A natural language driven multi-robot collaborative delivery method and system

Abstract

The application discloses a natural language driven multi-robot cooperative distribution method and system, which receives a user natural language distribution instruction, judges whether task information is complete in combination with historical interaction information; when the information is incomplete, a clarification question is generated, and the task context is updated according to the supplementary information; after the information is complete, an atomic tool function is called to generate a structured distribution task, a task allocation result is determined according to a robot state and a path cost, and a collision-free distribution path is planned based on a safety interval and reserved occupation information; the task and the path are issued to the robot for execution, and execution feedback is received; when the task is changed or the execution is abnormal, the task context is updated, and the task is regenerated, the robot is allocated or the path is planned. The application realizes closed-loop conversion from a natural language instruction to multi-robot cooperative distribution execution.

Description

Technical Field

[0001] This invention belongs to the field of multi-robot collaborative control and artificial intelligence technology, specifically relating to a natural language-driven multi-robot collaborative delivery method and system. Background Technology

[0002] Multi-robot systems, which utilize the collaborative operation of multiple automated guided vehicles (AGVs), can accomplish tasks such as goods handling, order picking, and delivery. They are increasingly being applied in warehousing and logistics, smart libraries, and public services. Compared to requiring professionals to input structured data such as coordinates, task numbers, and path nodes, controlling multiple robots via natural language commands lowers the barrier to entry, allowing users to directly describe the delivery objects, quantities, and delivery locations.

[0003] However, user-input natural language commands often suffer from semantic ambiguity, missing information, or non-standard expression. For example, the command "process the items at the front desk" doesn't specify the items to be delivered, the target location, or the task priority; the word "popular" in "find two popular science fiction books and deliver them to table number one" needs further clarification based on book information. If the system directly generates robot tasks based on such commands, it's prone to problems like incorrect object identification, missing task parameters, or execution results inconsistent with the user's true intentions. Therefore, the system not only needs to understand user commands but also needs to determine the completeness of task information and clarify the user's intentions through multiple rounds of interaction when information is insufficient.

[0004] Existing robot task planning methods based on large language models (LLMs) can leverage the semantic understanding and reasoning capabilities of LLMs to decompose natural language instructions into several task steps. However, these methods typically emphasize high-level semantic planning, and some methods generate a complete task plan all at once, making them highly dependent on the explicitness and completeness of user instructions. When users supplement or modify tasks during execution, or when the robot encounters abnormal situations such as delivery failures or path blockages, the original task plan is difficult to be promptly and specifically corrected based on execution feedback. Furthermore, the task steps generated by large language models mainly reflect semantic rationality and cannot directly guarantee the rationality of robot task allocation or the actual executability of paths under time, space, and kinematic constraints.

[0005] On the other hand, existing multi-robot path planning methods typically use structured data such as the current position of each robot, task start point, task end point, and environmental map as input, and generate collision-free paths through methods such as conflict-based search (CBS) and safe interval path planning. Path planning algorithms such as SIPPwRT can also coordinate the occupancy time of different robots at various positions using safe intervals and reservation tables, and consider kinematic constraints such as robot turning, straight-line movement, and collision avoidance time offsets. However, the above methods themselves lack the ability to understand natural language semantics and clarify user intent, and cannot directly convert ambiguous natural language instructions into the parameters required for low-level planning, such as task start point, task end point, delivery object, and robot allocation relationships.

[0006] In existing technologies, even when combining large language models with robot path planning algorithms, the task results generated by the large language model are typically only passed one-way to the underlying planning module, lacking a two-way closed-loop interaction between high-level task understanding and low-level execution results. When the robot's state changes, the original planned path becomes unavailable, some tasks fail, or the user adjusts delivery requirements midway, it is difficult to synchronously complete task parameter correction, robot reassignment, and conflict-free path replanning based on the new execution information. This results in a disconnect between the natural language task planning results and the robot's execution process in the actual physical environment.

[0007] Therefore, how to interactively clarify and decompose ambiguous or incomplete natural language delivery instructions, convert the clarified user intent into structured tasks required for multi-robot scheduling, further complete robot task allocation and collision-free path planning that meets kinematic constraints, and dynamically correct tasks and paths based on user supplementary information and underlying execution feedback are technical problems that need to be solved in the field of multi-robot collaborative delivery. Summary of the Invention

[0008] The technical problem to be solved by this invention is to address the shortcomings of the prior art by providing a natural language-driven multi-robot collaborative delivery method and system. This method solves the problem that existing multi-robot delivery systems are unable to interactively clarify ambiguous or incomplete natural language instructions and convert them into executable structured delivery tasks. Furthermore, the high-level task planning is disconnected from the low-level robot task allocation, collision-free path planning under kinematic constraints, and execution feedback. This results in difficulties in timely task correction, robot reallocation, and path replanning in the event of task changes or execution anomalies.

