A linkage control method, device and equipment of a space retail warehouse and a storage medium

Abstract

The application discloses a linkage control method, device and equipment of a space retail warehouse and a storage medium. The method comprises the following steps: in response to a dialogue request initiated by a user, performing a scene routing task and a dialogue reasoning task in parallel; wherein the scene routing task is used to determine a target scene corresponding to the dialogue request; the dialogue reasoning task is used to output a reply text and structured decision information; when it is determined that the target scene meets a preset output condition, a scene switching instruction carrying a global session identifier is sent to a space presentation end; the global session identifier is generated based on a first dialogue request initiated by the user; the intention parameters of the current user are obtained by analyzing the structured decision information, and when the intention parameters meet a preset trigger condition, a preparation instruction carrying the global session identifier is sent to a cabinet machine fulfillment device, so that the cabinet machine fulfillment device starts a fulfillment preparation process; and the immersive experience and transaction efficiency of the user can be improved.

Description

Technical Field

[0001] This application relates to the field of intelligent control, and more specifically, to a linkage control method, device, equipment, and storage medium for a spatial retail warehouse. Background Technology

[0002] In recent years, with the deep integration of spatial display technology and unmanned retail technology, spatial retail pods have rapidly developed as a new type of micro-smart retail terminal. These pods integrate multimodal human-computer interaction and immersive content display capabilities, providing users with an interactive experience that surpasses traditional vending machines. Currently, typical spatial retail pods are usually equipped with intelligent dialogue guide systems, large-screen or 3D spatial display systems, and automatic cabinet fulfillment systems, which can guide users to browse, inquire about, and purchase products through natural language interaction. However, in existing spatial retail pod systems, the dialogue, scene display, and cabinet fulfillment are independent of each other and lack an effective linkage mechanism. This results in the dialogue and spatial scene being out of sync, creating a fragmented experience. It also leads to longer time for users to retrieve their goods, reduced transaction efficiency, and ultimately, a significant impact on the user experience. Summary of the Invention

[0003] The main objective of this application is to provide a linkage control method, device, equipment, and storage medium for a spatial retail warehouse, in order to solve the technical problems in the prior art where dialogue, scene display, and cabinet fulfillment are independent and lack an effective linkage mechanism, resulting in a disconnect between dialogue and spatial scene, causing a fragmented experience, and a disconnect between dialogue intent and cabinet fulfillment, causing long waiting times for picking up goods and low transaction efficiency. This application achieves simultaneous, unblocked, and conflict-free linkage between the dialogue agent, spatial display scene, and cabinet fulfillment device, enabling the spatial scene to switch synchronously in real time according to the dialogue intent, and the cabinet to prepare for delivery in advance when the user's purchase intent is clear, thereby eliminating the fragmentation of experience and improving the immersive experience and transaction efficiency.

[0004] To achieve the above objectives, the first aspect of this application proposes a linkage control method for a spatial retail warehouse, applied to a server, comprising: In response to a user-initiated dialogue request, a scene routing task and a dialogue reasoning task are executed in parallel; wherein, the scene routing task is used to determine the target scene corresponding to the dialogue request; and the dialogue reasoning task is used to output the response text and structured decision information. When the target scene meets the preset output conditions, a scene switching instruction carrying the global session identifier is sent to the spatial presentation terminal; the global session identifier is generated based on the user's first dialogue request. The structured decision information is parsed to obtain the current user's intent parameters. When the intent parameters meet the preset triggering conditions, a preparatory instruction carrying the global session identifier is sent to the cabinet fulfillment device to enable the cabinet fulfillment device to start the fulfillment preparation process.

[0005] In one or more possible embodiments, when it is determined that the target scene meets the preset output conditions, sending a scene switching instruction carrying the global session identifier to the spatial rendering end includes: When the target scene is generated before the first reply text and the confidence level of the target scene is greater than the first confidence threshold, it is determined that the target scene meets the preset output conditions, and the scene switching instruction carrying the global session identifier is asynchronously sent to the spatial presentation terminal. When it is determined that the target scenario does not meet the preset output conditions, the target scenario, the confidence level corresponding to the target scenario, and the global session identifier are stored in the preset session state storage area.

