Multi-cleaning-robot formation cleaning method and system based on wireless communication network

Abstract

The invention discloses a multi-cleaning-robot formation cleaning method and system based on a wireless communication network. The multi-cleaning-robot formation cleaning method comprises the following steps that S1, a single robot autonomously perceives and performs environment modeling; s2, robot dynamic identification and adaptive networking; s3, global information fusion and task planning; and S4, performing formation cooperative execution and closed loop verification. The cleaning device has the beneficial effects that the cleaning efficiency is improved, repeated labor and missing areas are reduced, and the overall cleaning effect is improved; the robustness of the system is enhanced, the intelligent sensing ability of the robot is improved, the precise execution of the cleaning task is ensured, and the misoperation and collision risks are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network communication, in particular to a multi-cleaning robot formation cleaning method and system based on a wireless communication network. BACKGROUND

[0002] At present, a multi-robot cleaning system mainly realizes task allocation and collaborative work through a cloud scheduling platform, and typical technical solutions include the following two types: Centralized scheduling system: relying on a central server or a cloud platform, a robot is allocated a cleaning area according to a preset algorithm (such as a genetic algorithm or an auction algorithm), and the robot receives instructions and feeds back a state through a 4G / 5G network.

[0003] Static partition cleaning system: the cleaning area is manually divided by an operator before the task, each robot independently executes a fixed block task, and only a simple obstacle avoidance algorithm (such as an infrared sensor) is used to avoid collision.

[0004] Application and structural characteristics of the prior art: Hardware dependence: robots are usually equipped with basic sensors (such as ultrasonic waves and infrared rays) for obstacle avoidance, and some high-end models are equipped with single-line laser radars to realize SLAM mapping; Communication architecture: a "cloud-end" mode is adopted, and the robot needs to be continuously connected to the cloud server to obtain instructions; Task allocation logic: based on static priority (such as distance priority) or fixed time slice rotation to allocate tasks.

[0005] Therefore, the problems of the prior art are also obvious, for example: 1. Collision risk of cleaning robots: The recognition ability of existing cleaning robots is generally directed to static obstacles or common obstacles such as people, vehicles, and non-motor vehicles, and the recognition and behavior prediction ability for other cleaning robots is insufficient, which has potential collision risk; 2. Identification information is not shared: The garbage to be cleaned and the risk area recognized by a single robot are not shared with other robots, so that when multiple robots are dispatched, it is impossible to achieve global cleaning optimization and global obstacle avoidance optimization.

[0006] 3. Network environment dependence: The existing multi-robot scheduling relies on a cloud scheduling platform, and in a weak network or no network environment, collaborative work cannot be completed, and only an operator can manually divide a cleaning area and a cleaning task. SUMMARY

[0007] The purpose of the present application is to provide a multi-cleaning robot formation cleaning method based on a wireless communication network to solve the problems raised in the background art.

[0008] To achieve the above object, the present application provides the following technical solutions: In a first aspect, the present application provides a multi-cleaning robot formation cleaning method based on a wireless communication network, comprising the following steps: S1. Single robot autonomous perception and environment modeling: Each robot collects environmental data in real time through a multi-sensor fusion system, identifies obstacle avoidance areas, cleaning areas and special attention objects using target detection algorithms and semantic segmentation algorithms, and constructs a local grid map through a SLAM algorithm; S2. Robot dynamic identification and adaptive networking: After detecting other collaborative robots, the robot triggers a near-field communication protocol, establishes a Wi-Fi Direct ad hoc network, and elects the robot with the smallest number as the formation controller based on pre-assigned internal numbers; S3. Global information fusion and task planning: The formation controller integrates the local maps and perception information uploaded by each robot, generates a global semantic map, and divides the cleaning area into mutually exclusive sub-tasks based on a work time constant division algorithm and assigns them to each robot; S4. Formation cooperative execution and closed-loop verification: The controller issues cleaning instructions, robots with the same function execute tasks using a block method, and robots with complementary functions use a leader-following method, and after completing the task, the cleaning effect is verified by a visual sensor, and the area that does not meet the standard triggers secondary cleaning.

