Multi-scene adaptive AGV scheduling management system and path planning method

Abstract

The invention discloses a multi-scene adaptive AGV scheduling management system and a path planning method. Comprising a multi-scene perception and environment modeling module, a multi-AGV task scheduling and resource allocation module, a dynamic path planning and optimization module, a scene adaptive decision module, a multi-sensor fusion obstacle avoidance module, a heterogeneous communication and data synchronization module and a whole-process monitoring and fault diagnosis module. According to the invention, full-process adaptive control from environment modeling to decision optimization is realized. An SVM classification algorithm and dynamic matching degree calculation are adopted, so that the system can accurately identify characteristics of heterogeneous scenes such as warehousing, workshops and logistics hubs, and an optimal scheduling strategy is automatically switched; and a multi-dimensional cost function including path length, obstacle risk, smoothness and congestion degree is constructed, and seamless connection of global path pre-planning and local dynamic adjustment is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of AGV scheduling, in particular to a multi-scene adaptive AGV scheduling management system and path planning method. BACKGROUND

[0002] With the rapid development of intelligent manufacturing and smart logistics industry, AGV, as the core equipment of material handling and task execution, has been widely used in warehouse, workshop, logistics hub and other scenes. However, the existing AGV scheduling management system and path planning method has the following technical problems in practical application, which is difficult to meet the demand of multi-scene adaptive operation: The existing system is designed for a single scene, and uses fixed scheduling rules and path planning algorithms. When the scene is switched (such as switching from a wide aisle in the warehouse to a narrow aisle in the workshop, or switching from a static environment to a dynamic pedestrian environment), the system cannot automatically adjust the strategy, resulting in decreased scheduling efficiency and unreasonable path planning.

[0003] When multiple AGVs are working, there is a lack of efficient task allocation and conflict coordination mechanism, which is prone to AGV congestion, path occupation and task delay, especially in the scene of multiple task parallel and limited resources, the collaborative efficiency is low.

[0004] Traditional path planning is mostly based on static map to generate fixed path, without fully considering the dynamic changes of environment (such as temporary obstacles, AGV failure, task priority adjustment), resulting in low path feasibility and frequent manual intervention.

[0005] The data of laser radar, visual sensor and infrared sensor equipped by AGV are mostly used independently, without establishing an effective fusion mechanism, resulting in low obstacle detection accuracy, one-sided environmental perception and insufficient obstacle avoidance reliability.

[0006] There is a lack of whole-process state monitoring and fault diagnosis mechanism, which cannot master the AGV running state and task execution progress in real time, and it is difficult to quickly locate the cause and take emergency measures when faults occur, resulting in long system downtime.

[0007] There is no dynamic updated decision rule library, which cannot adaptively optimize the scheduling and path strategy according to the changes of scene characteristics (such as warehouse shelf layout adjustment and workshop production process change), and the system has poor expansibility. SUMMARY

[0008] The present application relates to the technical field of AGV scheduling, in particular to a multi-scene adaptive AGV scheduling management system and path planning method.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a multi-scenario adaptive AGV scheduling and management system, including a multi-scenario perception and environment modeling module, a multi-AGV task scheduling and resource allocation module, a dynamic path planning and optimization module, a scenario adaptive decision-making module, a multi-sensor fusion obstacle avoidance module, a heterogeneous communication and data synchronization module, and a full-process monitoring and fault diagnosis module. The multi-scene perception and environment modeling module is used to collect multi-scene environment data, extract scene features, dynamically update environmental information, and generate a unified map. The multi-AGV task scheduling and resource allocation module is used to parse task information, evaluate AGV status, sort task priorities, allocate task resources, and coordinate multi-AGV conflicts. The dynamic path planning and optimization module is used to generate global paths, dynamically adjust local paths, evaluate path costs, and smoothly optimize paths. The scenario adaptive decision-making module is used to identify scenario types, match optimal scheduling and path strategies, and dynamically update decision rules. The multi-sensor fusion obstacle avoidance module is used to detect obstacles, calibrate sensor data, fuse sensor information and generate an obstacle avoidance path. The heterogeneous communication and data synchronization module is used to realize multi-protocol communication, optimize data transmission, synchronize the time of each node, and verify data consistency. The full-process monitoring and fault diagnosis module is used to monitor the status of the system and AGV, detect and locate faults, perform emergency handling, and record and analyze data.

[0010] Preferably, the multi-scene perception and environment modeling module includes a multimodal perception unit, a scene feature extraction unit, a dynamic environment update unit, and a unified map generation unit; The multimodal sensing unit integrates lidar, vision camera, infrared sensor, RFID reader, and ultrasonic sensor to collect environmental 3D point cloud, images, tags, and distance data; The scene feature extraction unit extracts scene type features, environmental structure features, dynamic element features, and task association features; The dynamic environment update unit updates the feature data of the static environment, dynamic environment, and abnormal environment; the unified map generation unit generates a global map in a raster-point cloud hybrid format that includes a static layer, a dynamic layer, and a task layer.

[0011] Preferably, the multi-AGV task scheduling and resource allocation module includes a task parsing unit, an AGV status evaluation unit, a task priority sorting unit, a resource allocation optimization unit, and a conflict coordination unit. The task parsing unit breaks down the task and extracts basic task information and related information; The AGV status assessment unit assesses the basic status, performance status, and task status of the AGV. The task priority sorting unit calculates task priorities using a task priority calculation model, which is as follows: in As a task priority, , , , The weighting coefficients are satisfied. , This represents the urgency level of the task. This represents the task complexity coefficient. This is the task value coefficient. This refers to the AGV adaptability coefficient. The resource allocation optimization unit uses an improved genetic algorithm to allocate task resources; The conflict coordination unit detects and coordinates path conflicts, resource conflicts, and task conflicts.

[0012] Preferably, the dynamic path planning and optimization module includes a global path pre-planning unit, a local path dynamic adjustment unit, a path cost evaluation unit, and a path smoothing optimization unit; The global path pre-planning unit uses an improved A algorithm to generate a global initial path; The local path dynamic adjustment unit uses the RRT algorithm to adjust the local path; The path cost evaluation unit evaluates the quality of a path using a dynamic path cost function, which is: in The total cost of the path. , , , The weighting coefficients are and satisfy the following conditions: , For path length cost, As a consequence of obstacle risk, For path smoothness cost, The cost is path congestion. The path smoothing optimization unit uses a third-order Bézier curve to smooth the path.