[0006] In one or more possible embodiments, the method further includes: Semantic features are extracted from the dialogue request; When the extracted semantic features meet the preset first type of pickup semantic rules, it is determined that the current situation is an external order scenario, and a pop-up command carrying the global session identifier is sent to the associated display terminal; When the extracted semantic features meet the preset second type of pickup semantic rules, and there is pending transaction data in the current session context, it is determined that the current situation is an internal purchase scenario, and a preset delivery guidance reply is output.

[0007] In one or more possible embodiments, parsing the structured decision information to obtain the current user's intent parameters, and when the intent parameters meet preset triggering conditions, sending a preparatory instruction carrying the global session identifier to the counter fulfillment device, includes: Extract the intent category and the confidence level corresponding to the intent category from the structured decision information; The preparatory instruction is generated when the intent category belongs to a preset intent set and the confidence level corresponding to the intent category is greater than a second confidence level threshold, or when the number of consecutive occurrences of the intent category is greater than a preset number of occurrences threshold. The preparatory instruction is sent to the cabinet fulfillment device through a preset callback interface.

[0008] In one or more possible embodiments, the method further includes: The generated control commands carrying the global session identifier are scheduled and distributed according to a preset priority; wherein the priority levels are configured in descending order from high to low as follows: user interrupt command, scene switching command, pop-up display command, and voice broadcast command.

[0009] In one or more possible embodiments, the method further includes: Upon receiving a new round of dialogue request containing interruption semantic features, the user interruption instruction is generated and triggered. In response to the user interruption command dispatched by the scheduler, the pending sentence queue in the voice broadcast module is cleared, and the currently executing voice broadcast processing is terminated.

[0010] In one or more possible embodiments, the method further includes: After the spatial presentation terminal executes the scene switching command and switches to the target scene, monitor the duration of continuous no user interaction. When the duration reaches the preset silent time window, a rollback command is sent to the spatial presentation terminal so that the spatial presentation terminal executes the rollback command and switches to the standby scene; Once the scene switching operation is completed, the current scene state information will be synchronized and updated to the global session state storage area.

[0011] Secondly, this application provides a linkage control device for a spatial retail warehouse, comprising: The task execution module is used to respond to a user-initiated dialogue request and execute a scene routing task and a dialogue reasoning task in parallel; wherein, the scene routing task is used to determine the target scene corresponding to the dialogue request; the dialogue reasoning task is used to output the response text and structured decision information. The scene switching instruction sending module is used to send a scene switching instruction carrying the global session identifier to the spatial presentation end when the target scene meets the preset output conditions; the global session identifier is generated based on the first dialogue request initiated by the user. The preparatory instruction sending module is used to parse the structured decision information to obtain the intent parameters of the current user. When the intent parameters meet the preset triggering conditions, it sends a preparatory instruction carrying the global session identifier to the cabinet fulfillment device so that the cabinet fulfillment device can start the fulfillment preparation process.

[0012] Thirdly, this application provides a linkage control device for a spatial retail warehouse, the device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method as described in any of the first aspects.

[0013] Fourthly, this application provides a storage medium storing a computer program for causing a computer to perform the method described in any of the first aspects.