[0009] Preferably, in step S1, the multi-sensor fusion system includes a visual sensor, a laser radar, an IMU and a GPS, the target detection algorithm is a YOLO series algorithm, and the semantic segmentation algorithm is used to distinguish the geometric range and pollution level of obstacle avoidance areas, cleaning areas and special attention objects.

[0010] Preferably, in step S2, the near-field communication protocol includes a collision avoidance strategy and a communication establishment strategy, the collision avoidance strategy adjusts the motion trajectory in real time through a speed obstacle method, and the communication establishment strategy realizes the exchange of global positioning information and task state between robots through Wi-Fi Direct in a weak or no network environment.

[0011] Preferably, in step S3, the global semantic map is generated by spatiotemporal alignment algorithm and graph optimization method to splice local maps, and the task allocation is divided into floor washing, dust collection or disinfection mutually exclusive sub-tasks based on the function type of the robot.

[0012] In a second aspect, an embodiment of the present application provides a multi-cleaning robot formation cleaning system based on perception interaction, comprising the following modules: Module A: Autonomous perception and environment modeling module, configured in each robot, including a multi-sensor fusion unit, a target detection unit and a semantic segmentation unit, for collecting environmental data in real time and constructing a local grid map; Module B: dynamic identification and networking module, configured in each robot, including a near field communication unit and a master controller election unit, for detecting other robots and establishing an ad hoc network, and electing a formation controller; Module C: global information fusion and task planning module, configured in the formation controller, including a map stitching unit and a task allocation unit, for integrating a global semantic map and dividing mutually exclusive sub-tasks; Module D: formation execution and verification module, configured in each robot and controller, including a block cleaning unit, a navigation-following unit and a visual verification unit, for executing cleaning instructions and feeding back cleaning results.

[0013] Preferably, the multi-sensor fusion unit in module A includes a vision sensor, a laser radar, an IMU and a GPS, the target detection unit adopts a YOLO series algorithm, and the semantic segmentation unit determines a cleaning area priority through a pollution level. Preferably, the near field communication unit in module B supports a Wi-Fi Direct protocol, and the master controller election unit realizes automatic upgrading of the controller based on a pre-assigned internal number priority.

[0014] Preferably, the task allocation unit in module C dynamically allocates a washing, dusting or disinfecting sub-task according to a robot function type, and optimizes path planning based on a work time constant division algorithm.

[0015] Compared with the prior art, the present application has the following advantages: 1. The present application ensures mutual exclusion of tasks of each robot and optimizes paths by global information fusion and task planning, task allocation based on a work time constant division algorithm, significantly improves cleaning efficiency, reduces repetitive labor and missed areas, and improves overall cleaning effect.

[0016] 2. The present application can realize stable communication and cooperative operation in a weak network / no network environment through a Wi-Fi Direct ad hoc network, enhances the robustness of the system, and ensures smooth completion of cleaning tasks even in a complex environment.

[0017] 3. The present application dynamically allocates scene cleaning tasks to available robots through formation execution tasks, and is more flexible in use.

[0018] 4. The present application can identify and classify cleaning areas, obstacle avoidance areas and special attention objects in real time through a multi-sensor fusion system (vision, laser radar, IMU, GPS) and advanced perception algorithms (such as YOLO series, semantic segmentation algorithm), improving the intelligent perception ability of the robot, ensuring the accurate execution of the cleaning task, and reducing the risk of misoperation and collision.

[0019] 5. The present application supports multiple functional modules (such as floor washing, dust collection, disinfection), and dynamically allocates robot functions according to task requirements, improving the functional diversity and adaptability of the system, and meeting the cleaning needs in different scenarios.