[0013] Preferably, the scene adaptive decision-making module includes a scene recognition unit, a decision rule base unit, an adaptive matching unit, and a strategy update unit; The scene recognition unit uses the SVM classification algorithm to identify scene types; The decision rule base unit stores a subset of scheduling rules, a subset of path planning strategies, and a subset of parameter configurations; The adaptive matching unit calculates the optimal matching rules and strategies based on the scene adaptive matching degree. The formula for calculating the scene adaptive matching degree is as follows: in For matching degree, The number of scene feature dimensions, For the first The weight coefficients of each feature and satisfying , For the current scene The actual values ​​of each feature. For the rule corresponding to the first Each characteristic standard value; The policy update unit uses a reinforcement learning algorithm to update the decision rules and parameters.

[0014] Preferably, the multi-sensor fusion obstacle avoidance module includes an obstacle detection unit, a sensor data calibration unit, a fusion decision unit, and an obstacle avoidance path generation unit; The obstacle detection unit detects static obstacles, dynamic obstacles, and temporary obstacles; The sensor data calibration unit performs time calibration, spatial calibration, and accuracy calibration. The fusion decision unit fuses sensor data using a multi-sensor fusion confidence model, which is as follows: in To integrate confidence levels, For the number of sensors, For the first The reliability weight of each sensor, For the first The detection confidence level of each sensor; The obstacle avoidance path generation unit uses an improved artificial potential field method to generate obstacle avoidance paths.

[0015] Preferably, the heterogeneous communication and data synchronization module includes a multi-protocol communication unit, a data transmission optimization unit, a time synchronization unit, and a data consistency verification unit; The multi-protocol communication unit supports WiFi 6, Bluetooth 5.0, LoRa, Ethernet, and CAN bus protocols, and can adaptively switch communication protocols. The data transmission optimization unit employs classified transmission, error control, and adaptive bandwidth adjustment to optimize data transmission. The time synchronization unit achieves time synchronization based on the NTP protocol; The data consistency verification unit uses the Raft algorithm to ensure data consistency.

[0016] Preferably, the full-process monitoring and fault diagnosis module includes a status monitoring unit, a fault detection unit, a fault location unit, an emergency handling unit, and a data recording and analysis unit; The status monitoring unit monitors the AGV's operating status, task execution status, and system module status. The fault detection unit uses threshold detection, trend analysis, model prediction, and redundancy detection methods to detect faults. The fault location unit locates faults through fault code analysis, circuit testing, component replacement testing, and log analysis. The emergency response unit takes emergency measures in response to hardware failures, software failures, system failures, and security risks. The data recording and analysis unit stores operational data and fault data, and uses big data analysis algorithms to analyze the data.

[0017] A multi-scenario adaptive AGV path planning method, based on the aforementioned multi-scenario adaptive AGV scheduling and management system, includes the following steps: Step 1: System initialization and scene perception, initialization of each module, multi-scene perception and environment modeling module collects environmental data and generates an initial global map; Step 2: Scene recognition and decision matching. The scene adaptive decision module identifies the scene type and matches the optimal scheduling and path strategy. Step 3: Task reception and parsing. The task parsing unit breaks down the task and extracts task information, and the task priority sorting unit sorts the task priorities. Step 4: Multi-AGV status assessment and task allocation. The AGV status assessment unit assesses the AGV status, the resource allocation optimization unit allocates task resources, and the conflict coordination unit coordinates conflicts. Step 5: Global path pre-planning and cost evaluation. The global path pre-planning unit generates the global initial path, the path cost evaluation unit evaluates the path cost, and the path smoothing optimization unit smooths the path. Step 6: Dynamic path adjustment and obstacle avoidance. The multi-sensor fusion obstacle avoidance module detects obstacles and generates an obstacle avoidance path, while the local path dynamic adjustment unit adjusts the local path. Step 7: AGV execution and status monitoring, AGV executes tasks, heterogeneous communication and data synchronization module synchronizes data, full-process monitoring and fault diagnosis module monitors status and handles faults; Step 8: Task completion and system optimization. The AGV provides feedback on task completion, the data recording and analysis unit records and analyzes the data, and the strategy update unit optimizes decision rules and parameters.

[0018] Preferably, in step 2, scene recognition uses the SVM classification algorithm, and decision matching is achieved through scene adaptive matching degree calculation; in step 3, task priority is calculated through a task priority calculation model; in step 5, path cost is evaluated through a dynamic path cost function; and in step 6, obstacle detection confidence is calculated through a multi-sensor fusion confidence model.

[0019] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves end-to-end adaptive control from environmental modeling to decision optimization. Employing an SVM classification algorithm and dynamic matching degree calculation, the system can accurately identify heterogeneous scene characteristics such as warehouses, workshops, and logistics hubs, and automatically switch to the optimal scheduling strategy. A multi-dimensional cost function, including path length, obstacle risk, smoothness, and congestion, is constructed to achieve seamless integration of global path pre-planning and local dynamic adjustment. By integrating heterogeneous data from LiDAR, visual cameras, and other sources through a multi-sensor fusion confidence model, obstacle detection accuracy is improved and obstacle avoidance response time is shortened. A dynamic update mechanism based on a reinforcement learning-based decision rule base enables the system to autonomously optimize scheduling parameters without manual intervention when scenarios such as changes in shelf layout and production process adjustments occur, resulting in improved task execution efficiency compared to traditional methods. Attached Figure Description

[0020] Figure 1 This is a system module diagram of the present invention; Figure 2 This is a schematic diagram of the multi-scene perception and environment modeling module of the present invention; Figure 3 This is a schematic diagram of the multi-AGV task scheduling and resource allocation module of the present invention; Figure 4 This is a schematic diagram of the dynamic path planning and optimization module of the present invention; Figure 5 This is a flowchart of the method of the present invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figures 1-5This invention provides a multi-scenario adaptive AGV scheduling and management system, including a multi-scenario perception and environment modeling module, a multi-AGV task scheduling and resource allocation module, a dynamic path planning and optimization module, a scenario adaptive decision-making module, a multi-sensor fusion obstacle avoidance module, a heterogeneous communication and data synchronization module, and a full-process monitoring and fault diagnosis module. The multi-scene perception and environment modeling module is used to collect multi-scene environment data, extract scene features, dynamically update environmental information, and generate a unified map. The multi-AGV task scheduling and resource allocation module is used to parse task information, evaluate AGV status, sort task priorities, allocate task resources, and coordinate multi-AGV conflicts. The dynamic path planning and optimization module is used to generate global paths, dynamically adjust local paths, evaluate path costs, and smoothly optimize paths. The scene adaptive decision-making module is used to identify scene types, match optimal scheduling and path strategies, and dynamically update decision rules. The multi-sensor fusion obstacle avoidance module is used to detect obstacles, calibrate sensor data, fuse sensor information and generate obstacle avoidance paths; The heterogeneous communication and data synchronization module is used to implement multi-protocol communication, optimize data transmission, synchronize the time of each node, and verify data consistency. The end-to-end monitoring and fault diagnosis module is used to monitor the status of the system and AGV, detect and locate faults, perform emergency handling, and record and analyze data.