[0014] The technical solutions provided by the embodiments of this application may include the following beneficial effects: In this application, by responding to user dialogue requests and executing scene routing and dialogue inference tasks in parallel, and issuing scene switching instructions to the spatial presentation terminal based on a global session identifier, while parsing the structured decision information output by dialogue inference, the fulfillment preparation process of the vending machine is triggered in advance when the intent parameters meet the conditions. This achieves simultaneous, low-latency linkage between the dialogue agent, the spatial display scene, and the machine's fulfillment actions within the spatial retail cabin. On the one hand, the parallel execution mechanism solves the problem of asynchronous dialogue and scene and fragmented experience caused by traditional serial processing. On the other hand, the pre-fulfillment mechanism driven by dialogue intent breaks down the traditional process barrier of "preparing for delivery only after payment," significantly shortening the user's actual waiting time for picking up goods and significantly improving transaction efficiency and the user's immersive interactive experience in unmanned retail scenarios. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 A flowchart illustrating a linkage control method for a spatial retail warehouse provided in this application; Figure 2 A schematic diagram of a linkage control device for a spatial retail warehouse provided in this application; Figure 3 This is a block diagram of a linkage control device for a spatial retail warehouse according to an embodiment; Figure 4 This is a block diagram of a storage medium provided according to an embodiment. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] In existing technologies, the generation of sales scripts and the switching of spatial display scenes (such as large-screen content rendered by Unreal Engine or naked-eye 3D images) operate independently. When a user asks "how to prevent counterfeiting" or "brand story" via voice, the dialogue system can provide a corresponding response, but the spatial display scene often remains on the default product list page or advertisement page. Scene switching usually relies on manual triggering or simple preset logic in the backend, and cannot automatically and in real-time match according to the user's current dialogue intent. This asynchronous phenomenon of "hearing and seeing" directly causes a fragmented interactive experience, weakens the sense of immersion, and also affects the effective delivery of key information. At the same time, in existing unmanned retail systems... In this system, the activation logic of the POS terminal's fulfillment devices (such as delivery channels and conveyor belts) is "payment followed by delivery." This means that the delivery process for the corresponding goods (such as unlocking the delivery channel and preheating the conveyor belt) is only initiated after the user completes the payment and receives a payment success instruction. This implies that from the time the user clearly expresses their purchase intention, such as "I want to buy this bottle of water," until the final payment is completed, the POS terminal remains idle and waiting. After the user completes the payment, they still need to wait 3 to 5 seconds or even longer for the mechanical delivery time to retrieve the goods. This disconnect between the user's intention and the physical fulfillment significantly prolongs the entire process from decision-making to retrieval, reduces transaction efficiency, and particularly impacts the user experience in high-frequency consumption scenarios.

[0019] Based on the aforementioned problems in the prior art, this application provides a linkage control method for a spatial retail warehouse, such as... Figure 1 As shown, it is applied to the server and includes: Step 101: In response to a user-initiated dialogue request, a scene routing task and a dialogue reasoning task are executed in parallel; wherein, the scene routing task is used to determine the target scene corresponding to the dialogue request; and the dialogue reasoning task is used to output the response text and structured decision information. In one or more possible embodiments, when the server receives a dialogue request initiated by a user, it obtains the corresponding voice stream or text information. For example, if it is a voice signal, it converts it into text-format dialogue request data through an internal automatic speech recognition module. At the moment the dialogue request is received, the server checks the current user's interaction state. When it determines that the user is initiating a dialogue request for the first time, it generates a globally unique global session ID. When the user interacts with the dialogue request again, the global session ID is used as the index key to quickly extract the current user's historical session context from the server's high-speed memory database (the session state storage area in this application). This includes, for example, the dialogue records of the previous rounds, the status of the currently viewed products, and the transaction data to be processed within the session.

[0020] To overcome the problems of increased latency in the first character (first token) and severe disconnect between terminal screen and voice response caused by the serial processing architecture of traditional unmanned retail terminals, which adopts either "routing the scene first, then performing dialogue inference" or "switching the scene after dialogue inference is completed," this application introduces an asynchronous, non-blocking, parallel dual-task architecture. After obtaining the dialogue text and the current session context, the server immediately starts the scene routing task and the dialogue inference task simultaneously through a multi-threaded or asynchronous coroutine mechanism, allowing the two to run independently in terms of computing resources without blocking each other. On one hand, the server inputs the aforementioned dialogue request data and session context into a pre-deployed lightweight scene routing model. Considering the extremely high real-time requirements of the scene routing task, the model typically uses an intent classification algorithm with a small number of parameters or an efficient rule matching engine. After rapid inference, it outputs the target scene corresponding to the current dialogue request and the confidence level corresponding to the target scene. The target scenarios encompass various 3D spatial displays or large-screen interface states that can be presented within the retail cabin, such as standby scenarios (scene_0), product 3D details scenarios, brand story immersive scenarios, and product anti-counterfeiting and traceability scenarios, etc. These will not be listed individually. The goal of the scenario routing task is to ensure that the visual end displays the appropriate image before the main agent speaks. On the other hand, the server will simultaneously call a larger-scale leading sales dialogue agent (such as a sales agent built based on a large language model) to perform dialogue reasoning tasks. This leading sales dialogue agent performs deep semantic understanding, multimodal logic analysis, and business rule fusion based on the same input data. It is worth noting that the dialogue reasoning task in this application not only generates human-like interactive dialogue but also provides control basis for the underlying hardware. Therefore, it is configured to output two different formats simultaneously: one is the reply text, which mainly carries the sales agent's service scripts, product introductions, or business guidance, and is pushed to the downstream text-to-speech (TTS) module in a streaming manner for direct voice interaction with the user. Secondly, there is structured decision information, which encapsulates precise business instructions extracted from the user's fuzzy dialogue, such as intent parameters (e.g., intent category is "Add to Cart ADD_TO_CART" or "Product Inquiry PRODUCT_INQUIRY"), the confidence score of the intent, and the specific product entity ID involved.