[0020] 6. The present application detects cleaning effect through visual sensor and triggers secondary cleaning in unqualified area, ensuring high-quality completion of cleaning task and improving product use experience. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The method flowchart of the embodiment of the present application is shown in the figure. Figure 2 The schematic diagram of the multi-sweeping robot formation of the present application is shown in the figure. Figure 3 The system module diagram of the present application is shown in the figure. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] Embodiment 1 Please refer to Figure 1 、 Figure 2 The present embodiment provides a technical solution: a multi-cleaning robot formation cleaning method based on wireless communication network, comprising the following steps: S1. Single robot autonomous perception and environment modeling: each robot collects environmental data in real time through a multi-sensor fusion system, identifies obstacle avoidance areas, cleaning areas and special attention objects using target detection algorithm and semantic segmentation algorithm, and constructs a local grid map through SLAM algorithm.

[0024] In the present embodiment, the multi-sensor fusion system includes visual sensor, laser radar, IMU and GPS. The target detection algorithm is YOLO series algorithm, and the semantic segmentation algorithm is used to distinguish the geometric range and pollution level of obstacle avoidance areas, cleaning areas and special attention objects.

[0025] In one specific embodiment, the specific process of detecting obstacles by YOLO series algorithm includes data collection, target detection, and coordinate conversion, which includes the following processes: The robot collects environmental images in real time through visual sensors such as RGB cameras; The input image is sent to a pre-trained YOLO model (such as YOLOv5 or YOLOv8), which outputs the bounding box, class, and confidence of all detected targets in the image through a single forward propagation; Obstacle classification: the YOLO model identifies common obstacles (such as pedestrians, furniture, vehicles, etc.) and labels their bounding box coordinates and class labels (such as "pedestrian" and "table"); Convert the detected obstacle two-dimensional pixel coordinates (image coordinate system) to the three-dimensional space coordinate system of the robot (such as laser radar coordinate system or global coordinate system) through camera calibration parameters; Combine laser radar point cloud data to further verify the physical location of the obstacle and eliminate false positives (such as virtual shadows caused by mirror reflection).

[0026] In one specific embodiment, the process of dividing cleaning areas, cleaning areas, and special attention objects by semantic segmentation algorithm includes: Use a pre-trained semantic segmentation model (such as U-Net or DeepLab) to classify each pixel of the ground image and output the classification label. Further, the classification label includes an obstacle avoidance area, a cleaning area, and a cleaned area, the obstacle avoidance area is marked as an impassable area (such as the bottom of furniture, the edge of a step); the cleaning area is divided into different pollution levels (high, medium, and low) according to the ground texture (such as stains, dust density); the cleaned area is identified by color or texture difference.

[0027] Align the segmentation result with the local grid map constructed by SLAM, assign a pollution level label to each grid, and form a semantic map.

[0028] As the robot moves, update the pollution status of the ground area in real time.

[0029] In one specific embodiment, constructing a local grid map by SLAM algorithm includes extracting feature points (such as corners and furniture edges) from laser radar scan data, matching consecutive frame point clouds using ICP (Iterative Closest Point) algorithm, and estimating the robot pose change.

[0030] Capture consecutive frame images through the camera, calculate the optical flow or feature matching, and further optimize the pose estimation.

[0031] Fuse multi-sensor data (Lidar, IMU, GPS) and loop closure results using a graph optimization framework (e.g. g2o or GTSAM) to construct a globally consistent grid map.

[0032] When the robot re-enters an explored area, recognize loop closure through feature matching or Lidar point cloud comparison, and correct map accumulated errors.

[0033] Embed obstacle positions detected by YOLO and clean area labels segmented by semantic segmentation into the grid map to form a hybrid map containing both geometric and semantic information. For example, mark "highly contaminated areas" as red, "static obstacles" as black, and "cleaned areas" as green in the grid map.

[0034] S2. Dynamic identification and adaptive networking of robots: When the robot detects other collaborative robots, it triggers a near-field communication protocol, establishes a Wi-Fi Direct ad hoc mesh network, and elects the robot with the smallest internal number as the formation controller based on pre-assigned internal numbers.

[0035] In this embodiment, the near-field communication protocol includes a collision avoidance strategy and a communication establishment strategy. The collision avoidance strategy adjusts the motion trajectory in real time through the velocity obstacle method, and the communication establishment strategy exchanges global positioning information and task status between robots through Wi-Fi Direct in weak or no network environment.