[0009] The present invention adopts the following technical solution: A natural language-driven multi-robot collaborative delivery method includes the following steps: S1. Receive the user's natural language delivery instructions and obtain delivery environment map information and status information of multiple robots; S2. Construct a task context based on the delivery instructions and historical interaction information, and use a large language model to determine whether the delivery task information is complete; if it is incomplete, generate a clarification question, and update the task context according to the user's supplementary information until complete delivery task information is obtained. S3. Using the large language model to call predefined atomic tool functions, the delivery task information is converted into a set of structured delivery tasks, each of which includes a task start point and a task end point. S4. Based on the structured delivery task set and robot state information, calculate the path cost for each robot to execute each structured delivery task, and determine the task allocation result based on the path cost. S5. Based on map information, task allocation results, and robot's reserved occupancy information on map locations, determine the safe zone and plan a collision-free delivery path that meets kinematic constraints. S6. Send the task allocation results and collision-free delivery path to the corresponding robot and receive task execution feedback; S7. Upon receiving task change information or determining from task execution feedback that there is a task execution anomaly, update the task context and regenerate the structured delivery task, redetermine the task allocation result, and replan the collision-free delivery route, at least one of the following:

[0010] Furthermore, the task context includes at least one of historical natural language instructions, historical dialogue information, user supplementary information, and execution feedback from atomic tool functions.

[0011] Furthermore, the large language model generates interactive actions based on a reasoning-action-observation loop, which include intent clarification actions, atomic tool function call actions, and task correction actions.

[0012] Furthermore, the predefined atomic tool functions include at least one of the following: delivery object retrieval function, similar delivery object recommendation function, target location acquisition function, robot status acquisition function, task allocation function, and robot task issuance function.

[0013] Furthermore, the structured delivery task also includes the delivery object, the task starting point is the location of the delivery object, and the task ending point is the delivery location specified by the user.

[0014] Furthermore, the path cost is the shortest path distance on the map from the robot's current position to the task starting point of the corresponding structured delivery task, and the task allocation result is determined with the goal of minimizing the sum of the path costs corresponding to each robot.

[0015] Furthermore, determining the safe zone includes: dividing the time axis of each map location into continuous time intervals that are not occupied by robots according to the reserved occupancy information, and recording the time each robot occupies the map location through a reservation table.

[0016] Furthermore, the kinematic constraints include at least one of the robot's steering constraints, straight-line constraints, and collision avoidance time offset constraints; the collision-free delivery path includes the times when the robot arrives at and leaves each map location within the corresponding safe zone.

[0017] Furthermore, the task execution anomaly includes at least one of delivery task execution failure and delivery route blockage; when the existence of the task execution anomaly is determined, the task execution feedback is input into the large language model as new observation information to generate a task correction action.

[0018] Secondly, embodiments of the present invention provide a natural language-driven multi-robot collaborative delivery system, comprising: The instruction interaction module is used to receive the user's natural language delivery instructions, construct a task context based on the delivery instructions and historical interaction information, use a large language model to determine whether the delivery task information is complete, generate clarification questions when the delivery task information is incomplete, and update the task context based on the user's supplementary information. The task generation module is used to call predefined atomic tool functions using the large language model to convert complete delivery task information into a structured delivery task set; The task allocation module is used to calculate the path cost for each robot to execute each structured delivery task based on the structured delivery task set and the state information of multiple robots, and to determine the task allocation result based on the path cost. The path planning module is used to determine the safe zone and plan a collision-free delivery path that meets kinematic constraints based on the delivery environment map information, the task allocation results, and the robot's reserved occupancy information on map positions. The task execution module is used to send the task allocation results and the collision-free delivery path to the corresponding robot, and to receive the task execution feedback returned by the robot. The dynamic correction module is used to update the task context and control at least one of the task generation module, the task allocation module and the path planning module to re-execute the corresponding processing when receiving task change information or determining that there is a task execution abnormality based on the task execution feedback.

[0019] Thirdly, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described natural language-driven multi-robot collaborative delivery method.

[0020] Fourthly, embodiments of the present invention provide a computer-readable storage medium including a computer program, which, when executed by a processor, implements the steps of the above-described natural language-driven multi-robot collaborative delivery method.

[0021] Fifthly, a chip includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described natural language-driven multi-robot collaborative delivery method.

[0022] In a sixth aspect, embodiments of the present invention provide an electronic device including a computer program, which, when executed by the electronic device, implements the steps of the above-described natural language-driven multi-robot collaborative delivery method.

[0023] Compared with the prior art, the present invention has at least the following beneficial effects: A natural language-driven multi-robot collaborative delivery method combines the natural language understanding capabilities of a large language model with multi-robot task allocation and safe interval path planning. By performing integrity checks and interactive clarification on natural language delivery instructions, errors in identifying delivery objects, pickup locations, or delivery locations caused by ambiguous instruction meanings or missing task information can be reduced. Atomic tool functions convert the clarified natural language information into structured delivery tasks, allowing high-level semantic information to serve as input for lower-level scheduling. Path cost calculation, task allocation, and safe interval path planning enable multiple robots to execute deliveries according to corresponding tasks and collision-free paths. By reintegrating task change information and task execution feedback into the task context, the system can correct tasks in the event of instruction changes, delivery failures, or path blockages. The purpose of these technical measures is to form a closed loop between natural language interaction and actual robot execution, improving the executability, adaptability, and reliability of multi-robot delivery.

