Robot navigation method and system based on MCP architecture

Abstract

The invention discloses a robot navigation method and system based on an MCP architecture. The method comprises the following steps: firstly, receiving a user instruction through voice recognition, and analyzing the instruction into an MCP request instruction by adopting a large language model; secondly, generating an optimized path on the static map based on an improved A * algorithm; a dynamic cost map is constructed through real-time point cloud data, a double-layer fusion mechanism of a global buffer map and a local cost map is innovatively adopted, double-rate attenuation is introduced to realize intelligent management of obstacle information, and the problems of jitter and ghosting of dynamic obstacle perception are effectively solved; and finally, a navigation instruction is executed based on an MCP protocol, and dynamic obstacle avoidance and accurate path tracking are realized in combination with a three-level re-planning mechanism. According to the invention, an MCP protocol is deeply integrated, full-process automation from voice analysis, path planning to motion execution is realized through a standardized instruction system, and the interaction intelligence and navigation reliability of the navigation robot in complex environments such as an exhibition hall are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of mobile robot navigation and human-computer interaction technology, and particularly relates to a robot navigation method and system based on the MCP (Model Context Protocol) architecture. Background Technology

[0002] Existing tour guide robots mostly adopt a hierarchical paradigm of "global static planning + local dynamic obstacle avoidance": the upper layer solves the global path on a static map, while the lower layer avoids obstacles in real time based on sensors and outputs motion control. A common industrial production line uses "A* global + DWA local," which involves using A* to search for an approximate shortest path on a two-dimensional grid, and then using Dynamic Windowing (DWA) to sample and predict the velocity space, integrating target orientation, obstacle avoidance, and velocity efficiency scoring to select the current velocity command. This approach is simple to implement and highly real-time, but it still has shortcomings in engineering problems such as dynamic passenger flow fluctuations, multi-round explanation program arrangement, and cross-module constraint consistency. The uniform decay of a single-layer cost map can easily create "ghosting" or "over-clearing," leading to frequent replanning and tracking oscillations.

[0003] Existing publicly available solutions improve overall performance by using "RRT* to generate a static near-optimal path → ant colony algorithm for secondary optimization (improving smoothness, reducing length and inflection points) → combining with DWA for dynamic obstacle avoidance," and provide parameter settings, procedures, and simulation results, making them suitable for museum navigation applications. Their value lies in improving the quality of static paths and dynamic safety, but focusing on the concatenation of algorithm layers, they still fall short in addressing systemic issues such as "protocol standardization," "cross-layer rate / constraint consistency," "dynamic obstacle 'ghosting' jitter management," and "stable switching after replanning."

[0004] The classic A* algorithm can search for approximate shortest paths on a static 2D grid using distance-based heuristics (Euclidean / Manhattan). However, when used in conjunction with DWA, it often relies on single-layer cost map rolling updates and uniform decay, failing to reflect chassis steering / curvature constraints. This can easily produce geometrically reachable but kinematically difficult-to-track polygonal paths, resulting in high local layer correction pressure and a tendency to trigger oscillations and re-planning at narrow channels / sharp bends. Summary of the Invention

[0005] To address the shortcomings of existing "A* global + DWA local" pipelines in cross-layer collaboration, dynamic jitter suppression, replanning stability, and business loop closure, this invention aims to solve the following problems: 1) Existing technologies lack protocol-based standardization across the entire process: There is a lack of unified semantic / routing / flow control and idempotency mechanisms between voice semantics and navigation control; new plugins and business functions (explanation, statistics, safety monitoring) significantly intrude on the navigation link; 2) Existing technologies lack sufficient global path tracing capability: Classic A* does not explicitly incorporate chassis orientation / turning radius constraints, leading to high-frequency backend control actions and large curvature peaks; 3) Dynamic obstacle "ghosting" and perception jitter: Uniform attenuation in a single-layer costmap makes it difficult to distinguish between short-term false detections and stable obstacles, easily leading to frequent switching and unnecessary replanning; 4) Scattered replanning triggers: The lack of tiered backoff and unlocking criteria makes it difficult to control state oscillations and stagnation duration. Therefore, this invention proposes a robot navigation method and system based on the MCP architecture.

