Dynamic routing inspection path planning method, medium and equipment

Abstract

The invention discloses a polling path dynamic planning method, medium and equipment, and the method comprises the steps: obtaining environment information in real time, dynamically switching an RTK-GPS mode, a visual inertial navigation mode and a laser SLAM positioning mode, generating the current positioning information of a robot in combination with a global map, and constructing a local environment map; and planning an adaptive second preset inspection path based on the first preset inspection path. Environmental feature information is actively collected in an outdoor environment, a real-time obstacle avoidance path is generated by dynamically updating a local environment map, and the real-time obstacle avoidance path is fused to a second preset inspection path, so that smooth switching and dynamic obstacle avoidance cooperation of a positioning mode in an indoor and outdoor scene are realized; the problem of planning failure caused by multiple positioning source conflicts, local path oscillation and environment sudden change in a complex scene is solved, and the continuity of the inspection process and the path planning robustness are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of robot navigation, and in particular to a dynamic planning method, medium and device for inspection paths. Background Art

[0002] Existing inspection robots face dual challenges of positioning robustness and dynamic obstacle avoidance coordination in complex scenarios. In the indoor-outdoor transition area, navigation drift is likely to occur due to the lag in positioning mode switching. At present, although multi-source heterogeneous sensors have been integrated, problems such as a sharp increase in the cumulative error of visual odometry and laser point cloud feature mismatch still occur due to sudden changes in environmental light or unstructured terrain, and it is difficult to balance the real-time performance and confidence of multi-modal data fusion. At the communication level, in an occluded environment, instruction adhesion is likely to occur due to protocol redundancy in a conventional system, resulting in a lag in the emergency obstacle avoidance response. In addition, the existing path planning algorithm lacks the intention prediction of dynamic obstacles, and it is difficult to construct continuous spatio-temporal associations for multi-sensor cross-modal data, resulting in frequent oscillations of local paths and affecting the inspection efficiency and safety. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to propose a dynamic planning method, medium and device for inspection paths, so as to solve the problem of unstable inspection path planning caused by inaccurate switching of multi-positioning modes, insufficient cross-modal obstacle avoidance coordination, and communication delay in a dynamic environment.

[0004] To achieve the above technical objectives, in a first aspect, the present application provides a dynamic planning method for inspection paths, including:

[0005] Obtain environmental information, and select a positioning mode according to the environmental information. The positioning modes include RTK-GPS mode, visual inertial navigation mode, and laser SLAM positioning mode, and the environmental information includes indoor environment and outdoor environment;

[0006] Obtain global map information, and generate positioning information of the current inspection robot according to the positioning mode;

[0007] Generate a local environmental map within a preset range according to the positioning information and the global map information;

[0008] Moreover, obtain a first preset inspection path associated with the global map information, and generate a second preset inspection path according to the first preset inspection path and the local environmental map;

[0009] If the environmental information is an outdoor environment, the method further includes:

[0010] Collect environmental feature information in the outdoor environment, and extract environmental features;

[0011] Update the local environmental map according to the environmental features, and generate a local obstacle avoidance path, and update the local obstacle avoidance path to the second preset inspection path.

[0012] In some embodiments, if the environmental information is an indoor environment, the positioning mode is a visual inertial navigation mode or a laser SLAM positioning mode;

[0013] If the environmental information is an outdoor environment, the positioning mode is an RTK-GPS mode.

[0014] In some embodiments, the RTK-GPS mode is configured as follows:

[0015] Receive the carrier phase observations of the satellite navigation system through the RTK-GPS module, and perform real-time kinematic differential positioning with a ground reference station to obtain the positioning data of the RTK-GPS module. The positioning data includes longitude and latitude coordinates;

[0016] When the RTK-GPS module outputs a fixed solution state and the horizontal positioning accuracy is set to a preset accuracy threshold, use the positioning data as the positioning information of the inspection robot in the current environmental information and output it;

[0017] When the RTK-GPS module outputs a float solution or a single-point positioning state, automatically switch to the visual inertial navigation mode or the laser SLAM positioning mode;

[0018] The visual inertial navigation mode is configured as follows:

[0019] Collect the environmental color image and depth information through an RGB-D depth camera;

[0020] Combine the environmental color image and depth information with the angular velocity and acceleration data output by the inertial measurement unit, and calculate the motion state of the current inspection robot through visual odometry;

[0021] Convert the motion state into the positioning information of the inspection robot in the current environmental information and output it;

[0022] The laser SLAM positioning mode is configured as follows:

[0023] Obtain the environmental point cloud data by scanning with a 3D lidar;

[0024] Perform voxel filtering and noise reduction processing on the environmental point cloud data;

[0025] Based on the filtered environmental point cloud data, calculate the matching relationship between the environmental point cloud data and the global point cloud map through the iterative closest point algorithm, and output the matching result;

[0026] Convert the matching result into the positioning information of the inspection robot in the current environmental information and output it.