[0036] In a specific embodiment, the process of adjusting the motion trajectory in real time through the velocity obstacle method includes: The robot scans the surrounding environment at a frequency of 10 Hz through the Lidar to obtain 360° point cloud data and identify the contours and distances of obstacles (including other robots).

[0037] The visual sensor (RGB camera) synchronously collects images and uses the YOLOv5 model to detect dynamic obstacles (such as other robots) to obtain their bounding boxes and class labels.

[0038] Based on the Kalman filter or SORT (Simple Online and Realtime Tracking) algorithm, continuously track the positions, velocities, and directions of other robots.

[0039] According to the current velocity vector (linear velocity and angular velocity), predict the motion trajectory of other robots in the next 1-2 seconds.

[0040] Based on the predicted trajectories of itself and other robots, calculate the relative velocity vector between them to generate a collision cone (Collision Cone) to determine the potential collision area.

[0041] Within the possible adjustment range of the robot's current speed (e.g. max acceleration / deceleration 0.5 m / s²), select the velocity vector that deviates the trajectory from the collision cone.

[0042] Adjust the linear or angular velocity of the robot according to the velocity obstacle calculation result, generate a collision-free path.

[0043] If it is predicted that complete avoidance is not possible (e.g. another robot suddenly stops), trigger emergency braking (decelerate to a stop) and send an alert to the formation controller.

[0044] In one specific embodiment, the process of establishing a Wi-Fi Direct ad hoc mesh network includes: Each robot periodically sends a broadcast signal (containing a unique ID and a functional tag) to search for nearby cooperative robots.

[0045] When another robot is detected, send a connection request and negotiate communication parameters (such as channel, encryption method), establish a point-to-point connection.

[0046] Based on the SLAM-based grid map coordinates (e.g. x=3.2m, y=5.1m, accuracy ±0.1m) or GPS latitude and longitude coordinates (e.g. longitude 113.5°E, latitude 22.3°N, accuracy ±1m).

[0047] Encapsulate the positioning information in JSON format and periodically broadcast (e.g. once every second) to all robots in the network.

[0048] Share the task status, and when the status changes (e.g. battery level below 20%), trigger an event-driven data push immediately. Task status includes: remaining battery level: in percentage form (e.g. 78%), cleaning mode: dust collection, floor washing or disinfection, current task: block number of cleaning or pilot-following state.

[0049] Each robot periodically sends a heartbeat packet (e.g. interval 1 second) to detect the survival status of network nodes.

[0050] If a node is detected to be offline (not responding for 3 consecutive times), trigger automatic reconnection or switch to a backup communication protocol.

[0051] In one specific embodiment, based on pre-allocated internal numbers, the robot with the smallest number is elected as the formation controller, which includes each robot pre-allocating a unique internal number (e.g. ID=001 to ID=003), and the one with the smallest number (ID=001) automatically becomes the formation controller. The controller's responsibilities include: integrating the global semantic map; assigning cleaning tasks and path planning; monitoring network status and coordinating fault recovery.

[0052] S3. Global information fusion and task planning: The formation controller integrates the local maps and perception information uploaded by each robot, generates a global semantic map, and divides the cleaning area into mutually exclusive sub-tasks based on the work time division algorithm and assigns them to each robot.

[0053] In this embodiment, the global semantic map is generated by splicing local maps through a space-time alignment algorithm and a graph optimization method, and the task allocation is divided into washing, dusting or disinfecting mutually exclusive sub-tasks based on the functional type of the robot.

[0054] In a specific embodiment, the formation controller integrates the local maps and perception information uploaded by each robot to generate a global semantic map, including data reception and preprocessing, space-time alignment and map splicing, and dynamic information integration. Specifically, data reception and preprocessing includes: The local map (resolution 5 cm x 5 cm) constructed by each robot through SLAM contains geometric structure, obstacle position and semantic label (such as pollution level); Real-time position, motion trajectory, task status (remaining power, cleaning mode) of other robots.