[0006] To achieve the above-mentioned objectives, the present invention specifically adopts the following technical solution:

[0007] In a first aspect, the present invention provides a robot-guided tour method based on an MCP architecture, comprising the following steps:

[0008] S1. The robot pose and keyframes are output using a laser-IMU tightly coupled odometry method. Loop closure detection and pose optimization are then used to eliminate accumulated errors, generating a high-precision real-time 3D point cloud map and a 2D grid map. Fixed coordinates and theme labels are configured for each booth in the exhibition area. Dynamic features generated during object movement are removed from the 2D grid map and settled into static occupied grids. At the same time, global buffer map parameters and local cost map parameters are initialized.

[0009] S2. When the robot receives a natural language service request from the client, it converts the request into text and parses it into a structured MCP request using a large language model. The MCP-Bus assigns a corresponding ID to the MCP request and performs routing, retries, and deduplication. When the MCP request is a query command, it outputs language signals to interact with the user, thereby completing the explanation and / or retrieval service. When the MCP request is a control command, it initializes the 3D point cloud map information, using the robot's current location as the starting point and the target display case location specified in the control command as the ending point, to prepare path planning.

[0010] S3. An improved A* algorithm based on orientation awareness is used to search for the optimal global path on a two-dimensional grid map; the heuristic function of the improved A* algorithm is obtained by adding a heading cost term to the heuristic function of the A* algorithm.

[0011] S4. A dynamic cost map is constructed in the robot coordinate system using point clouds collected by LiDAR. A two-layer fusion mechanism combining a global buffer map and a local cost map is used, and a dual-rate attenuation strategy is introduced to manage dynamic obstacle information. On the pure tracking controller baseline, path following and dynamic obstacle avoidance are achieved by combining parameter constraints issued by the MCP parameter table. During dynamic obstacle avoidance, motion execution and three-level replanning are adopted: when the dynamic cost map indicates that forward movement is blocked or the safety margin is less than the preset safety distance threshold, it is triggered step by step in the order of first-level trajectory sampling temporary obstacle avoidance, second-level local A* replanning, and third-level global A* replanning. When the previous level intervention is successful or the blocking condition is removed, it retreats to the next level or returns to the tracking controller baseline.

[0012] S5. Once the robot reaches the destination, it triggers an interaction service and returns to S2 after the interaction service ends, waiting for the next natural language service request.

[0013] Based on the above scheme, each step can be implemented in the following preferred manner.

[0014] As a preferred embodiment of the first aspect above, step S1 further includes distributing the robot's speed limit, acceleration limit, angular velocity limit, curvature limit, safety radius, sensing parameters, and attenuation parameters through the MCP parameter table.

[0015] As a preferred embodiment of the first aspect mentioned above, in step S2, the fields of the MCP request include {intent, target, deadline, constraints, context}; where intent represents the user's intent; target represents the target display location; deadline represents the timeliness constraint of the instruction; constraints represent the execution priority order constraint; and context represents the connecting context information.

[0016] As a preferred embodiment of the first aspect mentioned above, in step S3, the heuristic function of the A* algorithm is improved. Calculate using the following formula:

[0017] ;

[0018] in, Indicates the robot's current location To the target display case location The Euclidean distance; The angle between the robot's current orientation and the direction of the target display case; and All are weight coefficients greater than 0; This is the cost item for the course.

[0019] As a preferred embodiment of the first aspect mentioned above, in step S4, the multi-frame point cloud is aggregated through spatiotemporal stabilization to obtain a global buffer map; the near-field obstacles are updated in real time using the laser point cloud in the robot coordinate system to generate a local cost map; different exponential decay rates are applied to the global buffer map and the local cost map respectively, and the exponential decay rate of the global buffer map is less than that of the local cost map, and the timestamps of the global buffer map and the local cost map are unified as the fusion gating to form a dual-rate decay strategy.

[0020] As a preferred embodiment of the first aspect above, in step S4, the pure tracking controller uses a combined strategy of forward view point and curvature: the forward view distance is positively correlated with the linear velocity and is constrained by the upper limit of curvature; the rate of change of rotation angle is uniformly constrained by the MCP parameter table; wherein, the forward view distance is the safe distance between the robot and the waypoint.