[0023] The multi-scene perception and environment modeling module includes a multimodal perception unit, a scene feature extraction unit, a dynamic environment update unit, and a unified map generation unit; The multimodal sensing unit integrates various sensing devices such as LiDAR, vision camera, infrared sensor, RFID reader, and ultrasonic sensor to cover the sensing needs of different scenarios: LiDAR: Used to acquire 3D point cloud data of the environment to detect obstacle distances and contours; Visual cameras: collect environmental image information and identify scene feature markers (such as QR codes on warehouse shelves, workshop equipment markings, and logistics hub platform markings), pedestrians, and dynamic obstacles; Infrared sensors are used in harsh environments such as low light and dust, such as obstacle detection in welding areas of workshops and dark corners of warehouses, to compensate for the blind spots of lidar and visual sensors. RFID reader: Reads preset RFID tag information in the scene, such as shelf number, material location, and area permission, to achieve AGV positioning assistance and scene feature confirmation; Ultrasonic sensors: used for short-range obstacle detection, such as short-range collision avoidance between AGVs and between AGVs and equipment, supplementing the short-range sensing accuracy of lidar.

[0024] Data collected by each sensing device is transmitted to the processing unit in real time through a data interface to ensure the real-time nature of environmental information.

[0025] The scene feature extraction unit extracts features from the raw data collected by the multimodal perception unit, identifying kernel feature parameters for different scenes, specifically including: Scene type characteristics: Through image recognition and RFID tag parsing, scene type identifiers are extracted, such as "warehouse sorting scene", "workshop material transfer scene", and "logistics hub loading and unloading scene". Environmental structural features: Extracting channel width, obstacle distribution density, and fixed facility location parameters from lidar point cloud data; Dynamic element features: Extract the motion speed, direction, and size parameters of dynamic obstacles using data from visual cameras and infrared sensors; Task-related features: By combining RFID data with task instructions, features related to the current task are extracted, such as the location features of the target shelf in the warehouse and the accessibility features of the target workstation in the workshop.

[0026] The feature extraction process employs a deep learning algorithm, namely a CNN convolutional neural network, to ensure the accuracy and robustness of feature extraction.

[0027] The dynamic environment update unit monitors changes in environmental information in real time and dynamically updates the environmental model, specifically including: Static environment update: When fixed facilities in the scene change (such as warehouse rack relocation or workshop equipment adjustment), the changed area is identified and the fixed feature parameters in the environment model are updated through comparative analysis of multimodal perception data. Dynamic environment update: Real-time tracking of the movement status of dynamic obstacles, updating the position, speed and other parameters of the dynamic obstacles at preset intervals (adjusted adaptively by the system according to the complexity of the scene); Abnormal Environment Update: When a sudden abnormality is detected, an emergency update mechanism is triggered to prioritize updating the environmental data of the abnormal area and synchronize it to the dispatch center and other AGVs.

[0028] The unified map generation unit generates a global environment map in a unified format based on extracted scene features and dynamically updated environmental data, specifically including: Map layered design: The map is divided into a static layer, a dynamic layer, and a task layer. The static layer includes fixed facilities and passage boundaries, the dynamic layer includes dynamic obstacles and the real-time position of AGVs, and the task layer includes the position of the task target and the task execution area. Unified map format: A raster-point cloud hybrid map format is adopted. The static layer is stored as a raster map, and the grid size is adaptively adjusted according to the scene's accuracy requirements. The dynamic layer and task layer are supplemented with point cloud data to ensure the accuracy and lightweight nature of the map. Map adaptation output: Based on the needs of the scheduling center and the AGV local path planning module, output map data of different precision and dimensions to meet the calculation needs of different modules.

[0029] The multi-AGV task scheduling and resource allocation module includes a task parsing unit, an AGV status evaluation unit, a task priority sorting unit, a resource allocation optimization unit, and a conflict coordination unit. The task parsing unit receives task instructions from the upper-layer system, and breaks down and parses the task, specifically including: Task basic information parsing: Extract task ID, task type, task start point, task end point, required resources, and task requirements; Task association information parsing: Identifying the dependencies between tasks and the association between tasks and scenarios; Task decomposition: For complex tasks, they are decomposed into multiple subtasks, each of which contains an independent execution objective, path range, and time window.

[0030] The AGV status assessment unit collects real-time operational status data from each AGV, comprehensively evaluating the available resources and execution capabilities of the AGVs, specifically including: Basic status assessment: AGV position, speed, heading angle, remaining battery power, and load status; Performance status assessment: AGV's operational fault status, positioning accuracy, and motion performance, including maximum speed, acceleration, and turning radius; Task status assessment: the current task execution progress of the AGV, the task completion quality, and the available time for the next task to be scheduled.

[0031] The evaluation data is transmitted back in real time through a heterogeneous communication module to ensure the real-time performance and accuracy of AGV status evaluation.

[0032] The task priority sorting unit establishes a task priority calculation model based on task attributes and system operation requirements to sort all tasks to be executed, ensuring that high-priority tasks are executed first. The priority calculation model is as follows: in: This represents the task priority, with a value range of [0,1]. The larger the value, the higher the priority. , , , These are the weight coefficients of each influencing factor, satisfying... It can be adaptively adjusted according to the type of scene; This is the task urgency coefficient, calculated based on the task completion time limit, with a value range of [0,1]. The shorter the completion time limit, the higher the urgency coefficient. The larger; This is the task complexity coefficient, calculated based on the number of subtasks, path length, and execution difficulty. Its value ranges from [0,1]. Higher complexity results in higher performance. The larger; This is the task value coefficient, calculated based on the task's impact on the production / logistics process. Its value ranges from [0,1], with higher values ​​indicating greater impact. The larger; This is the AGV adaptability coefficient, calculated based on the degree of matching between AGV performance and task requirements. Its value ranges from [0,1]. A higher adaptability coefficient indicates better performance. The larger the value, the higher the priority of all tasks to be executed. These tasks are then sorted from highest to lowest priority to form a task execution queue.

