern. Each material frame can be a turnover box, storage box, or pallet container. The material can be wear-resistant plastic, metal sheet, or composite board. Status identification marks, sensor triggering structures, or category identification structures can be set on the side walls or bottom of the material frames to enable the control unit to identify the current storage category, full load status, or target type. The overall size of the sorting table is usually matched with the robot's maximum working radius and the layout of the material racks. The spacing between each material frame can meet the avoidance requirements of the robot's end effector.

[0124] The control unit is a processing device used to uniformly schedule and control the voice interaction results, sorting task logic and robot action execution. Its function is to parse the target type after receiving the voice sorting instruction, generate the mapping relationship between the item and the material box, and issue gripping, handling and placement control instructions to the robot to drive the robot to complete the sorting according to the target classification.

[0125] The control unit can integrate a processor, memory, communication module and motion control module, and be deployed in a robot, stand-alone control cabinet, edge computing device or industrial computer, and connected to the robot and status components on the sorting table via cable, Ethernet, wireless LAN or fieldbus.

[0126] See also Figure 2 The control unit includes a main unit and a display, which can be movably mounted near the main unit via a display bracket.

[0127] The voice-interactive item sorting system provided in this application, when the control unit receives a voice sorting command, parses it to obtain multiple target types of items to be sorted in the current batch, and then establishes a correspondence between the target types and the category identifiers, i.e., electronic tag information, of each material box on the sorting table, so that each target type forms an item mapping relationship that can be read and executed by the robot. Subsequently, the control unit generates a task queue based on the mapping relationship and schedules the robot to pick up the corresponding type of items from the material rack located on one side in sequence. After receiving the control command, the robot moves to the target picking position according to the planned trajectory, uses the end effector to complete the gripping, sucking or lifting action, and then transports the item to the top of the corresponding material box and completes the placement. After the placement is completed, the control unit continuously updates the subsequent tasks based on the robot feedback and the material box status, so that the voice recognition results, material box status updates and robot action execution form a continuous linkage.

[0128] Because the category name is directly converted into a material box mapping relationship after input and participates in robot control, there is no need for repeated manual reconfiguration when the item category changes. The sorting path and placement target can be adjusted synchronously with the task changes, thus enabling continuous operation of multiple types of items in dynamic sorting scenarios. Furthermore, the storage category of each material box is consistent with the robot's execution target, thereby suppressing sorting interruptions, response delays, and missorting / omissions. It should be understood that the above example is for demonstration purposes only and is not limiting. The specific structure, quantity, arrangement, and communication method of the material rack, sorting table, robot, and control unit can all be replaced and adjusted according to the actual application scenario.

[0129] Based on the aforementioned embodiments, the robot is further described as a humanoid robot with dexterous hands at the ends of its arms.

[0130] In this application, the two arms can move independently or in coordination, and their joints can perform rotation, swinging, lifting, extension, and contraction movements to bring the dexterous hand to a predetermined pick-up or put-down position. The dexterous hand can be installed on the end flange or end connector of the robot forearm and form a stable transmission relationship with the end of the robotic arm through bolt connection, quick-release snap-fit, or integrated flange fixation. Its internal drive can be a motor screw drive, tendon cable drive, hydraulic drive, or pneumatic drive to drive each finger joint to open, close, pinch, or cover according to a predetermined trajectory.

[0131] For example, a dexterous hand can be a five-finger bionic structure, a three-finger adaptive structure, or a multi-finger flexible structure. The five-finger bionic structure facilitates precise finger-separation operations similar to those of a human hand, the three-finger adaptive structure facilitates stable envelope grasping while simplifying the structure, and the multi-finger flexible structure is suitable for conformal grasping of irregular or fragile objects. Its finger skeleton can be made of metal skeleton, carbon fiber reinforced material, or high-strength engineering plastic, and covered with silicone, rubber, or elastomer coatings to balance support strength, flexibility, and friction performance. Some fingertips can also be equipped with replaceable wear-resistant pads or anti-slip textured layers to adapt to different objects such as cardboard boxes, plastic containers, metal parts, or glass products. Furthermore, the dimensions of the dexterous hand are typically matched to the maximum dimensions of the items and the height of the shelf. For example, the palm width, maximum opening span, and finger length can be set according to the maximum grasping size of the items to be sorted, and the finger length can be a predetermined proportion of the palm length to meet the requirements of enveloping grasping while maintaining hand dexterity. The robot's height, arm span, and working radius should cover the farthest retrieval point on the shelf and the farthest placement point on the sorting table to avoid situations where it cannot reach the item or where its posture interferes with the task. Based on the above structure, the robot can identify, approach, grasp, transport, and place target items according to the item mapping relationship issued by the control unit. When faced with items of different shapes, sizes, and placement postures, the arms and dexterous hand can achieve high fault tolerance through posture adjustment and adaptive envelopment of the knuckles, thereby reducing the reliance on manual reconfiguration of the end effector and improving the sorting system's adaptability to dynamic classification tasks and continuous operation capability. It should be understood that the above examples are for demonstration purposes only and are not limiting. The proportions, number of degrees of freedom, dexterous finger configuration, and material selection of humanoid robots can be adapted to different application scenarios.