[0024] Furthermore, by incorporating historical natural language commands, historical dialogue information, user-supplemented information, and execution feedback from atomic tool functions into the task context, the large language model can make continuous judgments based on the interaction process preceding the current command and the tool execution results, avoiding task generation based solely on a single input. This configuration helps maintain the consistency of task semantics across multiple rounds of interaction and provides a complete information foundation for subsequent intent clarification, task generation, and dynamic correction.

[0025] Furthermore, by employing a reasoning-action-observation loop, the large language model can select actions such as intent clarification, atomic tool function calls, or task correction based on the current task context in each round of processing, and continue processing the action results as new observation information. This avoids generating a complete delivery plan all at once, allowing the system to progressively advance the task based on user responses and underlying execution results, thus improving the flexibility of handling complex, multi-step delivery tasks.

[0026] Furthermore, by setting atomic tool functions such as delivery object retrieval, similar delivery object recommendation, target location acquisition, robot status acquisition, task allocation, and robot task distribution, different delivery operations are encapsulated into structured, callable basic functions. This limits the scope of action of the large language model, transforming its output from open natural language programming into function calls with explicit parameters and execution objects. This helps reduce inconsistencies between task expression and the underlying system interface, and facilitates the expansion of tool functionality for different delivery scenarios.

[0027] Furthermore, by setting the delivery object, task start point, and task end point in the structured delivery task, key information in the natural language instructions, such as "what object to deliver, where to pick it up from, and where to deliver it," is converted into explicit scheduling parameters. This establishes a correspondence between user intent and robot path planning input, enabling the task allocation module and path planning module to perform subsequent processing based on a unified data format.

[0028] Furthermore, by using the shortest path distance on the map from the robot's current position to the task's starting point as the path cost, and minimizing the sum of the path costs for each robot as the task allocation objective, the pairing relationship between robots and delivery tasks can be determined by considering the current distribution of robots. This helps reduce the unnecessary movement distance required for robots to reach the pickup location, lowers the overall task execution cost, and avoids uneven robot utilization caused by assigning tasks solely based on fixed numbers or random methods.

[0029] Furthermore, by dividing the timeline of each map location into continuous safe zones not occupied by robots, and using a reservation table to record the time a robot occupies a map location, path planning considers not only whether a map location is passable, but also the time relationships between different robots arriving at the same location. This avoids multiple robots occupying the same map location at the same time, providing a basis for time coordination and collision avoidance between multiple robot paths.

[0030] Furthermore, by considering steering constraints, straight-line constraints, and collision avoidance time offset constraints in collision-free path planning, and determining the robot's arrival and departure times at each map position within the corresponding safe zone, the generated path simultaneously satisfies spatial accessibility, temporal coordination, and robot kinematic requirements. This reduces problems such as paths that cannot be practically executed or insufficient turning time resulting from planning solely based on the grid's shortest distance, and provides the robot with a certain margin for adjusting arrival and departure times.

[0031] Furthermore, by treating delivery task failures and delivery route blockages as task execution anomalies, and inputting the corresponding task execution feedback as new observation information into the large language model, the upper-level task processing can perceive the lower-level delivery status. A feedback channel is established between the robot's execution results and the upper-level decision-making module, enabling the system to generate corrective actions for anomalies instead of continuing to execute the original, invalid task plan, thus improving the delivery process's adaptability to environmental changes and execution anomalies.

[0032] It is understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0033] In summary, this invention combines the natural language understanding capabilities of a large language model with multi-robot task allocation, kinematic constraints, and safe zone path planning to achieve fuzzy instruction clarification, structured task generation, collision-free delivery, and anomaly feedback correction, while taking into account interactive flexibility, task executability, and multi-robot collaborative safety.

[0034] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the overall framework of the system described in this invention; Figure 2 These are photos of the actual deployment environment of the present invention in a real library setting. The left image shows the global visualization interface, and the right image shows a photo of the AGV performing a delivery task on site. Figure 3 A schematic diagram of a computer device provided in an embodiment of the present invention; Figure 4 This is a block diagram of a chip according to an embodiment of the present invention; Figure 5 This is a flowchart of the method of the present invention.