[0021] As a preferred embodiment of the first aspect above, in the dynamic obstacle avoidance step S4, the corresponding cost function for each candidate anchor point sequence is calculated using the following formula. And select the candidate anchor sequence with the smaller cost function value for output:

[0022] ;

[0023] in, Weighting coefficients representing the heading; This indicates the angle of deviation between the current heading and the target heading; The weighting coefficient represents the minimum safe distance. This represents the minimum safe distance, which is the distance between the robot and its nearest obstacle. This indicates a correction term to prevent division by zero; Weighting coefficients representing speed; This represents the velocity stability constraint term; Weighting coefficients representing curvature; This represents the path smoothness constraint.

[0024] As a preferred embodiment of the first aspect, in step S4, the triggering condition for the third-level replanning is: when the minimum safe distance is less than a preset safe distance threshold, the local cost map is resampled to generate a new path, thereby achieving the purpose of temporarily bypassing the obstacle; when the continuous If all cycles fail to bypass obstacles temporarily, or if the path feasibility score falls below a preset feasibility threshold, the improved A* algorithm is re-executed on the local cost map for local path planning; when consecutive If local replanning fails in all cycles, or if the planned path cannot reach the destination, then backtrack to the starting point and re-execute the improved A* algorithm for global path planning. All are preset number of cycles.

[0025] Secondly, the present invention provides a robot-guided tour system based on the MCP architecture, comprising:

[0026] The data acquisition module is used to acquire natural language service requests sent by the client;

[0027] The result acquisition module is used to process the received natural language service request according to the robot navigation method based on the MCP architecture described in any of the above-mentioned first aspects. When it is a query command, it outputs language signals to interact with the user; when it is a control command, it performs path planning and completes the user's service request after the robot reaches the target display case location.

[0028] Thirdly, the present invention provides a computer electronic device, which includes a memory and a processor;

[0029] The memory is used to store computer programs;

[0030] The processor is configured to, when executing the computer program, implement a robot navigation method based on an MCP architecture as described in any of the solutions of the first aspect above.

[0031] Compared with the prior art, the present invention has the following advantages:

[0032] This invention standardizes and orchestrates the voice-semantic-navigation-control-service transition through MCP, allowing new services to be added to the navigation link with zero intrusion. The improved A* incorporates chassis constraints during the search phase, significantly reducing backend curvature peaks and control jitter. Through GBM+LCM fusion and dual-rate attenuation, it can distinguish between short-term false detections and steady-state obstacles, reducing unnecessary replanning and stagnation. Through first-level trajectory sampling for temporary obstacle avoidance, second-level local A* replanning, and third-level global A* replanning, the "obstacle avoidance → replanning → unlocking" process is interpretable and parameter-tunable. Experiments demonstrate that in dynamic passenger flow scenarios, the method of this invention reduces the number of replanning attempts, lateral error variance, and stagnation duration compared to the A* baseline. Furthermore, compared to "RRT* + ant colony + DWA," the method of this invention avoids the link complexity caused by secondary optimization required for random sampling paths. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method of the present invention;

[0034] Figure 2 This is a timing diagram of the MCP request / response and parameter table of the present invention;

[0035] Figure 3 This is a schematic diagram of a 3D point cloud map generated according to an embodiment of the present invention;

[0036] Figure 4This is a two-dimensional grid map and a schematic diagram of the display case's fixed location generated according to an embodiment of the present invention;

[0037] Figure 5 This is a schematic diagram of global path planning in an embodiment of the present invention;

[0038] Figure 6 This is a schematic diagram of dynamic obstacle avoidance based on a local cost map in an embodiment of the present invention;

[0039] Figure 7 This is a system block diagram of the present invention. Detailed Implementation

[0040] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in the various embodiments of the present invention can be combined accordingly without mutual conflict.

[0041] In the description of this invention, it should be understood that the terms "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" and "second" may explicitly or implicitly include at least one of those features.

[0042] like Figure 1 , Figure 2 As shown, in a preferred embodiment of the present invention, the robot navigation method based on the MCP architecture includes the following steps S1 to S5. The specific implementation process of each step will be described in detail below.