[0055] Check data integrity (such as whether the timestamp is continuous), filter abnormal data (such as coordinates beyond the map boundary).

[0056] The process of space-time alignment and map splicing includes: Align the timestamps of each robot's data through NTP protocol to ensure temporal consistency; Align each local map based on the reference points of the SLAM map (such as pre-calibrated two-dimensional code labels); Convert to a unified geographic coordinate system (such as WGS-84) through GPS coordinates.

[0057] Use a graph optimization framework (such as g2o or GTSAM) to match the geometric features (such as corners, corridor boundaries) of overlapping areas and optimize the geometric consistency of the global map.

[0058] Fuse semantic labels (such as "high pollution area" "dynamic obstacle") into the global map to form a hybrid semantic map.

[0059] Specifically, the process of dynamic information integration includes: According to the dynamic obstacle information (such as pedestrian trajectory) uploaded by each robot, mark temporary obstacle avoidance areas (red areas) in the global map; Receive real-time new stain information detected by each robot and update the pollution level of the corresponding grid (such as from "low pollution" to "medium pollution").

[0060] In one specific embodiment, the formation controller integrates local maps and perception information uploaded by each robot to generate a global semantic map, which includes: Step 1: Task Quantification and Priority Sorting Task quantification includes: Pollution level weight: Assign different weights (e.g., 3, 2, 1) according to pollution levels (high, medium, low); Cleaning area calculation: Based on the grid resolution in the global semantic map (e.g., 5 cm x 5 cm), calculate the area of each sub-task; Equivalent working time: Combine the robot's functional type to calculate the time required to clean a unit area (e.g., a scrubbing robot takes 2 minutes to clean 1㎡, and a vacuum robot takes 1.5 minutes).

[0061] Priority sorting includes: High-pollution areas (weight 3) are assigned first, and medium and low-pollution areas are assigned according to the remaining time; The task queue is arranged in descending order of priority to ensure that critical areas are processed quickly.

[0062] Step 2: Robot Capability Assessment. Includes: State parameter collection: Remaining power: In percentage form (e.g., robot A has 80% remaining, and robot B has 60% remaining); Cleaning efficiency: According to historical data, calculate the cleaning area per unit time for each robot (e.g., robot A has a scrubbing efficiency of 10㎡ / h, and robot B has a vacuuming efficiency of 15㎡ / h); Current position: Obtain the real-time coordinates of the robot through the global semantic map and calculate the path distance to the task area.

[0063] Comprehensive capability score: Formula: Weight coefficients (α, β, γ) are dynamically adjusted according to the scene (e.g., α weight is increased when power is insufficient).

[0064] Step 3: Task Allocation Strategy. Includes: Initial allocation: Traverse the task queue and assign tasks to the robot with the highest comprehensive capability score in turn; Update the robot's cumulative workload after allocation (equivalent working time = task area x unit time).

[0065] Load balancing adjustment: Calculate the standard deviation of the cumulative workload of all robots, and if it exceeds the threshold (e.g., standard deviation > 10 minutes), trigger redistribution; Transfer part of the tasks of overloaded robots to low-load robots, and preferentially transfer low-priority tasks.

[0066] Step 4: Dynamic adjustment mechanism. Includes: Fault response: If a robot is offline or has less than a threshold amount of power (e.g., <20%), immediately retrieve its unfinished tasks and redistribute them according to the remaining robot capabilities. New task processing: When a new stain or temporary obstacle is detected, insert it into the task queue and reorder it; Prioritize calling idle robots or low-load robots to handle.

[0067] S4. Formation coordination execution and closed-loop verification: the controller issues cleaning instructions, robots with the same function use block method to execute tasks, and robots with complementary functions use leader-following method. After completing the task, the cleaning effect is verified by the vision sensor, and the area that does not meet the standard triggers secondary cleaning.

[0068] In this embodiment, the controller issues cleaning instructions, and robots with the same function use block method to execute tasks, including task block division, parallel operation, and dynamic obstacle avoidance.

