CN-121982231-A - Robot navigation method and system based on MCP architecture

CN121982231ACN 121982231 ACN121982231 ACN 121982231ACN-121982231-A

Abstract

The invention discloses a robot navigation method and system based on an MCP architecture. The method comprises the steps of firstly receiving a user instruction through voice recognition, analyzing the instruction into an MCP request instruction by adopting a large language model, secondly generating an optimized path on a static map based on an improved A-type algorithm, constructing a dynamic cost map through real-time point cloud data, innovatively adopting a double-layer fusion mechanism of a global buffer map and a local cost map, introducing double-rate attenuation to realize intelligent management of barrier information, effectively solving the problems of jitter and ghost of dynamic barrier perception, and finally executing a navigation instruction based on an MCP protocol and realizing dynamic barrier avoidance and accurate path tracking by combining a three-level re-planning mechanism. The invention deeply integrates the MCP protocol, realizes full-flow automation from voice analysis, path planning to motion execution through a standardized instruction system, and remarkably improves the interaction intelligence and navigation reliability of the navigation robot in complex environments such as exhibition halls and the like.

Inventors

CHEN LI
HU WENQIN
WU LINTAO
TONG RUOFENG
TANG MIN
DU PENG

Assignees

浙江大学

Dates

Publication Date: 20260505
Application Date: 20251230

Claims (10)

1. The robot navigation method based on the MCP architecture is characterized by comprising the following steps of: S1, outputting a pose and a key frame of a robot by adopting an odometer method based on tight coupling of a laser and an IMU, eliminating accumulated errors through loop detection and pose optimization, generating a high-precision real-time 3D point cloud map and a two-dimensional grid map, and configuring fixed point coordinates and subject identifiers for each showcase in an exhibition area; S2, after the robot receives a natural language service request sent by a client, converting the request into a text form, analyzing the text form into a structured MCP request through a large language model, distributing a corresponding ID for the MCP request by an MCP-Bus, and completing routing, retry and deduplication; when the MCP request is a control instruction, initializing 3D point cloud map information, taking the current position of the robot as a starting point, taking the position of a target showcase appointed in the control instruction as an end point, and preparing path planning; S3, adopting an improved A-algorithm of orientation perception to search an optimal global path on a two-dimensional grid map, wherein a heuristic function of the improved A-algorithm is obtained by adding a heading cost term in a heuristic function of the A-algorithm; S4, constructing a dynamic cost map under a robot coordinate system by using point clouds acquired by a laser radar, using a double-layer fusion mechanism of a global buffer map and a local cost map, and introducing a double-rate attenuation strategy to manage dynamic obstacle information; in the dynamic obstacle avoidance process, adopting motion execution and three-level re-planning, namely when the forward blockage or the safety margin is smaller than a preset safety distance threshold value prompted by a dynamic cost map, sampling a first-level track, temporarily detouring, carrying out second-level local A re-planning and carrying out third-level global A re-planning in sequence to trigger step by step, and returning to the next level or returning to the tracking controller base line when the previous level of intervention is successful or the blocking condition is released; s5, triggering the interactive service when the robot reaches the end point, returning to S2 after the interactive service is finished, and waiting for the next natural language service request.
2. The robot guiding method based on the MCP architecture of claim 1, wherein step S1 further includes issuing the upper speed limit, the upper acceleration limit, the upper angular speed limit, the upper curvature limit, the safety radius, the perception parameter, and the attenuation parameter of the robot in a lump through the MCP parameter table.
3. The robot navigation method based on the MCP architecture of claim 1, wherein in step S2, the field of the MCP request contains { intent, target, deadline, constraints, context }, wherein the intent represents the user intent, the target represents the target showcase location, deadline represents the instruction timeliness constraint, constraints represents the execution priority order constraint, and context represents the join context information.
4. The robot guiding method based on MCP architecture according to claim 1, wherein in step S3, a heuristic function of a-algorithm is improved Calculated as follows: ; Wherein, the Indicating the current position of the robot To the target showcase position Is the euclidean distance of (2); the included angle between the current direction of the robot and the direction of the target showcase is set; And All are weight coefficients greater than 0; is a heading cost term.
5. The robot navigation method based on the MCP architecture as claimed in claim 1, wherein in the step S4, a global buffer map is obtained by aggregation of multi-frame point clouds through space-time stabilization, near-field obstacles are updated in real time by laser point clouds under a robot coordinate system to generate a local cost map, different exponential decay rates are respectively applied to the global buffer map and the local cost map, the exponential decay rate of the global buffer map is smaller than that of the local cost map, and the time stamps of the global buffer map and the local cost map are unified as fusion gating to form a double-speed-rate decay strategy.
6. The robot navigation method based on the MCP architecture as claimed in claim 1, wherein in the step S4, the pure tracking controller uses a forward view point and curvature combination strategy that a forward view distance is positively correlated with a linear velocity and is constrained by an upper curvature limit, and a rotation angle change rate is uniformly constrained by an MCP parameter table, wherein the forward view distance is a safe distance between the robot and a passing point.
7. The robot navigation method based on MCP architecture of claim 1, wherein in the dynamic obstacle avoidance of step S4, the cost function corresponding to each candidate anchor sequence is calculated according to the following formula And selecting a candidate anchor point sequence with small cost function value for output: ; Wherein, the A weight coefficient representing heading; Representing a deviation angle of the current heading and the target heading; the weight coefficient representing the minimum safe distance, Representing a minimum safe distance, i.e. the distance of the robot from its nearest obstacle; a correction term indicating that 0 is prevented from being divided; A weight coefficient representing the speed; representing a speed stationarity constraint term; a weight coefficient representing curvature; representing a path smoothness constraint.
8. The robot navigation method based on MCP architecture of claim 1, wherein in step S4, the triggering condition of three-level re-planning is that when the minimum safe distance is smaller than the preset safe distance threshold, local cost map resampling is triggered to generate a new path so as to achieve the purpose of temporarily bypassing the obstacle, and when the minimum safe distance is continuous Triggering to execute the improved A algorithm again on the local cost map to carry out local path planning when the temporary obstacle detouring fails in all periods or the path feasibility score is lower than a preset feasibility threshold value, and when the path feasibility score is continuous If the period is failed in local re-planning or the planned path cannot reach the end point, returning to the start point, and re-executing the improved A algorithm to carry out global path planning; 、 The number of cycles is preset.
9. A robotic navigation system based on MCP architecture, comprising: the data acquisition module is used for acquiring a natural language service request sent by the client; The result acquisition module is used for processing the received natural language service request according to the robot navigation method based on the MCP architecture according to any one of claims 1-8, outputting a language signal to interact with a user when the query instruction is given, executing path planning when the query instruction is given, and completing the service request of the user after the robot reaches the target showcase position.
10. A computer electronic device comprising a memory and a processor; The memory is used for storing a computer program; The processor is configured to implement the robot navigation method based on the MCP architecture according to any one of claims 1 to 8 when executing the computer program.