[0033] The resource allocation optimization unit allocates tasks to the optimal AGV based on the task priority queue and AGV status evaluation results, thereby achieving rational utilization of resources. Specifically, this includes: Allocation objectives: The optimization objectives are to minimize the total task execution time, reduce AGV energy consumption, and maximize resource utilization. Assignment constraints: The AGV's load capacity meets the task requirements, the AGV's remaining power can support task execution, the AGV's current status is fault-free, and the AGV's permissions match the task execution area; Allocation Algorithm: An improved genetic algorithm is adopted to transform the task allocation problem into a multi-objective optimization problem. Through encoding, crossover, and mutation operations, the optimal allocation scheme is solved to ensure the global optimality of the allocation result.

[0034] The conflict coordination unit establishes a multi-dimensional conflict coordination mechanism to address potential path and resource conflicts that may occur during multi-AGV operations. Specifically, this includes: Conflict detection: Based on the real-time position of the AGV, the planned path, and the task execution time window, potential conflicts are identified through a spatiotemporal conflict detection algorithm; Conflict Classification: Conflicts are classified into path conflicts (AGVs arrive at the same location at the same time or their paths intersect), resource conflicts (multiple AGVs compete for the same task resource or channel resource), and task conflicts (multiple tasks compete for the same AGV). Conflict resolution strategies: Path conflicts: The "priority avoidance" strategy is adopted, where low-priority AGVs adjust their paths or pause and wait, while high-priority AGVs execute their original paths; if the priorities are the same, the "closest distance" strategy is adopted, where AGVs farther from the conflict point adjust their paths. Resource conflicts: Resources are allocated according to task priority, with AGVs corresponding to higher priority tasks having priority in occupying resources; if priorities are the same, resources are allocated according to the order of resource request time. Task conflict: Reassess the AGV status and task requirements, assign conflicting tasks to other available AGVs, or adjust the task execution order.

[0035] The dynamic path planning and optimization module includes a global path pre-planning unit, a local path dynamic adjustment unit, a path cost evaluation unit, and a path smoothing optimization unit; The global path pre-planning unit generates a globally optimal path from the starting point to the destination for the AGV based on a unified map generated by the multi-scene perception and environment modeling module. Specifically, this includes: Planning objectives: Shortest path length, least travel time, lowest energy consumption, and highest obstacle avoidance rate; Planning algorithm: An improved A* algorithm is adopted, and a scene-adaptive heuristic function is introduced. The weight of the heuristic function is adjusted according to the scene type to ensure the scene adaptability of the path planning. Path constraints: satisfy AGV kinematic constraints, scene physical constraints, and task time constraints.

[0036] The local path dynamic adjustment unit monitors real-time environmental changes and makes local adjustments to the global path to ensure path feasibility. Specifically, this includes: Adjust trigger conditions: new obstacle detected, AGV deviates from the planned path by more than a threshold, task target location changes, higher priority task is inserted; Adjustment scope: Only the affected local path segments are adjusted, without the need to replan the global path, reducing the amount of computation and improving response speed; Algorithm adjustment: The Fast Exploratory Random Tree (RRT*) algorithm is adopted, which combines dynamic environmental information to quickly generate local optimal paths, ensuring that the adjusted paths are smooth and feasible.

[0037] The path cost evaluation unit establishes a dynamic path cost function to quantitatively evaluate the merits of planned paths, providing a basis for path selection and optimization. The cost function is as follows: in: This represents the total cost of the path; the smaller the value, the better the path. , , , These are the weighting coefficients for each cost factor, satisfying... It can be adaptively adjusted according to the type of scenario and task requirements; This is the cost of the path length, i.e., the actual physical length of the path; The cost of obstacle risk is calculated based on the distance between the path and the obstacle, and the obstacle's motion state. The closer the distance and the faster the obstacle's movement speed, the greater the risk. The larger; The cost for path smoothness is calculated based on the rate of change of the path's curvature; the greater the change in curvature, the lower the cost. The larger; The cost of path congestion is calculated based on the number and density of other AGVs on the path; the higher the AGV density, the higher the cost. The larger the value, the greater the cost. The total cost of each candidate path is calculated using this formula, and the path with the lowest cost is selected as the optimal path.

[0038] The path smoothing optimization unit smooths the planned path to ensure smooth AGV movement and low energy consumption. Specifically, this includes: Smoothing algorithm: The path is fitted with a third-order Bézier curve. By adjusting the position of the control points, sharp corners and sharp bends in the path are eliminated, so that the curvature of the path changes continuously. Constraints: The smoothed path must satisfy the AGV kinematic constraints and must not collide with obstacles; Optimization objective: To minimize the rate of change of curvature of the path, ensure smooth changes in acceleration and angular velocity during AGV operation, and reduce energy consumption and mechanical losses.

[0039] The scene adaptive decision-making module includes a scene recognition unit, a decision rule base unit, an adaptive matching unit, and a policy update unit; The scene recognition unit, based on scene features extracted by the multi-scene perception and environment modeling module, achieves automatic scene type recognition and classification, specifically including: Identification features: scene structure features, task features, environmental dynamic features, facility features; Recognition Algorithm: The Support Vector Machine (SVM) classification algorithm is adopted. The extracted scene features are used as input to train the scene recognition model, which can recognize scenes such as "warehouse sorting", "workshop material transfer", "logistics hub loading and unloading", and "smart factory flexible production". Recognition Update: Real-time monitoring of scene feature changes; when scene features change significantly, scene recognition is re-performed to ensure the accuracy of the recognition results.