[0036] Among them, 60. Computer equipment; 61. Processor; 62. Memory; 63. Computer program; 600. Electronic device; 610. Processing unit; 620. Storage unit; 6201. Random access memory unit; 6202. Cache memory unit; 6203. Read-only memory unit; 6204. Program / utility; 6205. Program module; 630. Bus; 640. Display unit; 650. Input / output interface; 660. Network adapter; 700. External device. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0039] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0040] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0041] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0042] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0043] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0044] This invention provides a natural language-driven multi-robot collaborative delivery method that deeply integrates a large language model with the SIPPwRT algorithm. The upper layer utilizes ReAct's "reasoning-action-observation" loop to achieve proactive intent clarification and progressive task planning, solving the problem that existing LLM methods cannot handle ambiguous or incomplete instructions. The lower layer achieves globally optimal task allocation through minimum cost allocation of large model calls and uses SIPPwRT's safe interval mechanism to plan collision-free paths for multiple robots, balancing the flexibility of semantic understanding with the accuracy and safety of path execution. Furthermore, on a self-built simulated dataset, the system achieved an overall task success rate of 74.8%, validating the effectiveness of the hybrid framework. Simultaneously, multi-AGV delivery deployment was successfully completed in a real library scenario, demonstrating the feasibility and robustness of this invention in real-world physical environments.

[0045] Please see Figure 5 This invention provides a natural language-driven multi-robot collaborative delivery method, which is applied to a multi-robot collaborative delivery system including an upper-layer task planning part and a lower-layer scheduling and execution part.

[0046] The upper-level task planning component includes a large language model and an atomic tool function library, which are used to receive natural language delivery instructions input by users, judge and clarify user intentions, and generate corresponding atomic tool function calls based on the clarified delivery task information.

[0047] The underlying scheduling and execution part is used to obtain delivery environment map information and the status information of multiple robots, allocate tasks to robots according to structured delivery tasks, and generate collision-free delivery paths for each robot using the safe interval path planning method.

[0048] In this embodiment, the robot can be an automated guided vehicle (AGV). Multiple AGVs are individually connected to the system to receive task allocation results and delivery routes, and return their current location, operating status, and task execution results to the system. The delivery environment can be a warehousing and logistics scenario, a smart library scenario, or other scenarios that require multiple mobile robots to collaborate in picking up and delivering goods.

[0049] In this embodiment, the large language model can be a pre-trained model with natural language understanding, function calling, and multi-turn dialogue capabilities.

[0050] After the system starts, it loads the large language model, atomic tool function library, and delivery environment map information, and obtains the status information of each AGV. The status information should at least reflect the current position of the AGV and whether it is available to perform delivery tasks.

[0051] Users input natural language delivery instructions via an interactive terminal. For example, in a library delivery scenario, a user might input "Find two popular science fiction books and deliver them to table number one." The system then combines this natural language delivery instruction with previous historical interaction information to form the task context.

[0052] The large language model determines whether the information required for the delivery task is complete based on the task context. The information required for a delivery task may include the object to be delivered, the quantity, the task start point, and the task end point. For information that can be obtained through atomic utility functions, the user is not required to input coordinates directly; instead, the system calls the corresponding utility function to query based on the natural language description.

[0053] When a user inputs "Please process the items at the front desk," the instruction does not specify the items to be processed or the target delivery location. The large language model generates a clarifying question to inquire about the items requiring delivery and their destination. After the user replies, the system adds this reply as supplementary information to the task context, and the large language model then re-evaluates the completeness of the task information.

[0054] The above process follows a reasoning-action-observation loop. Specifically, the large language model infers and generates actions based on the current task context; when the action is an intent-clarification action, the system outputs a clarifying question to the user; the user's response constitutes new observation information and is used to update the task context for the next round. This loop continues until complete delivery task information sufficient to generate a structured delivery task is obtained.

[0055] After confirming that the delivery task information is complete, the large language model generates atomic tool function calls sequentially according to the current task context.

[0056] The atomic utility function library may include: The delivery object retrieval function is used to retrieve delivery objects based on keywords, authors, genres, ratings, or availability status. A similar delivery object recommendation function is used to recommend similar objects based on the type, style, content, or theme of the reference object; The target location retrieval function is used to retrieve the location of the delivery object; The robot state acquisition function is used to obtain the current state information of each robot. The task assignment function is used to calculate the task assignment result based on the target position and robot state; The robot task dispatch function is used to send delivery tasks and corresponding routes to the robot.

[0057] In the library scenario, for the instruction "find two popular science fiction books and deliver them to table number one", the large language model can first call the delivery object retrieval function to obtain candidate books based on conditions such as "science fiction", "popular" and "available for borrowing"; then call the target location retrieval function to determine the location of the selected books on the bookshelf; and finally combine the location of "table number one" to form a structured delivery task.

[0058] Each structured delivery task includes at least a task start point and a task end point. For a book delivery task, the task start point is the location of the corresponding book on the bookshelf, and the task end point is the table number one location specified by the user. A structured delivery task may also include the delivery object to identify the specific book that needs to be picked up by the AGV.

[0059] All atomic utility functions are called using a structured format. The large language model determines the next action based on the execution results returned by the utility functions, rather than generating all execution steps at once. Therefore, the delivery object retrieval results, target location, and robot state can sequentially enter the subsequent scheduling process.

[0060] After obtaining the set of structured delivery tasks, the system obtains the set of currently available AGVs and the current position of each AGV.

[0061] For each task in the structured delivery task set, the system calculates the path cost from the current position of each AGV to the starting point of the task. In this embodiment, the path cost is the shortest path distance in the delivery environment map.