[0043] S1. The robot pose and keyframes are output using the laser-IMU tightly coupled odometry method (Fast-LIO2). Loop closure detection and pose optimization are then used to eliminate accumulated errors, generating a high-precision real-time 3D point cloud map and a 2D grid map. Fixed coordinates and theme labels are configured for each booth in the exhibition area. Dynamic features generated during the movement of objects are removed from the 2D grid map and settled into static occupied grids. At the same time, the parameters of the global buffer map (GBM) and the local cost map (LCM) are initialized.

[0044] It should be noted that step S1 of the present invention also includes distributing the robot's speed limit, acceleration limit, angular velocity limit, curvature limit, safety radius, sensing parameters and attenuation parameters through the MCP parameter table to achieve consistent execution.

[0045] It should be noted that in step S1 of the present invention, the global buffer map parameters and local cost map parameters include update cycle, fusion radius, timestamp gating, etc.

[0046] S2. When the robot receives a natural language service request from the client, it converts the request into text and parses it into a structured MCP request using a large language model. The MCP-Bus (Model Context Protocol Bus) assigns a corresponding ID to the MCP request and performs routing, retries, and deduplication. When the MCP request is a query command, it outputs language signals to interact with the user, thereby completing the explanation and / or retrieval service. When the MCP request is a control command, it initializes the 3D point cloud map information, using the robot's current location as the starting point and the target display case location specified in the control command as the ending point, to prepare path planning.

[0047] It should be noted that in step S2 of this invention, interactive services are implemented through the MCP protocol. Automatic Speech Recognition (ASR) and a Large Language Model (LLM) are used to convert natural language service requests into one or more return instructions, thus obtaining a structured MCP (Model Context Protocol) request. This allows for the execution of corresponding operations based on different MCP request instructions. Return instructions include query instructions and control instructions. When a query instruction is received, the robot can perform corresponding services such as explanation and retrieval (i.e., outputting explanation content in voice form); when a control instruction is received, the robot enters a global path planning preparation state, starting from its current location and ending at the target display case location. The fields of the MCP request include {intent, target, deadline, constraints, context}, and are broadcast to subsequent global planning and motion control via a unified bus. Wherein, intent represents user intent; target represents target display location; deadline represents instruction timeliness constraint; constraints represent execution priority order constraint, in this embodiment specifying that control instructions have higher priority than query instructions; and context represents connecting context information, used to avoid repeated interactions.

[0048] S3. An improved A* algorithm based on orientation perception is used to search for the optimal global path on a two-dimensional grid map. The heuristic function of the improved A* algorithm is obtained by adding a heading cost term to the heuristic function of the A* algorithm.

[0049] It should be noted that in step S3 of this invention, the heuristic function of the improved A* algorithm... Calculate using the following formula:

[0050]

[0051] in, Indicates the robot's current location To the target display case location The Euclidean distance; The angle between the robot's current orientation and the direction of the target display case; and All are weighting coefficients greater than 0; This is the cost item for the course.

[0052] In this embodiment S3, the improved A* algorithm uses a heuristic function that adds a heading cost term to the traditional Euclidean distance and imposes steerability constraints on the turning radius / curvature during the expansion phase, prioritizing corridors with "consistent orientation and steerable passage." The included angle... Determined by map scale and chassis kinematic calibration.

[0053] Furthermore, in the improved A* algorithm described above, the anchor point sequence representing intermediate waypoints undergoes spline / curvature smoothing to generate a trackable global reference trajectory and desired heading, consistent with the constraints of the MCP parameter table (maximum velocity, maximum angular velocity, maximum curvature). Finally, it is issued as an MCP control command: {target, waypoints, v_limit, w_limit, k_limit}. Here, waypoints is the anchor point sequence; v_limit, w_limit, and k_limit represent the maximum velocity, maximum angular velocity, and maximum curvature, respectively.

[0054] S4. A dynamic cost map is constructed in the robot coordinate system using point clouds collected by LiDAR. A two-layer fusion mechanism combining a global buffer map and a local cost map is used, and a dual-rate attenuation strategy is introduced to manage dynamic obstacle information. On the pure tracking controller (path-follow) baseline, combined with the parameter constraints issued by the MCP parameter table, path following and dynamic obstacle avoidance are realized. During dynamic obstacle avoidance, motion execution and three-level replanning are adopted: when the dynamic cost map indicates that forward movement is blocked or the safety margin is less than the preset safety distance threshold, it is triggered step by step in the order of first-level trajectory sampling temporary obstacle avoidance, second-level local A* replanning, and third-level global A* replanning. When the previous level intervention is successful or the blocking condition is removed, it retreats to the next level or returns to the tracking controller baseline.