[0069] Specifically, task area division includes: dividing the cleaning area into isochronous task blocks according to equivalent working time (e.g., each block cleaning time ≤5 minutes), each block containing continuous grid areas (e.g., x=10~20, y=30~40); the block boundary is dynamically adjusted through the obstacle position in the global semantic map to avoid cutting obstacles.

[0070] Specifically, parallel operation includes: each robot receives the assigned block coordinates and path sequence, and starts cleaning from different entry points (e.g., the two ends of the block diagonal line); Use "arch-shaped" path to cover the block to ensure no omission. Specifically, dynamic obstacle avoidance includes: if a robot encounters a temporary obstacle (e.g., a moving pedestrian), it immediately reports to the main controller, and the controller dynamically adjusts the paths of other robots to avoid repeated cleaning; Each robot synchronizes progress in real time through a Wi-Fi Direct network to ensure seamless connection at adjacent block boundaries.

[0071] In this embodiment, the leader-following method includes: leader robot path planning, follower robot coordination strategy, and dynamic adjustment mechanism.

[0072] Specifically, leader robot path planning includes: the leader robot moves along the preset path (e.g., the center line of the corridor) to perform the main cleaning task (e.g., floor washing); avoid high dynamic obstacles (e.g., areas with high pedestrian density) when planning the path.

[0073] Specifically, the follower robot coordination strategy includes: The following robots maintain a fixed distance (e.g. 0.5 meters) from the leading robot, forming a snake-shaped formation, covering the area on both sides of the leading path; the leading robot cleans the main path, and the following robots handle the edge area (e.g. corners, furniture bottom).

[0074] Specifically, the dynamic adjustment mechanism includes: if the leading robot is temporarily suspended due to failure, the following robot automatically switches to autonomous mode and takes over the unfinished task; the main controller monitors the formation state in real time and dynamically adjusts the formation parameters (such as distance, speed).

[0075] In this embodiment, the cleaning effect is verified after the task is completed through a visual sensor, including: the robot collects the image of the cleaned ground through a visual sensor (RGB camera), with a resolution of 1920x1080 and a frame rate of 30fps; Compare with the baseline image before cleaning (stored locally or in the controller).

[0076] Divide the images before and after cleaning into 8x8 pixel blocks and calculate the brightness, contrast, and structural similarity; The comprehensive similarity threshold is set to 90%, and if it is lower than the threshold, it is determined as an unqualified area; Use a pre-trained CNN model (such as ResNet-18) to detect residual stains (such as oil stains, dust clumps), and output the confidence (threshold ≥ 0.8); The robot reports the detection results (including unqualified area coordinates, pollution level) to the main controller through the Wi-Fi Direct network; The controller marks the unqualified area as "second cleaning" in the global semantic map.

[0077] Embodiment 2 Please refer to the attached Figure 3 This embodiment provides a multi-cleaning robot formation cleaning system based on perception interaction, including the following modules: Module A: autonomous perception and environment modeling module, configured in each robot, including multi-sensor fusion unit, target detection unit and semantic segmentation unit, used for real-time collection of environment data and construction of local grid map; Module B: dynamic identification and networking module, configured in each robot, including near field communication unit and main controller election unit, used for detecting other robots and establishing ad hoc network, and electing formation controller; Module C: global information fusion and task planning module, configured in the formation controller, including map stitching unit and task allocation unit, used for integrating global semantic map and dividing mutually exclusive subtasks; Module D: formation execution and verification module, configured in each robot and controller, including block cleaning unit, leader-follower unit and visual verification unit, used for executing cleaning instructions and feeding back cleaning results.

[0078] In one specific embodiment of the present embodiment, the multi-sensor fusion unit in module A includes vision sensors, lidar, IMU and GPS, the target detection unit adopts YOLO series algorithm, and the semantic segmentation unit determines the cleaning area priority by pollution level. 7. The multi-cleaner robot formation cleaning method based on a wireless communication network according to claim 1, characterized in that: In one specific embodiment of the present embodiment, the near field communication unit in module B supports the Wi-Fi Direct protocol, and the master controller election unit realizes automatic upgrading of the controller based on the pre-allocated internal number priority.