[0027] In some embodiments, obtaining the global map information and generating the positioning information of the current inspection robot according to the positioning mode includes:

[0028] When the RTK-GPS mode is adopted, the latitude and longitude coordinates output by the RTK-GPS module are converted into position coordinates in the coordinate system corresponding to the global map information, which is recorded as the initial positioning information;

[0029] When the visual inertial navigation mode is adopted, the motion state output by the visual odometry algorithm is converted into the initial positioning information corresponding to the global map information;

[0030] When the laser SLAM positioning mode is adopted, the matching result output by the iterative closest point algorithm is converted into the initial positioning information corresponding to the global map information;

[0031] The matching degree between the initial positioning information and the global map information shown above is calculated in real time;

[0032] When the matching degree is lower than the preset matching threshold, trigger the re-initialization of the positioning mode and update the initial positioning information, recalculate the matching degree between the updated initial positioning information and the global map information until the matching degree meets the preset matching threshold, and record the updated initial positioning information as the final positioning information;

[0033] Output the final positioning information, which includes three-dimensional position coordinates and heading angle.

[0034] In some embodiments, generating the second preset inspection path according to the first preset inspection path and the local environment map includes:

[0035] Read the path point sequence of the first preset inspection path from the global map information, and convert the path point sequence into the coordinate system corresponding to the local environment map, which is recorded as the first path point sequence;

[0036] Perform density-based spatial clustering processing on the point cloud data in the local environment map to identify discrete point cloud clusters;

[0037] Calculate the minimum bounding cube for each point cloud cluster to obtain the three-dimensional position information and size information of the obstacle;

[0038] Mark the three-dimensional position information and size information of the obstacle in the local environment map;

[0039] When an obstacle is detected on the first path point sequence, it is recorded as a marked obstacle, and a local obstacle avoidance path is generated using the dynamic window method, including:

[0040] Taking the current motion state of the inspection robot as the initial condition, sampling multiple groups of velocity combinations in the velocity space, and the velocity combination includes linear velocity and angular velocity;

[0041] Generate the motion trajectory within a preset time period for each group of velocity combinations;

[0042] Evaluate the collision risk and execution efficiency of each motion trajectory with the marked obstacles;

[0043] Select the motion trajectory with zero collision risk and optimal efficiency as the local obstacle avoidance path;

[0044] Convert the local obstacle avoidance path into a second sequence of path points and splice it with the obstacle-free segment of the first sequence of path points;

[0045] Generate a continuous executable third sequence of path points;

[0046] Connect the third sequence of path points to form a second preset inspection path.

[0047] In some embodiments, the environmental feature information includes environmental color images and point cloud data. The environmental feature information in the outdoor environment is collected, and the environmental features extracted include:

[0048] Execute a deep learning-based object detection algorithm on the environmental color image to identify environmental objects under a preset category, denoted as the object detection result;

[0049] Extract plane features from the point cloud data to obtain geometric features, and the geometric features include the geometric information of the ground and the geometric information of the obstacles;

[0050] Spatially align and fuse the object detection result with the geometric features to obtain environmental features.

[0051] In some embodiments, updating the local environmental map according to the environmental features includes:

[0052] Associate and store the environmental features with the positioning data;

[0053] And establish a spatio-temporal index of the environmental features based on the timestamp and the positioning data;

[0054] When the environmental features at the same position are repeatedly detected, update the confidence score of the environmental features at the same position;

[0055] And output the environmental features extracted at the current moment to the local environmental map.

[0056] In some embodiments, generating a local obstacle avoidance path according to the environmental features and updating the local obstacle avoidance path to the second preset inspection path includes:

[0057] Identify local obstacle information in the current traveling direction according to the environmental features, and the local obstacle information includes the geometric features and confidence scores of the local obstacles in the current local environmental map;

[0058] Calculate the avoidance priorities of the respective local obstacles according to the local obstacle information;

[0059] Use the improved RRT* algorithm to generate multiple candidate obstacle avoidance paths around local obstacles;

[0060] Conduct a safety assessment on the candidate obstacle avoidance paths, and exclude the candidate obstacle avoidance paths with a distance less than the preset safety threshold from the local obstacles;

[0061] Conduct a smoothness assessment on the remaining candidate obstacle avoidance paths, and select the candidate obstacle avoidance path with the smallest curvature change, denoted as the local obstacle avoidance path;

[0062] Extract the continuous path segments in the second preset inspection path that are not affected by local obstacles;

[0063] Smoothly connect the local obstacle avoidance path and the continuous path segments with a B-spline curve to obtain a fused obstacle avoidance path;

[0064] Perform equidistant sampling on the fused obstacle avoidance path to generate an updated second preset inspection path.