[0132] The combination of humanoid robots and dual-arm dexterous hands can form a more flexible grasping and delivery channel between the rack and the sorting table. This allows the classification instructions updated by the control unit after voice triggering to be more directly translated into actual sorting actions, reducing intermediate human intervention. Furthermore, the combination of dual-arm collaboration and precise end-effector grasping capabilities improves the stability and efficiency of sorting various types of items, as well as the adaptability to complex item shapes.

[0133] In one possible implementation, each material frame is equipped with a display screen to show the corresponding electronic tag.

[0134] In this application, the display screen is a visual information display component set on the surface of the material frame or in the vicinity of the material frame. It is used to display the item category, sorting status or electronic tag content corresponding to the material frame to the operator or control system in real time. Its function is to provide a clear and identifiable dynamic identification for each material frame so that the robot, control unit or human reviewer can quickly identify the target item category of each material frame, thereby improving sorting accuracy and supporting dynamic classification updates after voice interaction.

[0135] The display screen is usually installed on the front end, top edge, outer surface of the side wall, or on an independent bracket of the material frame, and is connected to the control unit via data cable or wireless communication to receive tag refresh instructions and status display data; the electronic tag, as an updatable display content bound to the material frame, can display information such as text, graphics, color codes or QR codes, thereby expressing the category name or auxiliary identification information corresponding to the current material frame.

[0136] In one possible embodiment, the display screen can be a liquid crystal display (LCD), an electronic paper display, or an LED dot matrix display. The LCD can employ a glass substrate and backlight module structure, suitable for displaying high-resolution text and icons. The electronic paper display can use a flexible substrate to achieve low power consumption and long-term display. The LED dot matrix display can use a modular LED array to enhance long-distance visibility. The display screen housing material can be ABS plastic, aluminum alloy, or tempered glass, and the surface can be coated with dustproof, waterproof, or impact-resistant coatings to adapt to warehousing or industrial environments. The size of the display screen is typically matched to the opening size of the material frame, its installation position, and viewing distance. For example, its display area can be controlled within a local area of ​​the front view of the material frame to avoid obstructing the loading port while ensuring that the classification labels are clearly readable at normal operating distances. The aspect ratio of the exemplary display screen can be set according to the label text layout, the text height can meet the reading requirements within a range of several meters, and its thickness should be as thin as possible to reduce the space occupied by the material frame.

[0137] When in operation, the display screen can perform display actions such as switching between on and off, refreshing content, scrolling icons, and changing colors, and synchronously update the content of the electronic tags under the command of the control unit. Alternatively, the display screen can be replaced with an e-ink screen, a digital tube display module, a projection-type tag display device, or a pluggable card tag, and its installation method can be screw fixing, clip fixing, magnetic fixing, or guide rail sliding installation. The corresponding electronic tag information can also be updated by local storage, wireless transmission, or near-field communication.

[0138] When the system starts, the control unit receives the voice sorting command, parses it to obtain multiple target types, and assigns the multiple target types to the electronic tags bound to each material box. Then, it sends a refresh command to the display screen on each material box, so that each display screen synchronously displays the text, graphics, color codes or QR code information corresponding to the target type.

[0139] Because the display screen is directly installed on or near the material rack, the operator can quickly confirm the current electronic tag content of each material rack from the rack side, the sorting table side, and within the robot's visual coverage area. The robot, based on the item mapping relationship established by the control unit, identifies the various items to be sorted on the rack, grabs them, and places them into the corresponding material rack indicated by the display screen. As the voice command changes, the control unit can update the electronic tag again and synchronously drive the display screen to change its display content, thus ensuring that the material rack identification is consistent with the robot's execution path, avoiding manual resetting and information lag caused by temporary adjustments to classification rules. Based on the above process, it can be seen that the synchronous display of the display screen and electronic tag not only enhances the visualization of the target type in the material rack but also allows the electronic tag update results to more directly serve the robot's sorting decisions and manual verification, thereby improving the recognition efficiency, operational continuity, and classification accuracy during continuous sorting of multiple types of items.