[0062] For example, when there are multiple tasks to be delivered and multiple available AGVs, the system generates the path cost for each AGV to execute each delivery task, and calls the task allocation function to determine the pairing relationship between tasks and AGVs with the goal of minimizing the sum of the path costs corresponding to each AGV.

[0063] The task allocation results are used to indicate which AGV will perform each structured delivery task. This task allocation process mainly considers the movement cost of the AGV to reach the task's starting point, giving priority to AGVs that are closer to the corresponding task's starting point, while also coordinating the overall pairing relationships between multiple tasks and multiple AGVs.

[0064] After the task allocation results are determined, the system plans delivery routes from the current position to the task start point and from the task start point to the task end point for each AGV.

[0065] The delivery environment map can be represented as a grid map composed of multiple map locations. The system establishes a reservation table based on the planned paths of each AGV and the time it occupies map locations. The reservation table is used to record the reservation and occupancy information of each map location by each AGV within the corresponding time period.

[0066] For any given map location, the system divides the timeline of that location into several continuous time intervals that are not occupied by other AGVs, based on the reserved occupancy information, and uses these continuous time intervals as the safe intervals for the corresponding map location.

[0067] During path planning, the system searches between map locations and their safe zones, ensuring that AGVs only enter and leave the corresponding map locations within the designated safe zones. For time slots already reserved by other AGVs, the current AGV will not enter the same location during that time slot, thus preventing multiple AGVs from occupying the same map location simultaneously.

[0068] When determining the time it takes for an AGV to move from one map location to an adjacent map location, the system further considers kinematic constraints such as the AGV's turning, straight-line movement, and collision avoidance time offset. The generated collision-free delivery path includes not only the map locations traversed sequentially, but also the times when the AGV arrives at and leaves each map location.

[0069] Since the same map location can have different safety zones, AGVs can adjust their arrival or departure times while meeting collision avoidance requirements. When a predetermined location is temporarily occupied by another AGV, the current AGV can wait at the previous safety location and enter that location in the subsequent safety zone.

[0070] The system coordinates the paths of multiple AGVs according to the task allocation order and updates the occupancy information corresponding to the planned paths using a reservation table, so that the path planning of subsequent AGVs can avoid the space-time positions that have been reserved by the AGVs.

[0071] After completing task allocation and collision-free path planning, the system calls the robot task dispatch function to send the corresponding structured delivery task and collision-free delivery path to each AGV.

[0072] Each AGV moves to the corresponding task start point according to the received path, retrieves the object to be delivered, and then moves to the task end point to complete the delivery. During the execution process, each AGV returns task execution feedback to the system. The task execution feedback may include whether the task was completed, whether the delivery failed, and whether the path was blocked.

[0073] The system displays the task execution status of each AGV on a global visualization interface and updates the robot status information and the reserved occupancy information of the map position based on the execution feedback. Figure 2 The deployment environment of the present invention in a real library setting is shown, wherein a global visualization interface is used to display the operating status of the delivery system, and on-site AGVs are used to perform corresponding book delivery tasks.

[0074] During the execution of a delivery task, users can supplement or modify the original delivery instructions. For example, users can change the delivery recipient, delivery quantity, or delivery destination. After receiving the task change information, the system adds the task change information to the current task context.

[0075] When the AGV returns a delivery task that has failed or the delivery path is blocked, or when other task execution errors occur, the system adds the corresponding task execution feedback as new observation information to the task context.

[0076] The large language model re-executes the reasoning-action-observation loop based on the updated task context and generates task-correction actions according to task changes or anomaly types. When the delivery object changes, the delivery object retrieval function and target location acquisition function can be called again to regenerate the structured delivery task; when the robot state changes or the originally assigned robot cannot continue to perform the task, the task allocation function can be called again to redetermine the task allocation result; when the path in the map is blocked, the safe zone can be re-determined based on the updated reserved occupancy information and a collision-free delivery path can be replanned.

[0077] Therefore, user-supplemented information and feedback from the underlying task execution can both have a reverse effect on the upper-level task planning, making natural language understanding, structured task generation, robot task allocation, and path planning a closed-loop processing process.

[0078] In another embodiment of the present invention, a natural language-driven multi-robot collaborative delivery system is provided. This system can be used to implement the above-mentioned natural language-driven multi-robot collaborative delivery method. Specifically, the natural language-driven multi-robot collaborative delivery system includes an instruction interaction module, a task generation module, a task allocation module, a path planning module, a task execution module, and a dynamic correction module.

[0079] The instruction interaction module is used to receive the user's natural language delivery instructions, construct a task context based on the delivery instructions and historical interaction information, use a large language model to determine whether the delivery task information is complete, generate clarification questions when the delivery task information is incomplete, and update the task context based on the user's supplementary information. The task generation module is used to call predefined atomic tool functions using the large language model to convert complete delivery task information into a structured delivery task set; The task allocation module is used to calculate the path cost for each robot to execute each structured delivery task based on the structured delivery task set and the state information of multiple robots, and to determine the task allocation result based on the path cost. The path planning module is used to determine the safe zone and plan a collision-free delivery path that meets kinematic constraints based on the delivery environment map information, the task allocation results, and the robot's reserved occupancy information on map positions. The task execution module is used to send the task allocation results and the collision-free delivery path to the corresponding robot, and to receive the task execution feedback returned by the robot. The dynamic correction module is used to update the task context and control at least one of the task generation module, the task allocation module and the path planning module to re-execute the corresponding processing when receiving task change information or determining that there is a task execution abnormality based on the task execution feedback.