[0021] Step 102: When the target scene meets the preset output conditions, send a scene switching instruction carrying the global session identifier to the spatial presentation terminal; the global session identifier is generated based on the user's first dialogue request. In one or more possible embodiments, when the target scene is generated before the first reply text and the confidence level of the target scene is greater than a first confidence threshold, it is determined that the target scene meets the preset output conditions, and the scene switching instruction carrying the global session identifier is asynchronously sent to the spatial presentation end; when it is determined that the target scene does not meet the preset output conditions, the target scene, the confidence level of the target scene, and the global session identifier are stored in a preset session state storage area; specifically, in actual operation, since the dialogue reasoning task (leading purchase agent) is usually generated word by word by text units (Tokens), generating it before the first reply text means that the scene routing model must complete the reasoning before the leading purchase agent outputs the first syllable (first Token); at the same time, in order to avoid the user's ambiguous expression leading to the wrong scene switching, the confidence level corresponding to the target scene must be given. When the confidence level is greater than the first confidence threshold (for example, set to 0.85), it can be determined that the target scene meets the preset output conditions. At this time, in order not to block the execution of any main thread, a scene switching instruction carrying the global session identifier is asynchronously sent to the spatial rendering end through a low-latency network communication protocol (such as UDP protocol). This ensures that when the voice broadcast module just starts playing the first reply (or when the first text content is replied), the 3D spatial display device (spatial rendering end) has completed the smooth transition from the previous scene to the target scene, eliminating the sense of audiovisual disconnect in traditional systems.

[0022] Secondly, in response to abnormal situations caused by complex network environments or fluctuations in model computing power, when the target scenario fails to meet the preset output conditions (e.g., the scenario routing task takes too long, resulting in a result only after the first reply text is output; or although the calculation speed is fast, the calculated confidence of the target scenario is lower than the first confidence threshold), the server will not forcibly block or delay the already started voice reply process, nor will it blindly send uncertain scenario switching instructions to the spatial presentation end. Instead, it will securely store the target scenario obtained from this routing, the specific confidence value, and the current global session identifier as an implicit context feature in the preset session state storage area. On the one hand, this ensures the consistency and traceability of states between multiple terminals and multiple services. On the other hand, when the user adds relevant information in subsequent conversations, such as when the user casually mentioned "Is this thing real?" in the previous round, the scenario was not switched because the confidence was only 0.6, but the state has been stored; in the next round, the user says "Let me see the certificate," and the main agent and the scenario routing model can combine this stored record to instantly increase the confidence, thereby accurately triggering scenario switching in the next round.

[0023] Furthermore, in a more refined implementation scenario, if the target scene output by the scene routing task takes too long, but the confidence level of the target scene is still greater than the first confidence threshold, the server can also trigger a parallel remedial strategy. For example, it can obtain the progress of the audio streaming output queue in real time and determine the processing method of the scene switching command by evaluating the remaining percentage of the current dialogue. Specifically, if the server detects that the current audio playback progress has not exceeded the preset progress threshold (for example, the entire reply text has just been played for less than 10%), it can send the scene switching command to the spatial presentation end. In order to avoid the screen from abruptly switching during the audio playback, the server does not immediately send the scene switching command, but performs symbol matching on the text units of the streaming output to find the next "semantic breakpoint" (such as a period, question mark, or long comma in the text). When the audio playback just reaches the audio pause and breathing gap corresponding to the punctuation mark, the scene switching command is sent to smoothly switch to the new scene. This ensures a balance between smooth dialogue and rich spatial display, maintaining high system stability and a good user experience within the retail cabin under various extreme interactive conditions.

[0024] Step 103: Parse the structured decision information to obtain the current user's intent parameters. When the intent parameters meet the preset triggering conditions, send a preparatory instruction carrying the global session identifier to the cabinet fulfillment device so that the cabinet fulfillment device can start the fulfillment preparation process.