[0040] The decision rule base unit establishes a rule base containing multi-scenario scheduling rules, path planning strategies, and parameter configurations to provide a basis for adaptive decision-making, specifically including: Subset of scheduling rules: Task allocation rules, conflict coordination rules, and priority weight configurations for different scenarios, such as prioritizing task execution efficiency in warehouse sorting scenarios and prioritizing path smoothness and obstacle avoidance safety in workshop material transfer scenarios. Path planning strategy subset: Path planning algorithm selection and cost function weight configuration for different scenarios, such as using the A algorithm to pursue the shortest path in wide-channel scenarios and the RRT algorithm to pursue path flexibility in narrow-channel scenarios; Parameter configuration subset: Optimal configuration of AGV motion parameters, sensor perception parameters, and communication parameters under different scenarios; Rule base update: Supports manual input of new scenario rules and automatic learning and updating. When the system connects to a new scenario, it can update the rule base by learning the optimal operating parameters under that scenario.

[0041] The adaptive matching unit matches the optimal scheduling rules, path planning strategies, and parameter configurations from the decision rule base based on the scene recognition results. The matching degree is calculated as follows:

[0042] in: This represents the degree of matching between the scenario and the rules / strategies; the higher the value, the higher the degree of matching. The number of dimensions for scene features; For the first The weight coefficients of each scene feature satisfy the following conditions: The settings are based on the degree of influence of features on scenario decisions; For the current scene The actual values ​​of each feature; For a certain rule / strategy in the decision rule base, the first... The standard values ​​for each feature are used. This formula is used to calculate the matching degree between the current scenario and each rule / strategy in the rule base, and the rule / strategy with the highest matching degree is selected as the execution plan for the current scenario.

[0043] The strategy update unit monitors the system's operational performance in real time and dynamically updates decision rules and strategy parameters to ensure optimal system adaptability when scenarios change. Specifically, this includes: Evaluation metrics: task completion efficiency, AGV collaboration efficiency, path planning success rate, obstacle avoidance success rate, and energy consumption level; Update trigger conditions: When the evaluation metric is lower than the preset threshold, the scene characteristics change significantly, or a new scene or task type is added; Update method: Reinforcement learning algorithm is adopted, with system operation evaluation indicators as reward signals, to continuously optimize the parameter configuration and strategy selection of decision rules, so as to realize dynamic iterative update of the rule base.

[0044] The multi-sensor fusion obstacle avoidance module includes an obstacle detection unit, a sensor data calibration unit, a fusion decision unit, and an obstacle avoidance path generation unit; The obstacle detection unit, based on sensor data from the multimodal sensing unit, achieves comprehensive detection of different types of obstacles, specifically including: Static obstacle detection: Using Euclidean clustering algorithms on LiDAR point cloud data, the location, size, and outline of fixed obstacles (shelves, equipment, walls) are identified; image segmentation algorithms from visual cameras are used to assist in confirming the type of static obstacle. Dynamic obstacle detection: Tracks the movement trajectory, speed, and direction of dynamic obstacles such as pedestrians and mobile devices using target tracking algorithms from visual cameras (such as KCF and YOLO); supplements dynamic obstacle detection in low-light environments with data from infrared sensors. Temporary obstacle detection: By fusing close-range data from ultrasonic sensors and lidar, small temporary obstacles such as scattered materials and temporarily placed tools on the ground can be identified to prevent AGVs from running over or colliding with them.

[0045] The sensor data calibration unit calibrates the data collected by each sensor to eliminate sensor errors and ensure data accuracy. Specifically, this includes: Time calibration: Based on the time synchronization signal of the heterogeneous communication and data synchronization module, the data acquisition timestamps of each sensor are aligned to avoid fusion errors caused by data time differences; Spatial calibration: Through hand-eye calibration and sensor extrinsic parameter calibration, the coordinate system of each sensor is unified to ensure that the obstacle position data detected by different sensors are consistent in the same coordinate system; Accuracy calibration: Establish a sensor error model and perform error compensation on the collected data according to the sensor type (such as the ranging error of lidar and the pixel error of vision camera) to improve data accuracy.

[0046] The fusion decision unit employs a multi-sensor data fusion algorithm to fuse calibrated sensor data, improving the accuracy and reliability of obstacle detection. The fusion confidence level is calculated as follows: in: The confidence level of obstacle detection after fusion, with a value range of [0,1]. The larger the value, the more reliable the detection result. The number of sensors participating in the fusion; For the first The reliability weight of a sensor is set based on its detection accuracy, environmental adaptability, and historical performance; the higher the reliability, the better. The larger; For the first The confidence level of an obstacle detection by each sensor is output by its own detection algorithm, such as the ranging confidence level of a LiDAR sensor or the target recognition confidence level of a visual camera. This formula is used to fuse the detection results from each sensor, resulting in a fused confidence level. When the value exceeds a preset threshold, the existence of the obstacle is confirmed, and the precise location, size, and motion status of the obstacle are output; when... If the value is less than the threshold, it is considered a suspected obstacle, and a secondary detection is initiated, such as increasing the sensor detection frequency or adjusting the detection angle.

[0047] When an obstacle is detected, the obstacle avoidance path generation unit generates a safe and reliable obstacle avoidance path, specifically including: Obstacle avoidance strategy: Different obstacle avoidance strategies are selected according to the type (static / dynamic), location, and motion state of the obstacle. The "detour" strategy is used for static obstacles, and the "predictive avoidance" strategy is used for dynamic obstacles. The collision point is predicted based on the motion trajectory, and the path is adjusted in advance. Path constraints: The obstacle avoidance path must meet the kinematic constraints of the AGV, the safe distance constraints from obstacles (set according to the type of obstacle, such as the safe distance for pedestrians being greater than that for static equipment), and not deviate from the original task objective; Generation algorithm: An improved artificial potential field method is adopted, which treats the AGV as a "particle". Obstacles generate a repulsive potential field and target points generate an attractive potential field. By calculating the potential field forces, an obstacle avoidance path is generated to ensure that the path is smooth and the obstacle avoidance effect is reliable.