[0080] This invention provides a terminal device comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or function. The processor described in this embodiment can be used in the operation of a natural language-driven multi-robot collaborative delivery method, including: The system receives user natural language delivery instructions and acquires delivery environment map information and multiple robot status information. Based on the delivery instructions and historical interaction information, it constructs a task context and uses a large language model to determine the completeness of the delivery task information. If incomplete, it generates clarification questions and updates the task context based on user-supplemented information until complete delivery task information is obtained. Using the large language model, it calls predefined atomic tool functions to convert the delivery task information into a set of structured delivery tasks, each including a task start point and a task end point. Based on the set of structured delivery tasks and robot status information, it calculates the path cost for each robot to execute each structured delivery task and determines the task allocation result based on the path cost. Based on map information, task allocation results, and robot reserved map positions, it determines a safe zone and plans a collision-free delivery path that satisfies kinematic constraints. It sends the task allocation result and collision-free delivery path to the corresponding robot and receives task execution feedback. Upon receiving task change information or determining a task execution anomaly based on task execution feedback, it updates the task context and regenerates at least one of the following: re-generating a structured delivery task, re-determining the task allocation result, and re-planning a collision-free delivery path.

[0081] Please see Figure 3The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the natural language-driven multi-robot collaborative delivery method described in this embodiment. To avoid repetition, details are omitted here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the natural language-driven multi-robot collaborative delivery system described in this embodiment. To avoid repetition, details are omitted here.

[0082] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 3 This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.

[0083] The processor 61 may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0084] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.

[0085] Furthermore, the memory 62 may include both internal storage units of the computer device 60 and external storage devices. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0086] Please see Figure 4 The terminal device is an electronic device 600, which is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including storage unit 620 and processing unit 610), a display unit 640, etc.

[0087] The storage unit stores program code, which can be executed by the processing unit 610 to perform the steps described in the above-described method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 610 can perform the steps shown in Figure 5.

[0088] Storage unit 620 may include readable media in the form of volatile storage units, such as random access memory (RAM) 6201 and / or cache memory 6202, and may further include read-only memory (ROM) 6203.

[0089] Storage unit 620 may also include a program / utility 6204 having a set (at least one) program module 6205, such program module 6205 including but not limited to: 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.

[0090] Bus 630 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0091] Electronic device 600 can also communicate with one or more external devices 700 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other computing devices (e.g., router, modem). This communication can be performed via input / output interface 650. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network, wide area network, and / or public network, such as the Internet) via network adapter 660. Network adapter 660 can communicate with other modules of electronic device 600 via bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0092] This invention also provides a storage medium, specifically a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the terminal device and extended storage media supported by the terminal device; it can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). More specific examples of the computer-readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical fiber, portable compact disk read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0093] Computer-readable storage media also include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium can also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, radio frequency, etc., or any suitable combination thereof.

[0094] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0095] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the natural language-driven multi-robot collaborative delivery method in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps: The system receives user natural language delivery instructions and acquires delivery environment map information and multiple robot status information. Based on the delivery instructions and historical interaction information, it constructs a task context and uses a large language model to determine the completeness of the delivery task information. If incomplete, it generates clarification questions and updates the task context based on user-supplemented information until complete delivery task information is obtained. Using the large language model, it calls predefined atomic tool functions to convert the delivery task information into a set of structured delivery tasks, each including a task start point and a task end point. Based on the set of structured delivery tasks and robot status information, it calculates the path cost for each robot to execute each structured delivery task and determines the task allocation result based on the path cost. Based on map information, task allocation results, and robot reserved map positions, it determines a safe zone and plans a collision-free delivery path that satisfies kinematic constraints. It sends the task allocation result and collision-free delivery path to the corresponding robot and receives task execution feedback. Upon receiving task change information or determining a task execution anomaly based on task execution feedback, it updates the task context and regenerates at least one of the following: re-generating a structured delivery task, re-determining the task allocation result, and re-planning a collision-free delivery path.

[0096] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0097] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0098] This invention provides a multi-robot collaborative delivery system based on ReAct-SIPPwRT joint decision-making. The system includes an upper-level large language model planning module, an atomic tool function library, and an lower-level task allocation and path planning module. The implementation method is described in detail below with reference to each step: S1. Upper-level task understanding: Multi-turn interactive intent clarification and task decomposition based on ReAct; Please see Figure 1 The upper-level planning module adopts the ReAct paradigm, interleaving reasoning trajectories with task-related actions. Let the... Step state is ,in For historical commands and dialogue context, This provides feedback on the execution of the previous action or supplementary information from the user. The large language model is based on the strategy. Generate Action Its types include These correspond to intent clarification, function call, and dynamic correction, respectively.