[0025] In one or more possible embodiments, the intent category and the confidence level corresponding to the intent category are extracted from the structured decision information; when the intent category belongs to a preset intent set and the confidence level corresponding to the intent category is greater than a second confidence level threshold, or when the number of consecutive occurrences of the intent category is greater than a preset number of occurrences threshold, the preparatory instruction is generated; and the preparatory instruction is sent to the cabinet fulfillment device through a preset callback interface. Specifically, traditional unmanned retail cabinets only issue addressing and delivery instructions after confirming receipt of a "payment success" signal from a third-party payment platform. This application explicitly states that the dialogue reasoning task outputs structured decision information, and extracts the intent category representing the user's true business needs and the confidence level corresponding to the intent category by parsing the structured decision information. First, it determines whether the intent category is included in a preset intent set. If it is included, it further confirms whether the confidence level corresponding to the intent category is greater than a second confidence threshold (e.g., set to 0.9 or 0.95). If so, it indicates that the user's purchase intent is highly likely. In real natural language interaction, users often show hesitation or unclear expression, causing the single confidence level given by the large model to hover around 0.80, failing to directly trigger the high threshold. Therefore, when the user interacts with the same product multiple times, and the number of consecutive occurrences of the intent category exceeds the preset number threshold (e.g., mentioning or asking twice or more consecutively), the triggering condition can also be determined to be met.

[0026] Once the triggering conditions are met, the server generates a preparatory instruction. This instruction not only carries a global session identifier to ensure traceability but also encapsulates the SKU code of the target product and the cargo channel coordinate mapping information. It is then sent to the corresponding cabinet fulfillment device via a pre-defined high-performance callback interface (such as WebSocket communication based on long connections, the MQTT IoT lightweight protocol, or direct RPC remote procedure calls). Upon receiving the preparatory instruction, the cabinet fulfillment device immediately initiates the fulfillment preparation process. At this point, the user may still be listening to the sales guide's voice or have just opened their phone to scan the code for payment. While the machine is in operation, the internal motors are already preparing. For example, a multi-axis robotic arm or XY coordinate lifting platform will move in advance and precisely stop at the specific cargo channel where the target product is located. At the same time, the electromagnetic lock at the front of the cargo channel may also enter a semi-unlocked state. Through this series of advance displacement and electromechanical preparation actions in physical space, the originally lagging mechanical addressing time is moved forward to the interval between user interaction and payment. When the user finally completes the scan payment and the server sends the final delivery confirmation signal, the mechanical device that is already in place only needs to perform the final step of pushing the product out, and the product can instantly fall into the retrieval compartment.

[0027] According to the linkage control method of the spatial retail warehouse provided in this application, the core technical pain point of the separation between the dialogue system, spatial display scene and underlying electromechanical performance in the existing unmanned retail terminal can be completely solved, thereby realizing seamless and low-latency linkage between multimodal interactive terminal and physical hardware.

[0028] In the practical application of unmanned retail, the act of users "picking up goods" includes two different scenarios: one is that users have placed an order in advance through an external mobile device and then enter the offline retail pod to pick up the goods with a verification code; the other is that users have just completed their on-site selection through multiple rounds of interaction with the dialogue agent inside the retail pod and casually ask how to get the goods. In traditional space retail pods, when encountering these two situations, a window will pop up on the screen saying "Please enter the pickup code". This will not only cause serious confusion for users who make purchases on-site (because they did not place an order online and therefore do not have a pickup code), but also raise the risk of complaints about violations.