[0055] It should be noted that in step S4 of the present invention, the multi-frame point cloud is aggregated to obtain a global buffer map through spatiotemporal stabilization; the near-field obstacles are updated in real time using the laser point cloud in the robot coordinate system to generate a local cost map; different exponential decay rates are applied to the global buffer map and the local cost map respectively, and the exponential decay rate of the global buffer map is less than that of the local cost map. The timestamps of the global buffer map and the local cost map are unified as the fusion gate to form a dual-rate decay strategy.

[0056] In this embodiment, the Global Buffer Map (GBM) aggregates multiple frames of point clouds at a slower frequency (e.g., once every 5-20 seconds) and performs spatiotemporal stabilization to suppress sensor noise and short-term false detections. The Local Cost Map (LCM) updates nearby obstacle information in real time at a higher frequency (e.g., once every 0-5 seconds). Different exponential decay rates are applied to the GBM and LCM: obstacles that are only briefly observed by the LCM but not appearing in the GBM are subject to a faster exponential decay rate, while obstacles that appear in the GBM are subject to a slower exponential decay rate. That is, the exponential decay rate of the Global Buffer Map is always satisfied. The exponential decay rate is less than that of the local cost map. , Furthermore, this embodiment uses timestamp consistency for fusion gating, forming a "dual-rate decay" strategy to weaken dynamic obstacle "ghosting" and perception jitter, outputting a unified fusion cost map for tracking and replanning. Specifically, this embodiment embeds a safety buffer zone and a weight enhancement strategy for crowd-sensitive areas into the fusion cost map, with parameters uniformly applied by the MCP. The weight enhancement strategy for crowd-sensitive areas means that in the global buffer map, static objects have higher weights, and these static objects are avoided as much as possible during navigation; in the local cost map, not only static objects have higher weights, but dynamic objects also have particularly high weights, thus alerting the robot to these dynamic objects.

[0057] It should be noted that in step S4 of the present invention, the pure tracking controller uses a joint strategy of forward view point and curvature: the forward view distance is positively correlated with the linear velocity and is constrained by the upper limit of curvature; the rate of change of rotation angle is uniformly constrained by the MCP parameter table; wherein, the forward view distance is the safe distance between the robot and the waypoint.

[0058] It should be noted that in the dynamic obstacle avoidance step S4 of the present invention, the cost function corresponding to each candidate anchor point sequence is calculated according to the following formula. And select the candidate anchor sequence with the smaller cost function value for output:

[0059]

[0060] in, Weighting coefficients representing the heading; This indicates the angle of deviation between the current heading and the target heading; The weighting coefficient represents the minimum safe distance. This represents the minimum safe distance, which is the distance between the robot and its nearest obstacle. This indicates a correction term to prevent division by zero; Weighting coefficients representing speed; This represents the velocity stability constraint term; Weighting coefficients representing curvature; This represents the path smoothness constraint.

[0061] It should be noted that, in step S4 of this invention, the triggering condition for the three-level replanning is: when the minimum safe distance... Less than the preset safe distance threshold When this happens, the local cost map is resampled to generate a new path, thereby temporarily bypassing the obstacle (i.e., first-level trajectory sampling for temporary obstacle bypass); when continuous If all cycles fail to bypass obstacles temporarily, or if the path feasibility score falls below the preset feasibility threshold, the improved A* algorithm is re-executed on the local cost map for local path planning (i.e., second-level local A* replanning). If local replanning fails in all cycles, or if the planned path cannot reach the destination, the process reverts to the starting point and re-executes the improved A* algorithm for global path planning (i.e., three-level global A* replanning). All are preset number of cycles.

[0062] In this embodiment, when the obstacle is removed or the score is restored, the system unlocks and reverts in the order of "Level 3 → Level 2 → Level 1 → Pure Tracking" to avoid jitter and overreaction.