[0099] S101, Multi-round Dialogue-based Clarification of Intent When user instructions contain missing or ambiguous information (e.g., "Please handle the items at the front desk"), the model uses its own reasoning to determine whether the information is sufficient; if insufficient, it generates a clarifying action. Ask users questions (including) (For user input), user responses as new observations renew The model is then re-evaluated. This loop continues until the model deems the intent clear and proceeds to the function call phase.

[0100] S102, Inference-driven step-by-step function calls Once the intent is clear, the model generates an atomic function call at each step. ,in This is a predefined utility function (see S2 for details). After execution, the system returns an observation. (For example, the result of function execution, robot status), the model infers the next action based on this, until the task objective is achieved or user intervention is required.

[0101] S103, Feedback-based dynamic correction During execution, the model receives two types of feedback as new observations—(1) information supplemented or modified by the user; and (2) feedback on the underlying execution results (such as delivery failure or route blockage). In the next step of inference, the model detects changes or anomalies and generates corrective actions, i.e. ,in Indicates context, Input for users This indicates the function feedback during task execution, which involves re-calling the relevant function to modify parameters or reassign the robot.

[0102] S2. Designing an atomic utility function library To support the execution of actions within the ReAct framework, this invention designs a set of structured utility functions that encapsulate the basic operations for interacting with the multi-machine delivery system. The function library is shown in Table 1.

[0103] Table 1

[0104] All function calls are completed in structured JSON format. The large model generates function call requests or natural language queries that conform to the format based on the ReAct framework.

[0105] S3, Low-level scheduling: Centralized task allocation and SIPPwRT conflict-free path planning After the upper layer of ReAct completes intent parsing and structured task generation, it enters the lower layer scheduling stage.

[0106] Let the current set of tasks to be delivered be... Each task Including starting point (Book pick-up point) and destination (Delivery point); Available AGV collections are , Current location is Define the single-machine execution cost. (Shortest path distance on the map).

[0107] The large model calls the atomic function calculate_optimal_allocation to solve for the globally optimal task-AGV pairing scheme with the goal of minimizing the total cost.

[0108] After allocation, the system calls the SIPPwRT algorithm to plan a collision-free path for each AGV. SIPPwRT, based on a safety zone mechanism, divides the time axis of each grid point into unoccupied continuous windows and coordinates multiple AGV occupancy through a reservation table. Simultaneously, it introduces kinematic constraints (turning, straight-line travel, and collision avoidance time offsets) to generate the shortest time path that satisfies speed and geometric constraints, allowing AGVs to flexibly adjust their arrival and departure times within the safety zone, providing time redundancy and fault tolerance. Finally, the system sends the paths for each AGV for execution via `send_agvs_task_workflow`.

[0109] S4. System Deployment and Operation Process In practical applications, users interact with the system via natural language. After system startup, the large model loads pre-trained weights (e.g., Qwen3-instruct-4B) and predefined atomic tool libraries and map information. After the user inputs a command, the system executes steps S1 to S3 until the task is completed or manual intervention is required. The entire process supports user modifications to commands mid-process, and the system re-plans its workflow through a dynamic correction mechanism.

[0110] This invention was tested on a self-built simulated dataset containing 900 test cases, covering intent types such as book recommendation, delivery, and request clarification. Experimental results show an overall task success rate of 74.8%, with multi-step, multi-function call tasks achieving a success rate as high as 96.1%, verifying the effectiveness of the ReAct mechanism in multi-round reasoning and complex task execution. Experimental results are shown in Appendix Table 2; Table 2

[0111] Meanwhile, the invention deployed four AGVs in a real library setting. Through multiple rounds of dialogue, the system clarified ambiguous user commands (such as "find two popular science fiction books and deliver them to table number 1"). The system successfully completed the multi-round interaction and conflict-free delivery task, further verifying the practicality and robustness of the invention.

[0112] In summary, this invention provides a natural language-driven multi-robot collaborative delivery method and system, establishing a closed-loop collaborative delivery mechanism comprised of natural language interaction, structured task generation, robot task allocation, safe zone path planning, and execution feedback correction. The system first determines the completeness of the user's delivery instructions based on historical interaction information. When semantic ambiguity or missing information exists, it proactively generates clarifying questions and obtains complete task information through multiple rounds of interaction. Subsequently, it uses predefined atomic tool functions to convert the user's intent into a structured task containing the delivery object, task start point, and task end point, reducing inconsistencies between open natural language output and the underlying execution interface. In terms of underlying scheduling, the system calculates the path cost based on the robot's current position and task start point, determines the task allocation relationship with a low overall movement cost, and coordinates the time occupied by different robots on map positions using safe zones and reservation tables. It also considers kinematic constraints such as turning, straight-line movement, and collision avoidance time offsets, thereby generating a collision-free delivery path that can be actually executed by the robots. When the user changes the task midway, or when a robot experiences task failure, path blockage, or other anomalies, the system feeds relevant information back to the large language model as new observations and regenerates the task, adjusts robot allocation, or updates the delivery path. Therefore, this invention balances the flexibility of natural language interaction, the accuracy of underlying scheduling, and the safety of multi-robot execution, thereby improving the task adaptability, collaboration efficiency, and operational reliability in complex delivery scenarios.