[0048] The heterogeneous communication and data synchronization module includes a multi-protocol communication unit, a data transmission optimization unit, a time synchronization unit, and a data consistency verification unit; The multi-protocol communication unit supports multiple communication protocols to adapt to the communication needs of different scenarios, specifically including: Protocol types: WiFi 6 (suitable for scenarios with high bandwidth requirements and medium communication distance, such as warehousing and logistics hubs), Bluetooth 5.0 (suitable for short-range, low-power communication between AGVs, such as workshop AGV collaboration), LoRa (suitable for long-range, low-speed scenarios, such as large warehouses and outdoor logistics hubs), Ethernet (suitable for wired communication between fixed equipment and the dispatch center, such as workshop control stations), CAN bus (suitable for communication between internal modules of AGVs). Protocol adaptive switching: Automatically select the optimal communication protocol according to the communication environment of the scenario. For example, when there is severe electromagnetic interference in the workshop, switch to the LoRa protocol with strong anti-interference ability; when the bandwidth demand in the warehouse is high, switch to the WiFi6 protocol. Communication interface: Equipped with a multi-protocol communication module, it enables unified access and switching of different protocols through interface adapter circuit.

[0049] The data transmission optimization unit optimizes the data transmission process to ensure the real-time performance and reliability of data transmission, specifically including: Data classification and transmission: Data is divided into critical data (AGV position, task instructions, obstacle information) and non-critical data (historical operation data, status statistics). Critical data adopts a priority transmission mechanism, which prioritizes communication bandwidth to ensure real-time transmission; non-critical data adopts batch transmission or transmission during idle periods to reduce the impact on critical data. Transmission error control: CRC check and retransmission mechanism are used to detect and correct errors in transmitted data. When data transmission is lost or erroneous, retransmission is automatically triggered to ensure data integrity. Adaptive bandwidth adjustment: Dynamically adjusts the data transmission rate and data packet size according to changes in the bandwidth of the communication environment to avoid data congestion caused by insufficient bandwidth.

[0050] The time synchronization unit enables time synchronization between the scheduling center, each AGV, and scene devices, ensuring data and decision consistency. Specifically, it includes: Synchronization reference: The Network Time Protocol (NTP) is adopted, and the clock of the dispatch center is used as the reference clock. Time synchronization signals are sent to each AGV and device through the communication network. Synchronization frequency: The synchronization frequency is adaptively adjusted according to the needs of the scenario. In scenarios with dense multi-AGV collaboration (such as material transfer in the workshop), the synchronization frequency is higher than that in single-AGV operation scenarios (such as single-point pickup in warehouse). Synchronization accuracy: Through timestamp calibration and delay compensation algorithms, the time synchronization accuracy of each node is ensured to reach the millisecond level, meeting the time consistency requirements of multi-AGV collaborative scheduling and path planning.

[0051] The data consistency verification unit ensures data consistency between the scheduling center and each AGV, avoiding scheduling errors caused by data inconsistency. Specifically, it includes: Data verification includes: AGV status data, task execution data, environmental map data, and path planning data. Verification mechanism: A distributed data consistency algorithm is adopted to elect a master node (scheduling center), and each AGV is a slave node. The master node periodically sends data synchronization instructions to the slave nodes, and the slave nodes compare their local data with the master node's data. If there is a discrepancy, the data is updated. Conflict resolution: When data conflicts occur between multiple nodes, the data from the master node shall prevail, or the conflict shall be resolved based on the timestamp and reliability of the data to ensure that the final data is consistent.

[0052] The full-process monitoring and fault diagnosis module includes a status monitoring unit, a fault detection unit, a fault location unit, an emergency handling unit, and a data recording and analysis unit; The status monitoring unit collects real-time operational status data of the AGV and various modules of the system, enabling full-process monitoring, specifically including: AGV operating status monitoring: AGV position, speed, heading angle, remaining battery power, load status, motor speed, steering angle, and sensor working status; Task execution status monitoring: task execution progress, task completion quality, task delay status, and subtask execution status; System module status monitoring: working status, data transmission status, and computing resource usage of each core module; Monitoring data display: Through the visual interface of the dispatch center, the location distribution of AGVs, task execution progress, and system module status are displayed in real time, allowing managers to intuitively grasp the system operation status.

[0053] The fault detection unit automatically detects faults based on monitored status data, specifically including: Fault types: AGV hardware faults (motor faults, sensor faults, battery faults, communication module faults), software faults (algorithm execution errors, data processing anomalies, program crashes), system faults (unresolved scheduling conflicts, path planning failures, communication interruptions); Detection methods include: threshold detection (e.g., battery voltage below threshold, motor current exceeding threshold), trend analysis (e.g., abnormal fluctuation trend of sensor data), model prediction (establishing a normal operating state model, and judging it as a fault if it deviates from the model range), and redundancy detection (comparing data from multiple sensors, and judging it as a sensor fault if the data is inconsistent and exceeds the error range). Detection and alarm: After fault detection, an alarm signal is immediately triggered, and the time, location, and relevant status data of the fault occurrence are recorded.

[0054] The fault location unit accurately locates the detected fault, determining the module, component, and cause of the fault, specifically including: Hardware fault location: By analyzing fault codes (fault feedback codes of each hardware module), circuit testing (such as motor drive circuit and sensor power supply circuit), and component replacement testing, faulty hardware components can be located, such as a faulty lidar or a faulty left motor. Software fault localization: Through log analysis and breakpoint debugging, locate the code location and cause of software faults, such as incorrect algorithm parameters or abnormal data format; System fault location: By backtracking data (system status data and communication data before the fault occurred) and conflict analysis (scheduling conflicts and data conflict records), the root cause of the system fault (such as unreasonable scheduling rules or communication protocol mismatch) can be located.

[0055] The emergency response unit takes corresponding emergency measures for different types of faults to reduce the impact of the faults and ensure the continuous operation of the system. These measures include: Hardware failure emergency response: When the AGV motor fails, immediately stop the AGV's operation and the dispatch center will reassign the task to other available AGVs; when the sensor fails, enable redundant sensor data or switch to a degraded operation mode (such as retaining only the core sensors, reducing path planning accuracy but ensuring basic operation); when the battery fails, guide the AGV to the nearest charging area. Software failure emergency response: When the program crashes, the relevant software modules will be automatically restarted and normal operation will be restored; when the algorithm execution is incorrect, the backup algorithm will be switched to. For example, when the path planning algorithm fails, a simplified backup algorithm will be activated. System failure emergency response: When communication is interrupted, the AGV switches to local autonomous operation mode and continues to execute tasks based on the local map and task information. Data is synchronized after communication is restored. When scheduling conflicts cannot be resolved, low-priority tasks are suspended and high-priority tasks are given priority in execution. Safety Emergency: If a malfunction may pose a safety risk, the emergency stop mechanism will be triggered immediately to cut off the AGV's power supply and ensure the safety of personnel and equipment.