[0065] In a second aspect, the present invention also provides a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method in the first aspect is implemented.

[0066] In a third aspect, the present invention also provides an electronic device, including a memory and a processor, where the memory is used to store one or more computer program instructions, and one or more computer program instructions are executed by the processor to implement the method in the first aspect.

[0067] Adopting the above technical solutions, compared with the prior art, the present invention has the following beneficial effects:

[0068] The above technical solutions dynamically switch the RTK-GPS mode, visual inertial navigation mode, and laser SLAM positioning mode by real-time obtaining environmental information, generate the current positioning information of the robot in combination with the global map and construct a local environmental map, and plan an adaptable second preset inspection path based on the first preset inspection path. Actively collect environmental feature information in the outdoor environment, generate a real-time obstacle avoidance path by dynamically updating the local environmental map, and fuse it into the second preset inspection path to achieve smooth switching of the positioning mode and coordinated dynamic obstacle avoidance in indoor and outdoor scenarios, solve the problems of multiple positioning source conflicts, local path oscillation, and planning failure caused by environmental mutations in complex scenarios, and ensure the continuity of the inspection process and the robustness of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0070] Figure 1 It is a method step diagram of steps S101 to S105 of the planning method described in the specific implementation manner;

[0071] Figure 2 It is a schematic structural diagram of the electronic device described in the specific implementation manner.

[0072] 1. Electronic device;

[0073] 11. Memory;

[0074] 12. Processor. Specific implementation manner

[0075] The following will further describe the present invention in detail in conjunction with the accompanying drawings and embodiments. It should be specifically noted that the following embodiments are only used to illustrate the present invention, but do not limit the scope of the present invention. Similarly, the following embodiments are only some embodiments of the present invention rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0076] Please refer to Figure 1 , in the first aspect, this embodiment provides a dynamic planning method for inspection paths, including:

[0077] S101. Obtain environmental information, select a positioning mode according to the environmental information. The positioning modes include RTK-GPS mode, visual inertial navigation mode, and laser SLAM positioning mode. The environmental information includes indoor environment and outdoor environment;

[0078] S102. Obtain global map information and generate positioning information of the current inspection robot according to the positioning mode;

[0079] S103. Generate a local environment map within a preset range according to the positioning information and the global map information;

[0080] And obtain a first preset inspection path associated with the global map information, and generate a second preset inspection path according to the first preset inspection path and the local environment map;

[0081] If the environmental information is the outdoor environment, the method further includes:

[0082] S104. Collect environmental feature information in the outdoor environment and extract environmental features;

[0083] S105. Update the local environmental map according to the environmental features, generate a local obstacle avoidance path, and update the local obstacle avoidance path to the second preset inspection path.

[0084] In this embodiment, the system uses Pixhawk V6X as the chassis controller, which is responsible for motor drive (outputting PWM signals through the M1 / M3 interface), sensor data fusion (IMU, RTK), and execution of low-level control instructions; uses C-RTK9Ps as the high-precision RTK module to provide centimeter-level positioning outdoors (NMEA-0183 protocol, output frequency 10Hz); uses Jetson Orin Nano as the airborne computing platform to run Ubuntu20.04 + ROS Noetic to process algorithm fusion and decision-making tasks; the dual-drive motor controller receives PWM signals (1000 - 2000μs) through the M1 / M3 interface of Pixhawk to drive the tracked differential steering; uses the H16 remote controller to provide video transmission, data transmission, and communication links; establishes a two-way data transmission channel with the airborne computer and the flight controller through a laptop: remotely connects the C-RTK9Ps mobile terminal and the ground station through the remote controller data transmission, and realizes information interaction between the C-RTK9Ps mobile terminal and the C-RTK9Ps base station through QGC to form an RTK positioning network; directly connects the remote controller to the flight controller to send the real-time dynamics of the flight controller to the remote controller terminal, and connects the network port of the remote controller receiver to Jetson Orin Nano to create a network passthrough channel between the laptop and the airborne computer to realize remote control of the airborne computer. The serial communication used is to connect Pixhawk and Jetson through the Telem2 port, with a baud rate of 921600 and MAVLink protocol to transmit control instructions; the network communication is to achieve low-latency data transmission between ROS nodes through gigabit Ethernet.