Description

Robot navigation method and system based on MCP architecture Technical Field The invention belongs to the technical field of mobile robot navigation and man-machine interaction, and particularly relates to a robot navigation method and system based on an MCP (Model Context Protocol ) architecture. Background The existing navigation robot mostly adopts a layered paradigm of global static planning and local dynamic obstacle avoidance, wherein an upper layer solves a global path on a static map, and a lower layer carries out real-time obstacle avoidance according to a sensor and outputs motion control. The common industrial assembly line is 'A-global + DWA local', namely, a two-dimensional grid is searched for approximate shortest path by A-global search, and then a dynamic window method (DWA) is used for sampling and short-term prediction in a speed space, and the target guiding, obstacle avoidance and speed benefit scoring are combined to select a current speed instruction. The method has the advantages of simple realization and strong real-time performance, but the method still has the defects in engineering problems such as dynamic passenger flow jitter, multi-round explanation business arrangement, cross-module constraint consistency and the like, and uniform attenuation of a single-layer cost map is easy to form 'ghosting' or 'excessive clearing', so that frequent heavy specification and tracking oscillation are caused. The prior proposal improves the overall performance by generating a static approximate optimal path through RRT, secondarily optimizing an ant colony algorithm (smoothing, reducing the length and inflection points), combining with DWA dynamic obstacle avoidance, gives out parameter setting, flow and simulation results, and adapts to museum guiding application, has the value of improving the quality and dynamic safety of the static path, but the serial connection of focusing algorithm layers is still insufficient in the system problems of protocol standardization, cross-layer rate/constraint consistency, dynamic obstacle 'ghost' jitter management, re-planning stable switching and the like. Classical a-algorithm may search for approximate shortest paths on a static two-dimensional grid with distance heuristics (euler/manhattan). However, when the method is matched with the DWA, the method is often dependent on rolling update and unified attenuation of a single-layer cost map, the steering/curvature constraint of the chassis is not reflected, a broken line path which is geometrically reachable but difficult to smoothly track in kinematics is easy to produce, the correction pressure of a local layer is large, and the vibration and the heavy gauge are easy to trigger at a narrow channel/sharp folding position. Disclosure of Invention Aiming at the defects of the existing 'A-global + DWA local' assembly line in the aspects of cross-layer coordination, dynamic jitter suppression, reprofiling stability, service closed loop and the like, the invention aims to solve the following problems. The method comprises the steps of 1) carrying out full-flow standardization which is lack of protocol in the prior art, namely, lacking a unified semantic/route/flow control and idempotent mechanism between voice semantics and navigation control, greatly invading navigation links by newly added plug-in units and service functions (explanation, statistics and safety monitoring), 2) carrying out global path traceability which is insufficient in the prior art, namely, classical A is not explicitly incorporated into chassis orientation/turning radius constraint, so that high-frequency actions and curvature peaks of rear-end control are large, 3) carrying out dynamic barrier 'ghosting' and perception jitter, namely, carrying out single-layer costmap unified attenuation, wherein short-time false detection and stable barrier are difficult to distinguish, frequent switching and unnecessary re-planning are easy to cause, and 4) carrying out re-planning triggering on scattered, namely, lacking grading rollback and unlocking criteria, and carrying out state concussion and stagnation duration are difficult to control. Therefore, the invention provides a robot navigation method and system based on an MCP architecture. In order to achieve the above purpose, the present invention adopts the following technical scheme: In a first aspect, the present invention provides a robot navigation method based on MCP architecture, including the steps of: S1, outputting a pose and a key frame of a robot by adopting an odometer method based on tight coupling of a laser and an IMU, eliminating accumulated errors through loop detection and pose optimization, generating a high-precision real-time 3D point cloud map and a two-dimensional grid map, and configuring fixed point coordinates and subject identifiers for each showcase in an exhibition area; S2, after the robot receives a natural language service request sent by a client, converting the