[0029] In one or more possible embodiments, the method further includes: extracting semantic features from the dialogue request; when the extracted semantic features satisfy a preset first type of pickup semantic rule, determining that the current situation is an external order scenario, and sending a pop-up instruction carrying the global session identifier to the associated display terminal; when the extracted semantic features satisfy a preset second type of pickup semantic rule, and there is pending transaction data in the current session context, determining that the current situation is an internal purchase scenario, and outputting a preset delivery guidance reply. The first type of pickup semantic rule is typically configured to include a set of semantic features with strong external order orientation, such as recognizing that the user's speech contains clear and specific verification terms such as "pickup code", "scan to pick up", "online order", and "mini-program pickup"; based on the determination result of the first type of pickup semantic rule, determining that the current situation is an external order scenario, the server will directly send a pop-up instruction carrying the global session identifier to the display terminal (such as a touch screen or interactive panel), and the display terminal will actively invoke a verification interaction interface for inputting a pickup numeric code or displaying a QR code scanning box on the user interface, thereby efficiently guiding externally attracted users to complete the fulfillment loop. The second type of pickup semantic rules are typically configured to include only generalized, fuzzy, or weakly targeted pickup question features. When the extracted semantic features only satisfy the preset second type of pickup semantic rules, such as when the user simply asks "How do I pick it up?", "Where do I get it?", or "Where is the pickup point?", this is insufficient to determine the actual business attribution. The server will trigger a context-deep judgment mechanism: using the global session identifier generated in the aforementioned steps as a unique index key, a query request is initiated to the session state storage area to retrieve the user's operation trajectory within this round of session. If pending transaction data is found in the session context, such as when the user clearly expressed their intention to purchase a certain item in the previous two rounds of dialogue, a virtual purchase has been dynamically generated in the context. By analyzing vehicle data or a recently created / completed local machine order record, the server can accurately determine whether the user is currently in a continuous internal purchase scenario. Upon confirming this, the server triggers strict instruction isolation and state protection measures: First, it implements an interception strategy in the underlying instruction distribution center, forcibly blocking and intercepting any pop-up instructions intended to activate the pickup code input box, resolutely preventing misleading interactive interfaces from appearing on the screen at the wrong time. Second, the server informs the main dialogue agent of this status, prompting it to output a pre-set delivery guidance response. This response will be presented directly to the user through the display interface and / or voice system, such as: "Your goods are ready; please pick them up directly from the pickup slot below."

[0030] In multimodal retail cabin scenarios, the system needs to simultaneously drive a mediumless aerial imaging device (spatial presentation end), a touch interactive screen (display terminal), an audio player (voice broadcast module), and underlying physical hardware (cabinet fulfillment device). This can easily lead to competitive conflicts and resource competition among multiple terminal commands during frequent, multi-round human-computer interactions. For example, if there is a lack of overall coordination when a user suddenly changes their mind or interrupts, the system may experience screen tearing where the screen has switched to the new product, but the voice is still broadcasting the old product, or the old voice broadcast has not stopped while the new voice response has started playing.

[0031] In one or more possible embodiments, the method further includes: scheduling and distributing various control commands carrying the global session identifier according to a preset priority; wherein the priority levels are configured in descending order from high to low as: user interruption command, scene switching command, pop-up display command, and voice broadcast command; when a new round of dialogue request containing interruption semantic features is received, the user interruption command is generated and triggered; in response to the scheduled and distributed user interruption command, the pending sentence queue in the voice broadcast module is cleared, and the currently executing voice broadcast processing is terminated. Specifically, control commands carrying a global session identifier need to be distributed according to a priority queue. The priority of the queue is configured from high to low as follows: user interrupt command, scene switching command, pop-up display command, and voice broadcast command. When a new round of dialogue audio is received, and features containing clear interruption or negation semantics such as "wait a minute," "no," "change," or "I don't want it anymore" are extracted, a user interrupt command with the highest execution authority is generated. After this command is prioritized and distributed, the system performs an interruption operation: First, the queue of unprocessed sentences in the voice broadcast module is forcibly cleared, truncating unprocessed text data; second, the underlying audio driver is controlled to immediately stop the currently executing voice broadcast action. Through the above command scheduling and interruption handling mechanism, the system can quickly stop unfinished voice output, effectively avoiding the problem of new and old voice overlapping in multi-round dialogues, and timely releasing system resources so that the server can quickly receive and process new user requests, thereby ensuring smooth interaction in multi-terminal collaboration.

[0032] In open, offline, unmanned public retail spaces, user interactions are often highly random and fragmented. For example, if a user searches for a specific expensive product (such as entering a highly targeted 3D anti-counterfeiting display or a detailed brand story scene) and then leaves without exiting, traditional retail pod systems often wait indefinitely or remain on that screen. This not only leaks the previous user's browsing privacy but also leaves the next new user facing a confusing, in-depth details page, greatly undermining the initial immersive "ready to use" experience.