[0056] The data recording and analysis unit records system operation data and fault data, providing a basis for system optimization and maintenance, specifically including: Data recording content includes: AGV operating status data, task execution data, fault data (fault type, occurrence time, location, cause, and handling result), scheduling decision data, path planning data, and sensor data; Data storage: A distributed database is used to store massive amounts of operational data, supporting long-term data preservation and fast retrieval; Data Analysis: Employing big data analytics algorithms, we analyze patterns in fault occurrence, system bottlenecks, and optimization directions for path planning, providing data support for system upgrades and maintenance plan development.

[0057] A multi-scenario adaptive AGV path planning method, based on a multi-scenario adaptive AGV scheduling and management system, includes the following steps: Step 1: System Initialization and Scene Awareness The multi-scenario adaptive AGV scheduling and management system is launched, and each module completes initialization. The multimodal perception unit of the multi-scenario perception and environment modeling module begins to collect environmental data, the scene feature extraction unit extracts scene type features, environmental structure features, and dynamic element features, the dynamic environment update unit initializes dynamic environment data, and the unified map generation unit generates the initial global unified map.

[0058] Step 2: Scene Recognition and Decision Matching The scene recognition unit of the scene adaptive decision-making module identifies the current scene type based on extracted scene features using an SVM classification algorithm; the decision rule base unit calls the corresponding scene rule subset, and the adaptive matching unit uses formulas... Calculate the matching degree between the current scenario and each rule / strategy, and select the scheduling rule, path planning strategy and parameter configuration with the highest matching degree.

[0059] Step 3: Task Reception and Parsing The system receives task instructions from the upper-level management system. The task parsing unit breaks down the task (complex tasks are broken down into subtasks), extracting basic task information (start point, end point, type, requirements) and related information (task dependencies, scenario-related features); the task priority ranking unit uses formulas... Calculate the priority of each task and form a task execution queue.

[0060] Step 4: Multi-AGV Status Assessment and Task Assignment The AGV status assessment unit collects real-time operating status data (position, power, load, fault status, etc.) of each AGV to assess the AGV's execution capability; the resource allocation optimization unit uses an improved genetic algorithm to allocate tasks to the optimal AGV based on the task execution queue and AGV status assessment results; the conflict coordination unit detects and coordinates resource conflicts during the task allocation process to ensure that the allocation result is reasonable.

[0061] Step 5: Global Path Pre-planning and Cost Assessment The global path pre-planning unit of the dynamic path planning and optimization module generates global initial paths for each AGV based on a unified map and matching path planning strategy; the path cost evaluation unit uses formulas... The total cost of the path is calculated, and the path with the minimum cost is selected as the candidate global path. The path smoothing optimization unit uses a third-order Bézier curve to smooth the candidate global path to ensure the continuity of the path curvature.

[0062] Step 6: Dynamic Path Adjustment and Obstacle Avoidance The multi-sensor fusion obstacle avoidance module detects obstacles in real time, and the fusion decision unit uses formulas... The obstacle detection confidence level is calculated, and after confirming the existence of the obstacle, the obstacle avoidance path generation unit generates an obstacle avoidance path; the local path dynamic adjustment unit makes local adjustments to the global path based on obstacle information, AGV real-time status (such as deviation from the path), and task changes (such as priority adjustment) to ensure that the path dynamically adapts to environmental changes.

[0063] Step 7: AGV Execution and Status Monitoring The AGV executes tasks according to the planned path and scheduling instructions. The heterogeneous communication and data synchronization module ensures that the data between the AGV and the scheduling center is synchronized in real time. The full-process monitoring and fault diagnosis module monitors the AGV's operating status and task execution progress in real time. The fault detection unit detects faults and triggers alarms. The fault location unit locates the cause of the fault. The emergency handling unit takes corresponding emergency measures.

[0064] Step 8: Task Completion and System Optimization After the AGV reaches the target point and completes the task, it reports the task completion information to the scheduling center. The data recording and analysis unit records the running data of this task (path information, energy consumption, execution time). The strategy update unit of the scenario adaptive decision module optimizes the decision rules and parameter configuration based on the task execution effect (such as completion efficiency and obstacle avoidance success rate). The system is reset to standby state, waiting to receive new task instructions.

[0065] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-scene adaptive AGV scheduling management system, characterized in that: The multi-scene perception and environment modeling module, the multi-AGV task scheduling and resource allocation module, the dynamic path planning and optimization module, the scene adaptive decision module, the multi-sensor fusion obstacle avoidance module, the heterogeneous communication and data synchronization module, and the whole-process monitoring and fault diagnosis module are included. The multi-scene perception and environment modeling module is used for collecting multi-scene environment data, extracting scene features, dynamically updating environment information, and generating a unified map. The multi-AGV task scheduling and resource allocation module is used for analyzing task information, evaluating AGV states, sorting task priorities, allocating task resources, and coordinating multi-AGV conflicts. The dynamic path planning and optimization module is used for generating a global path, dynamically adjusting a local path, evaluating a path cost, and smoothing and optimizing a path. The scene adaptive decision module is used for identifying scene types, matching optimal scheduling and path strategies, and dynamically updating decision rules. The multi-sensor fusion obstacle avoidance module is used for detecting obstacles, calibrating sensor data, fusing sensor information, and generating an obstacle avoidance path. The heterogeneous communication and data synchronization module is used for realizing multi-protocol communication, optimizing data transmission, synchronizing node times, and checking data consistency. The whole-process monitoring and fault diagnosis module is used for monitoring system and AGV states, detecting and locating faults, executing emergency handling, and recording and analyzing data. 2.The multi-scene adaptive AGV scheduling management system according to claim 1, characterized in that: The multi-scene perception and environment modeling module includes a multi-modal perception unit, a scene feature extraction unit, a dynamic environment update unit, and a unified map generation unit. The multi-modal perception unit integrates a laser radar, a visual camera, an infrared sensor, an RFID card reader, and an ultrasonic sensor to collect environment three-dimensional point cloud, image, label, and distance data. The scene feature extraction unit extracts scene type features, environment structure features, dynamic element features, and task association features. The dynamic environment update unit updates feature data of static environment, dynamic environment, and abnormal environment. The unified map generation unit generates a global map in a grid-point cloud hybrid format including a static layer, a dynamic layer, and a task layer. 3.The multi-scene adaptive AGV scheduling management system according to claim 1, characterized in that: The multi-AGV task scheduling and resource allocation module includes a task analysis unit, an AGV state evaluation unit, a task priority sorting unit, a resource allocation optimization unit, and a conflict coordination unit. The task analysis unit disassembles tasks and extracts basic information and associated information of the tasks. The AGV state evaluation unit evaluates basic states, performance states, and task states of AGVs. The task priority sorting unit calculates task priorities through a task priority calculation model, and the task priority calculation model is as follows: wherein is a task priority, , , , is a weight coefficient and satisfies , is a task urgency coefficient, is a task complexity coefficient, is a task value coefficient, is an AGV fitness coefficient; The resource allocation optimization unit allocates task resources by using an improved genetic algorithm. The conflict coordination unit detects and coordinates path conflicts, resource conflicts, and task conflicts. 4.The multi-scene adaptive AGV scheduling management system according to claim 1, characterized in that: The dynamic path planning and optimization module includes a global path pre-planning unit, a local path dynamic adjustment unit, a path cost evaluation unit, and a path smoothing and optimization unit. The global path pre-planning unit generates a global initial path by using an improved A algorithm. The local path dynamic adjustment unit adjusts a local path by using an RRT algorithm. The path cost evaluation unit evaluates the path by a dynamic path cost function, which is: wherein is a total path cost, , , , is a weight coefficient and satisfies , is a path length cost, is an obstacle risk cost, is a path smoothness cost, is a path congestion cost. The path smoothing and optimization unit smoothes a path by using a third-order Bezier curve.