[0063] S5. Once the robot reaches the destination, it triggers an interaction service and returns to S2 after the interaction service ends, waiting for the next natural language service request.

[0064] It should be noted that in step S5 of the present invention, in the exhibition hall scene, a three-element binding of "location-theme-explanation script" is established for each display case, including: location identifier, exhibit theme identifier and explanation script identifier, and is registered in the MCP service directory; after the explanation ends or the visitor leaves, indicators such as visitor count and dwell time are transmitted back for operation statistics, and reset to S2 to wait for the next event.

[0065] In a dynamic visitor flow exhibition hall environment, a well-constructed 3D point cloud map, such as Figure 3 As shown, the constructed two-dimensional raster map is as follows: Figure 4 As shown. Figure 5In the diagram, the green track represents the planned global path. Figure 6 In the diagram, the purple trajectory represents a local obstacle avoidance path. It is evident that the combination of MCP integrated collaboration (unified parameters) and "dual-layer cost + dual-rate attenuation + three-level replanning" in this embodiment significantly reduces the number of replanning attempts per unit time, lateral error variance, and stagnation duration. Compared to the traditional "A*+DWA," the native tracing ability of the path is improved, and unnecessary detours caused by control jitter and "ghosting" are significantly reduced.

[0066] It should also be noted that the robot-guided tour method based on the MCP architecture in the above embodiments can essentially be executed by a computer program or module. Therefore, similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a robot-guided tour system based on the MCP architecture, corresponding to the robot-guided tour method based on the MCP architecture provided in the above embodiments, such as... Figure 7 As shown, it includes:

[0067] The data acquisition module is used to acquire natural language service requests sent by the client;

[0068] The result acquisition module is used to process the received natural language service request according to the robot navigation method based on the MCP architecture described in the above embodiments. When it is a query command, it outputs language signals to interact with the user; when it is a control command, it performs path planning and completes the user's service request after the robot reaches the target display case location.

[0069] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the robot guidance method based on the MCP architecture provided in the above embodiments, which includes a memory and a processor;

[0070] The memory is used to store computer programs;

[0071] The processor is configured to implement the robot navigation method based on the MCP architecture in the above embodiments when executing the computer program.

[0072] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0073] It is understood that the processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be 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, or discrete hardware components.

[0074] It should also be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the embodiments provided in this application, the division of steps or modules in the system and method is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules or steps may be combined or integrated together, and a module or step may also be split.

[0075] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A robot-guided tour method based on MCP architecture, characterized in that, Includes the following steps: S1. The robot pose and keyframes are output using a laser-IMU tightly coupled odometry method. Loop closure detection and pose optimization are then used to eliminate accumulated errors, generating a high-precision real-time 3D point cloud map and a 2D grid map. Fixed coordinates and theme labels are configured for each booth in the exhibition area. Dynamic features generated during object movement are removed from the 2D grid map and settled into static occupied grids. At the same time, global buffer map parameters and local cost map parameters are initialized. S2. When the robot receives a natural language service request from the client, it converts the request into text and parses it into a structured MCP request using a large language model. The MCP-Bus assigns a corresponding ID to the MCP request and performs routing, retries, and deduplication. When the MCP request is a query command, it outputs language signals to interact with the user, thereby completing the explanation and / or retrieval service. When the MCP request is a control command, it initializes the 3D point cloud map information, using the robot's current location as the starting point and the target display case location specified in the control command as the ending point, to prepare path planning. S3. An improved A* algorithm based on orientation awareness is used to search for the optimal global path on a two-dimensional grid map; the heuristic function of the improved A* algorithm is obtained by adding a heading cost term to the heuristic function of the A* algorithm. S4. A dynamic cost map is constructed in the robot coordinate system using the point cloud collected by the lidar. A two-layer fusion mechanism combining the global buffer map and the local cost map is used, and a dual-rate attenuation strategy is introduced to manage dynamic obstacle information. On the baseline of the pure tracking controller, combined with the parameter constraints issued by the MCP parameter table, path following and dynamic obstacle avoidance are realized. During dynamic obstacle avoidance, motion execution and three-level replanning are adopted: when the dynamic cost map indicates that forward movement is blocked or the safety margin is less than the preset safety distance threshold, the first-level trajectory sampling temporary obstacle avoidance, the second-level local A* replanning, and the third-level global A* replanning are triggered in sequence. When the previous level intervention is successful or the blocking condition is removed, the system will retreat to the next level or return to the tracking controller baseline. S5. Once the robot reaches the destination, it triggers an interaction service and returns to S2 after the interaction service ends, waiting for the next natural language service request.