[0033] In one or more possible embodiments, the method further includes: after determining that the spatial presentation terminal executes the scene switching instruction and switches to the target scene, monitoring the duration of continuous no user interaction; when the duration reaches a preset silent time window, sending a rollback instruction to the spatial presentation terminal to cause the spatial presentation terminal to execute the rollback instruction and switch to the standby scene; and after determining that the scene switching operation is completed, synchronously updating the current scene state information to the global session state storage area. Specifically, when it is determined that the spatial presentation terminal has successfully executed the scene switching instruction and switched from the default standby state (or the previous scene) to a specific business target scene (such as a non-standby interactive scene), the server's internal timer engine will immediately start a delayed scheduling task. This system monitors multimodal sensor data (including but not limited to microphone voice activity, touchscreen clicks, and human movement captured by visual sensors) in real time within the retail compartment to calculate the duration of continuous user interaction. If a user makes any valid interaction during the monitoring process, the timer will be instantly reset and restarted. However, when the duration accumulates and eventually reaches a preset silent time window (e.g., set to 30 or 45 seconds of no interaction), the server will send a rollback command to the spatial presentation end. Upon receiving the rollback command, the spatial presentation end will smoothly roll back the current in-depth product details or anti-counterfeiting scene and switch to the default standby scene (such as the system's initial scene_0 welcoming standby loop animation), ensuring that every new user sees the most standard and user-friendly initial state of the retail compartment.

[0034] In one or more possible embodiments, to ensure that "automatic rollback" triggered by timeout or "active switching" triggered by user intent does not disrupt the underlying logic consistency of the system, this application constructs a rigorous state feedback closed loop. Regardless of the cause of the scene switch, as long as the server determines that the scene switch operation has been completed, the server will forcibly update and overwrite the latest currently displayed scene state information to the global session state storage area. This ensures that the global session state storage area always serves as the sole real data source for the entire system. In this way, when the user speaks again, the purchasing agent can accurately read the current screen state from the context when performing the next round of dialogue reasoning, and output the most appropriate voice response based on the visual content in front of the user. At the same time, when the system encounters a sudden network interruption or process restart, it can also instantly achieve breakpoint resumption and screen restoration for multiple terminals based on the state information in the storage area.

[0035] This application also provides a linkage control device for a spatial retail warehouse, such as... Figure 2 As shown, it includes: The task execution module 201 is used to respond to a user-initiated dialogue request and execute a scene routing task and a dialogue reasoning task in parallel; wherein, the scene routing task is used to determine the target scene corresponding to the dialogue request; and the dialogue reasoning task is used to output the response text and structured decision information. The scene switching instruction sending module 202 is used to send a scene switching instruction carrying the global session identifier to the spatial presentation terminal when the target scene meets the preset output conditions; the global session identifier is generated based on the first dialogue request initiated by the user. The preparatory instruction sending module 203 is used to parse the structured decision information to obtain the intent parameters of the current user. When the intent parameters meet the preset triggering conditions, it sends a preparatory instruction carrying the global session identifier to the cabinet fulfillment device so that the cabinet fulfillment device can start the fulfillment preparation process.

[0036] Based on the same inventive concept, this application also provides a linkage control device for a spatial retail warehouse, including at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-described linkage control method for the spatial retail warehouse.

[0037] like Figure 3 As shown, the device includes a processor 301, a memory 302, a communication interface 303, and a bus 304. The processor 301, memory 302, and communication interface 303 are interconnected via the bus 304.

[0038] Processor 301 is used to read instructions from memory 302 and execute them, so that at least one processor can execute the linkage control method for the spatial retail warehouse provided in the above embodiments.

[0039] The memory 302 is used to store various instructions and programs for the bandwidth allocation method provided in the above embodiments.

[0040] Bus 304 can be 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 representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0041] Processor 301 can be a central processing unit (CPU), a network processor (NP), a graphics processing unit (GPU), or any combination of CPU, NP, and GPU. It can also be a hardware chip. The aforementioned hardware chip can be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The aforementioned PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0042] In addition, this application also provides a storage medium, such as Figure 4 As shown, the storage medium stores a computer program that enables the computer to perform any of the methods described in the above embodiments.

[0043] The memory may include readable media in the form of volatile memory, such as random access memory (RAM) 401 and / or cache memory 402, and may further include read-only memory (ROM) 403.