5. The multi-scene adaptive AGV scheduling management system according to claim 1, characterized in that: The scene adaptive decision module comprises a scene recognition unit, a decision rule library unit, an adaptive matching unit and a strategy updating unit. The scene recognition unit adopts an SVM classification algorithm to recognize the scene type. The decision rule library unit stores a scheduling rule subset, a path planning strategy subset and a parameter configuration subset. The adaptive matching unit calculates the optimal matching rule and strategy through scene adaptive matching degree, and the scene adaptive matching degree calculation formula is: wherein is a matching degree, is a number of scene feature dimensions, is a weight coefficient of the th feature and satisfies , is an actual value of the th feature of the current scene, is a standard value of the th feature corresponding to the rule; The strategy updating unit updates the decision rule and parameter by using a reinforcement learning algorithm. 6.The multi-scene adaptive AGV scheduling management system according to claim 1, characterized in that: The multi-sensor fusion obstacle avoidance module comprises an obstacle detection unit, a sensor data calibration unit, a fusion decision unit and an obstacle avoidance path generation unit. The obstacle detection unit detects static obstacles, dynamic obstacles and temporary obstacles. The sensor data calibration unit performs time calibration, space calibration and precision calibration. The fusion decision unit fuses sensor data through a multi-sensor fusion confidence model, and the multi-sensor fusion confidence model is: wherein is a fusion confidence, is a number of sensors, is a reliability weight of the th sensor, is a detection confidence of the th sensor; The obstacle avoidance path generation unit generates an obstacle avoidance path by using an improved artificial potential field method. 7.The multi-scene adaptive AGV scheduling management system according to claim 1, characterized in that: The heterogeneous communication and data synchronization module comprises a multi-protocol communication unit, a data transmission optimization unit, a time synchronization unit and a data consistency verification unit. The multi-protocol communication unit supports WiFi6, Bluetooth 5.0, LoRa, Ethernet and CAN bus protocols, and can adaptively switch communication protocols. The data transmission optimization unit optimizes data transmission by using classification transmission, error control and bandwidth adaptive adjustment. The time synchronization unit realizes time synchronization based on the NTP protocol. The data consistency verification unit ensures data consistency by using the Raft algorithm. 8.The multi-scene adaptive AGV scheduling management system according to claim 1, characterized in that: The full-process monitoring and fault diagnosis module comprises a state monitoring unit, a fault detection unit, a fault positioning unit, an emergency handling unit and a data recording and analysis unit. The state monitoring unit monitors the AGV running state, task execution state and system module state. The fault detection unit detects faults by using threshold detection method, trend analysis method, model prediction method and redundancy detection method. The fault positioning unit locates faults by fault code analysis, circuit detection, component replacement test and log analysis. The emergency handling unit takes emergency measures for hardware faults, software faults, system faults and safety risks. The data recording and analysis unit stores running data and fault data, and analyzes data by using big data analysis algorithm. 9.A multi-scenario adaptive AGV path planning method based on the multi-scenario adaptive AGV scheduling management system of any one of claims 1-8, characterized in that: The steps comprise: Step 1: system initialization and scene perception, each module is initialized, and the multi-scene perception and environment modeling module collects environment data and generates an initial global map; Step 2: scene recognition and decision matching, the scene adaptive decision module recognizes the scene type and matches the optimal scheduling and path strategy; Step 3: task receiving and analysis, the task analysis unit disassembles the task and extracts task information, and the task priority sorting unit sorts the task priority; Step 4: multi-AGV state evaluation and task allocation, the AGV state evaluation unit evaluates the AGV state, the resource allocation optimization unit allocates task resources, and the conflict coordination unit coordinates conflicts. Step 5: Global path pre-planning and cost evaluation, the global path pre-planning unit generates a global initial path, the path cost evaluation unit evaluates the path cost, and the path smoothing optimization unit smooths the path; Step 6: Dynamic path adjustment and obstacle avoidance, the multi-sensor fusion obstacle avoidance module detects obstacles and generates an obstacle avoidance path, and the local path dynamic adjustment unit adjusts the local path; Step 7: AGV execution and state monitoring, AGV executes tasks, heterogeneous communication and data synchronization module synchronizes data, and full-process monitoring and fault diagnosis module monitors state and handles faults; Step 8: Task completion and system optimization, AGV feeds back task completion information, data recording and analysis unit records and analyzes data, and strategy updating unit optimizes decision rules and parameters. 10.The multi-scene adaptive AGV path planning method of claim 9, wherein: In step 2, scene recognition uses SVM classification algorithm, and decision matching is realized through scene adaptive matching degree calculation; in step 3, task priority is calculated through a task priority calculation model; in step 5, path cost is evaluated through a dynamic path cost function; in step 6, obstacle detection confidence is calculated through a multi-sensor fusion confidence model.

Images