[0079] In one specific embodiment of the present embodiment, the task allocation unit in module C dynamically allocates the washing, dusting or disinfecting sub-tasks according to the robot function type, and optimizes the path planning based on the work time constant division algorithm.

[0080] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for multi-cleaning robot formation cleaning based on a wireless communication network, the method comprising: Comprising the following steps: S1. Single robot autonomous perception and environment modeling: Each robot collects environmental data in real time through a multi-sensor fusion system, identifies obstacle avoidance areas, cleaning areas and special attention objects using target detection algorithms and semantic segmentation algorithms, and constructs a local grid map through a SLAM algorithm; S2. Robot dynamic identification and adaptive networking: After detecting other collaborative robots, the robot triggers a near field communication protocol, establishes a Wi-Fi Direct ad hoc network, and elects the robot with the smallest number as the formation controller based on pre-assigned internal numbers; S3. Global information fusion and task planning: The formation controller integrates the local maps and perception information uploaded by each robot to generate a global semantic map, and divides the cleaning area into mutually exclusive sub-tasks based on the work time constant division algorithm and assigns them to each robot; S4. Formation cooperative execution and closed-loop verification: The controller issues cleaning instructions, robots with the same function execute tasks using the block method, and robots with complementary functions use the leader-following method, and after the task is completed, the cleaning effect is verified by a visual sensor, and the area that does not meet the standard triggers secondary cleaning. 2.The method of claim 1, wherein: In the step S1, the multi-sensor fusion system includes a vision sensor, a laser radar, an IMU and a GPS, the target detection algorithm is a YOLO series algorithm, and the semantic segmentation algorithm is used to distinguish the geometric range and pollution level of the obstacle avoidance area, the cleaning area and the special attention object. 3.The method of claim 2, wherein: In the step S2, the near field communication protocol includes a collision avoidance strategy and a communication establishment strategy, the collision avoidance strategy adjusts the motion trajectory in real time through the speed obstacle method, and the communication establishment strategy realizes the exchange of global positioning information and task state between robots through Wi-Fi Direct in a weak or no network environment.

4. The method of claim 3, wherein the method further comprises: In the step S3, the global semantic map is generated by splicing local maps through a space-time alignment algorithm and a graph optimization method, and the task allocation is divided into washing, dust collection or disinfection mutually exclusive sub-tasks based on the function type of the robot.

5. A multi-cleaning robot formation cleaning system based on perception interaction, characterized in that, Comprising the following modules: Module A: autonomous perception and environment modeling module, configured in each robot, including a multi-sensor fusion unit, a target detection unit and a semantic segmentation unit, for collecting environmental data in real time and constructing a local grid map; Module B: dynamic identification and networking module, configured in each robot, including a near field communication unit and a master controller election unit, for detecting other robots and establishing an ad hoc network, and electing a formation controller; Module C: global information fusion and task planning module, configured in the formation controller, including a map splicing unit and a task allocation unit, for integrating a global semantic map and dividing mutually exclusive sub-tasks; Module D: formation execution and verification module, configured in each robot and the controller, including a block cleaning unit, a leader-following unit and a visual verification unit, for executing cleaning instructions and feeding back cleaning results.

6. The system of claim 5, wherein, The multi-sensor fusion unit in module A includes a vision sensor, a laser radar, an IMU and a GPS, the target detection unit uses a YOLO series algorithm, and the semantic segmentation unit determines the cleaning area priority by pollution level. 7.The method of claim 1, wherein: The.

8. The system of claim 5, wherein, The near field communication unit in the module B supports the Wi-Fi Direct protocol, and the master controller election unit realizes automatic upgrading of the controller based on pre-allocated internal number priority.

9. The system of claim 5, wherein, The task allocation unit in the module C dynamically allocates the washing, dusting or disinfecting sub-tasks according to the function type of the robot, and optimizes the path planning based on the work time constant division algorithm.

Images