[0044] The memory may also include a program / utility 405 having a set (at least one) of program modules 404, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0045] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0046] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0047] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0048] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0049] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for linkage control of a space retail warehouse, characterized by, Applied to servers, including: In response to a user-initiated dialogue request, a scene routing task and a dialogue reasoning task are executed in parallel; wherein, the scene routing task is used to determine the target scene corresponding to the dialogue request; and the dialogue reasoning task is used to output the response text and structured decision information. When the target scene meets the preset output conditions, a scene switching instruction carrying the global session identifier is sent to the spatial presentation terminal; the global session identifier is generated based on the user's first dialogue request. The structured decision information is parsed to obtain the current user's intent parameters. When the intent parameters meet the preset triggering conditions, a preparatory instruction carrying the global session identifier is sent to the cabinet fulfillment device to enable the cabinet fulfillment device to start the fulfillment preparation process.

2. The method according to claim 1, characterized in that, When the target scene is determined to meet the preset output conditions, a scene switching instruction carrying the global session identifier is sent to the spatial presentation end, including: When the target scene is generated before the first reply text and the confidence level of the target scene is greater than the first confidence threshold, it is determined that the target scene meets the preset output conditions, and the scene switching instruction carrying the global session identifier is asynchronously sent to the spatial presentation terminal. When it is determined that the target scenario does not meet the preset output conditions, the target scenario, the confidence level corresponding to the target scenario, and the global session identifier are stored in the preset session state storage area.

3. The method according to claim 1, characterized in that, The method further includes: Semantic features are extracted from the dialogue request; When the extracted semantic features meet the preset first type of pickup semantic rules, it is determined that the current situation is an external order scenario, and a pop-up command carrying the global session identifier is sent to the associated display terminal; When the extracted semantic features meet the preset second type of pickup semantic rules, and there is pending transaction data in the current session context, it is determined that the current situation is an internal purchase scenario, and a preset delivery guidance reply is output.

4. The method according to claim 1, characterized in that, The process involves parsing the structured decision information to obtain the current user's intent parameters. When the intent parameters meet preset triggering conditions, a preparatory instruction carrying the global session identifier is sent to the cabinet fulfillment device, including: Extract the intent category and the confidence level corresponding to the intent category from the structured decision information; The preparatory instruction is generated when the intent category belongs to a preset intent set and the confidence level corresponding to the intent category is greater than a second confidence level threshold, or when the number of consecutive occurrences of the intent category is greater than a preset number of occurrences threshold. The preparatory instruction is sent to the cabinet fulfillment device through a preset callback interface.

5. The method according to claim 1, characterized in that, The method further includes: The generated control commands carrying the global session identifier are scheduled and distributed according to a preset priority; wherein the priority levels are configured in descending order from high to low as follows: user interrupt command, scene switching command, pop-up display command, and voice broadcast command.

6. The method as described in claim 5, characterized in that, The method further includes: Upon receiving a new round of dialogue request containing interruption semantic features, the user interruption instruction is generated and triggered. In response to the user interruption command dispatched by the scheduler, the pending sentence queue in the voice broadcast module is cleared, and the currently executing voice broadcast processing is terminated.

7. The method according to claim 1, characterized in that, The method further includes: After the spatial presentation terminal executes the scene switching command and switches to the target scene, monitor the duration of continuous no user interaction. When the duration reaches the preset silent time window, a rollback command is sent to the spatial presentation terminal so that the spatial presentation terminal executes the rollback command and switches to the standby scene; Once the scene switching operation is completed, the current scene state information will be synchronized and updated to the global session state storage area.

8. A linkage control device for a spatial retail warehouse, characterized in that, include: The task execution module is used to respond to a user-initiated dialogue request and execute a scene routing task and a dialogue reasoning task in parallel; wherein, the scene routing task is used to determine the target scene corresponding to the dialogue request; the dialogue reasoning task is used to output the response text and structured decision information. The scene switching instruction sending module is used to send a scene switching instruction carrying the global session identifier to the spatial presentation end when the target scene meets the preset output conditions; the global session identifier is generated based on the first dialogue request initiated by the user. The preparatory instruction sending module is used to parse the structured decision information to obtain the intent parameters of the current user. When the intent parameters meet the preset triggering conditions, it sends a preparatory instruction carrying the global session identifier to the cabinet fulfillment device so that the cabinet fulfillment device can start the fulfillment preparation process.

9. A linkage control device for a spatial retail warehouse, characterized in that, The device includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-7.

10. A storage medium, characterized in that, The storage medium stores a computer program that enables a computer to perform the method according to any one of claims 1-7.

Images