[0113] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0114] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0115] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0116] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and 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 through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0117] 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.

[0118] Furthermore, 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0119] If the integrated module / unit 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, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random-access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0120] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It will 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... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0121] 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.

[0122] 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.

[0123] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A natural language-driven multi-robot collaborative delivery method, characterized in that, Includes the following steps: S1. Receive the user's natural language delivery instructions and obtain delivery environment map information and status information of multiple robots; S2. Construct a task context based on the delivery instructions and historical interaction information, and use a large language model to determine whether the delivery task information is complete; if it is incomplete, generate a clarification question, and update the task context according to the user's supplementary information until complete delivery task information is obtained. S3. Using the large language model to call predefined atomic tool functions, the delivery task information is converted into a set of structured delivery tasks, each of which includes a task start point and a task end point. S4. Based on the structured delivery task set and robot state information, calculate the path cost for each robot to execute each structured delivery task, and determine the task allocation result based on the path cost. S5. Based on map information, task allocation results, and robot's reserved occupancy information on map locations, determine the safe zone and plan a collision-free delivery path that meets kinematic constraints. S6. Send the task allocation results and collision-free delivery path to the corresponding robot and receive task execution feedback; S7. Upon receiving task change information or determining from task execution feedback that there is a task execution anomaly, update the task context and regenerate the structured delivery task, redetermine the task allocation result, and replan the collision-free delivery route, at least one of the following:

2. The natural language-driven multi-robot collaborative delivery method according to claim 1, characterized in that, The task context includes at least one of the following: historical natural language instructions, historical dialogue information, user-added information, and execution feedback from atomic tool functions.

3. The natural language-driven multi-robot collaborative delivery method according to claim 1, characterized in that, The large language model generates interactive actions based on a reasoning-action-observation loop, which include intent clarification actions, atomic tool function call actions, and task correction actions.

4. The natural language-driven multi-robot collaborative delivery method according to claim 1, characterized in that, The predefined atomic tool functions include at least one of the following: delivery object retrieval function, similar delivery object recommendation function, target location acquisition function, robot status acquisition function, task allocation function, and robot task issuance function.

5. The natural language-driven multi-robot collaborative delivery method according to claim 1, characterized in that, The structured delivery task also includes the delivery object, the task start point is the location of the delivery object, and the task end point is the delivery location specified by the user.

6. The natural language-driven multi-robot collaborative delivery method according to claim 1, characterized in that, The path cost is the shortest path distance on the map from the robot's current position to the starting point of the corresponding structured delivery task, and the task allocation result is determined with the goal of minimizing the sum of the path costs corresponding to each robot.

7. The natural language-driven multi-robot collaborative delivery method according to claim 1, characterized in that, Determining the safe zone includes: dividing the time axis of each map location into continuous time intervals that are not occupied by robots according to the reserved occupancy information, and recording the time each robot occupies the map location through a reservation table.

8. The natural language-driven multi-robot collaborative delivery method according to claim 1, characterized in that, The kinematic constraints include at least one of the robot's steering constraints, straight-line constraints, and collision avoidance time offset constraints; the collision-free delivery path includes the times when the robot arrives at and leaves each map location within the corresponding safe zone.

9. The natural language-driven multi-robot collaborative delivery method according to claim 1, characterized in that, The task execution anomaly includes at least one of delivery task execution failure and delivery route blockage; when the task execution anomaly is determined to exist, the task execution feedback is input into the large language model as new observation information to generate a task correction action.

10. A natural language-driven multi-robot collaborative delivery system, characterized in that, include: The instruction interaction module is used to receive the user's natural language delivery instructions, construct a task context based on the delivery instructions and historical interaction information, use a large language model to determine whether the delivery task information is complete, generate clarification questions when the delivery task information is incomplete, and update the task context based on the user's supplementary information. The task generation module is used to call predefined atomic tool functions using the large language model to convert complete delivery task information into a structured delivery task set; The task allocation module is used to calculate the path cost for each robot to execute each structured delivery task based on the structured delivery task set and the state information of multiple robots, and to determine the task allocation result based on the path cost. The path planning module is used to determine the safe zone and plan a collision-free delivery path that meets kinematic constraints based on the delivery environment map information, the task allocation results, and the robot's reserved occupancy information on map positions. The task execution module is used to send the task allocation results and the collision-free delivery path to the corresponding robot, and to receive the task execution feedback returned by the robot. The dynamic correction module is used to update the task context and control at least one of the task generation module, the task allocation module and the path planning module to re-execute the corresponding processing when receiving task change information or determining that there is a task execution abnormality based on the task execution feedback.

Images