2. The robot-guided tour method based on MCP architecture as described in claim 1, characterized in that, Step S1 also includes distributing the robot's speed limit, acceleration limit, angular velocity limit, curvature limit, safety radius, sensing parameters, and attenuation parameters through the MCP parameter table.

3. The robot-guided tour method based on MCP architecture as described in claim 1, characterized in that, In step S2, the fields of the MCP request include {intent, target, deadline, constraints, context}; where intent represents the user's intent; target represents the target display location; deadline represents the timeliness constraint of the instruction; constraints represent the execution priority order constraint; and context represents the connecting context information.

4. The robot-guided tour method based on MCP architecture as described in claim 1, characterized in that, In step S3, the heuristic function of the improved A* algorithm is... Calculate using the following formula: ； in, Indicates the robot's current location To the target display case location The Euclidean distance; The angle between the robot's current orientation and the direction of the target display case; and All are weight coefficients greater than 0; This is the cost item for the course.

5. The robot-guided tour method based on MCP architecture as described in claim 1, characterized in that, In step S4, the multi-frame point cloud is aggregated through spatiotemporal stabilization to obtain a global buffer map; the near-field obstacles are updated in real time using the laser point cloud in the robot coordinate system to generate a local cost map; different exponential decay rates are applied to the global buffer map and the local cost map respectively, and the exponential decay rate of the global buffer map is less than that of the local cost map. The timestamps of the global buffer map and the local cost map are unified as the fusion gating to form a dual-rate decay strategy.

6. The robot-guided tour method based on MCP architecture as described in claim 1, characterized in that, In step S4, the pure tracking controller uses a combined strategy of forward view point and curvature: the forward view distance is positively correlated with the linear velocity and is constrained by the upper limit of curvature; the rate of change of rotation angle is uniformly constrained by the MCP parameter table; where the forward view distance is the safe distance between the robot and the waypoint.

7. The robot-guided tour method based on MCP architecture as described in claim 1, characterized in that, In the dynamic obstacle avoidance step S4, the cost function corresponding to each candidate anchor point sequence is calculated using the following formula. And select the candidate anchor sequence with the smaller cost function value for output: ； in, Weighting coefficients representing the heading; This indicates the angle of deviation between the current heading and the target heading; The weighting coefficient represents the minimum safe distance. This represents the minimum safe distance, which is the distance between the robot and its nearest obstacle. This indicates a correction term to prevent division by zero; Weighting coefficients representing speed; This represents the velocity stability constraint term; Weighting coefficients representing curvature; This represents the path smoothness constraint.

8. The robot-guided tour method based on MCP architecture as described in claim 1, characterized in that, In step S4, the trigger condition for the third-level replanning is: when the minimum safe distance is less than the preset safe distance threshold, the local cost map is resampled to generate a new path, thereby achieving the purpose of temporarily bypassing the obstacle; When continuous If all cycles fail to bypass obstacles temporarily, or if the path feasibility score falls below a preset feasibility threshold, the improved A* algorithm is re-executed on the local cost map for local path planning; when consecutive If local replanning fails in all cycles, or if the planned path cannot reach the destination, then backtrack to the starting point and re-execute the improved A* algorithm for global path planning. , All are preset number of cycles.

9. A robot-guided tour system based on MCP architecture, characterized in that, include: The data acquisition module is used to acquire natural language service requests sent by the client; The result acquisition module is used to process the received natural language service request according to the robot navigation method based on the MCP architecture as described in any one of claims 1 to 8. When it is a query command, it outputs language signals to interact with the user; when it is a control command, it performs path planning and completes the user's service request after the robot reaches the target display case location.

10. A computer electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement the robot navigation method based on the MCP architecture as described in any one of claims 1 to 8 when executing the computer program.

Images