[0140] For example, the sorting table can be a circular sorting table. Figure 3 A schematic diagram of a circular sorting table provided in this application is shown below. Figure 3 As shown, during sorting, the robot is located at the center of the circular sorting table, which is uniformly arranged with several material frames (or boxes) along the circumference. Each material frame is equipped with an electronic tag with a screen.

[0141] A heat dissipation and dustproof grid is installed on the outer bottom of the circular sorting table (the side away from the robot). This grid has a perforated mesh structure, which serves two purposes: firstly, it provides a convection cooling channel for the internal control, drive, and power components, promptly dissipating the heat generated during equipment operation and preventing overheating shutdown; secondly, it prevents dust and small workpiece debris from entering the equipment cavity, while also isolating internal electrically charged components for safety protection. The mesh structure also allows for direct observation of the internal cooling fan's operating status, facilitating maintenance and inspection.

[0142] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the methods provided above.

[0143] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method provided above.

[0144] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0145] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0146] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0147] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0148] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0149] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0150] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0151] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for sorting items based on voice interaction, characterized in that, include: Receive voice sorting instructions and determine multiple target types based on the voice sorting instructions; Based on the multiple target types, the electronic tags of multiple material boxes are updated, and based on the electronic tags of each material box, the item mapping relationship is updated; the item mapping relationship is robot-readable and is used to characterize the mapping relationship between each material box and the item type. Based on the updated item mapping relationship, the robot is controlled to sort the various items piled on the rack into the corresponding material boxes.

2. The method according to claim 1, characterized in that, The process of determining multiple target types based on the voice sorting instructions includes: The voice sorting instruction is input into an end-to-end voice processing model. The audio features of the voice sorting instruction are extracted by the end-to-end voice processing model, and the multiple target types are determined based on the audio features.

3. The method according to claim 1, characterized in that, The process of determining multiple target types based on the voice sorting instructions includes: Convert the voice sorting instructions into text instructions; Extract keywords from the text instructions; Based on the extracted keywords, the multiple target types are determined.

4. The method according to claim 3, characterized in that, The extraction of keywords from the text instructions includes: Based on a natural language processing model trained using a dataset from the warehousing and sorting vertical industry, keywords are extracted from the text instructions.

5. The method according to claim 4, characterized in that, The determination of the multiple target types based on the extracted keywords includes: Identify ambiguous words in the extracted keywords; Based on the natural language processing model, the ambiguous words are disambiguated to obtain disambiguated keywords, and based on the disambiguated keywords, the multiple target types are determined.

6. The method according to any one of claims 1-5, characterized in that, The process of updating the electronic tags of multiple material frames based on the multiple target types, and based on the electronic tags of each material frame, includes: Based on the item type corresponding to each material box in the current item mapping relationship and the multiple target types, determine the material box to be adjusted and its corresponding target type from the multiple material boxes; Update the electronic tag of the type to be adjusted to the corresponding target type electronic tag.

7. The method according to claim 6, characterized in that, The step of determining the frame to be adjusted and its corresponding target type from the multiple frame boxes based on the item type corresponding to each frame in the current item mapping relationship and the multiple target types includes: Based on the multiple target types and the item types corresponding to each material box in the current item mapping relationship, at least one material box to be adjusted is determined. Based on the attributes of the material frame to be adjusted and historical sorting data, the target type corresponding to the material frame to be adjusted is determined.

8. The method according to any one of claims 1-5, characterized in that, Based on the updated item mapping relationship, controlling the robot to sort the various types of items piled on the rack into the corresponding material boxes includes: The robot is controlled to repeatedly perform the following steps until a voice sorting command is received again or the items in the material basket are empty: Based on the collected images of various items stacked on the rack, the type and spatial pose of the items to be sorted are determined. Based on the updated item mapping relationship and the type of the items to be sorted, the target material box is determined; Based on the spatial pose of the items to be sorted and the target material box, the sorting planning path of the robot is generated; Based on the sorting planning path, the robot is controlled to grab the items to be sorted and place the grabbed items into the target material box.

9. A control device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-8 to control the robot to sort items.

10. A voice-interactive item sorting system, characterized in that, include: Robots, racks, sorting tables, and control units; The rack is configured to hold multiple types of items; The sorting table is equipped with multiple material frames, each of which is configured to store one type of item. The control unit is used to execute the method of any one of claims 1-8 to control the robot to sort multiple types of items stacked on the rack into multiple material boxes.

11. The system according to claim 10, characterized in that, The robot is a humanoid robot with dexterous hands at the ends of its arms.

12. The system according to claim 10, characterized in that, Each of the aforementioned material frames is equipped with a display screen to show the corresponding electronic tag.

Images