[0085] In step S101, preferably, when the environmental information is an indoor environment, the positioning mode is mainly based on the visual inertial navigation mode and the laser SLAM positioning mode; when the environmental information is an outdoor environment, the positioning mode is mainly based on the RTK-GPS mode. Preferably, the sensor system uses Intel RealSense D435i, MID360 lidar, and IMU. Intel RealSense D435i collects data through an RGB-D camera to achieve visual inertial navigation and YOLOv8 target detection (real-time frame rate 30FPS); the laser SLAM positioning mode uses MID360 lidar to collect data through 3D LiDAR, with a scanning frequency of 20Hz, which is used for the FAST-LIO2 algorithm to construct an environmental point cloud map and provide odom odometer information and IMU information;

[0086] The RTK-GPS mode adopts an IMU. By means of the dual IMUs and magnetic compass built into the Pixhawk V6X, the RTK data is fused to improve the positioning robustness.

[0087] Furthermore, seamless positioning switching is achieved through dynamic weight allocation. Specifically, when the RTK signal quality deteriorates, the weight coefficients of VINS and FAST-LIO are automatically increased to ensure positioning continuity.

[0088] In step S102, the global map information can be constructed by fusing multi-source surveying and mapping data with the historical inspection path, and combining the preset environmental model and real-time sensor calibration to generate the global topological structure, ensuring dynamic adaptation to the spatial coordinate system of the current positioning mode.

[0089] Preferably, the RTK-GPS mode fuses the RTK position and IMU data through an EKF2 filter to output the high-precision pose in the global coordinate system; the visual inertial navigation mode performs positioning through VINS-Fusion, tightly couples the RGB images of the D435i with the IMU data, and outputs the 6DoF pose in the local coordinate system, and: uses the Kalibr tool to calibrate the external parameters of the camera-IMU, and the reprojection error < 0.15 pixels; the laser SLAM positioning mode processes the point cloud through FAST-LIO2 SLAM, and the MID360 point cloud (100,000 points / frame) is denoised through voxel filtering (resolution 0.1 m), and a 3D grid map (resolution 0.2 m) is constructed in real time to achieve a positioning error < 5 cm in the indoor environment.

[0090] In step S103, preferably, the acquisition of the first preset inspection path can be understood as: through the A* algorithm, an initial path is generated based on the 3D grid map (i.e., the global map information), which is the first preset inspection path, and the cost function is distance + height change; perform RRT* optimization to generate a smooth path in a complex environment, with a maximum curvature limit of 0.3 m-1; concatenate the position coordinates collected by RTK to form a global path, with the current position of the inspection robot as the 0th position, and move straight to the first position, and execute the instructions of the local path planner during the movement.

[0091] For the second preset inspection path, common DWA and TEB local path planners are deployed for the local path planner, and at the same time, the self-developed QZN-Planner is configured for local path planning. The self-developed local path planner has a fast start, occupies less running resources, and can flexibly insert specified actions.

[0092] In step S104, preferably, the environmental feature information includes the image information and point cloud data information of the current outdoor environment.

[0093] In step S105, the generation of the local obstacle avoidance path can be achieved through lidar obstacle avoidance and visual obstacle avoidance: Lidar obstacle avoidance includes point cloud clustering and the dynamic window method. Point cloud clustering uses the DBSCAN algorithm to detect obstacles and extract bounding boxes. The dynamic window method (DWA) generates candidate trajectories based on the current speed and selects the path with the minimum collision risk. Visual obstacle avoidance includes VINS-Fusion and semantic map update. VINS-Fusion converts depth information into obstacle information, and obstacles within 2 meters ahead are updated to RVIZ, providing obstacle inflation information for the local path planner. The semantic map update maps the detection results to the global map and marks them as temporary obstacles. The lidar and visual data are fused through Kalman filtering to improve the confidence of obstacle detection (for example, when the overlap rate between the lidar point cloud and the YOLO detection frame is >70%, it is confirmed as a real obstacle).

[0094] In this embodiment, the environmental perception module dynamically selects the RTK-GPS mode, the visual inertial navigation mode, or the lidar SLAM positioning mode, generates robot positioning information in combination with the global map, constructs a local environment map, and generates an adapted second preset inspection path based on the first preset inspection path; in an outdoor scene, it fuses environmental features to update the local map and generate an obstacle avoidance path, realizing dynamic path optimization. It adopts multi-source positioning fusion (RTK-GPS, VINS-Fusion, FAST-LIO2) and a dynamic weight allocation mechanism, and through the EKF2 filter, point cloud clustering, and semantic map update technology, it ensures the smooth switching between visual / laser positioning when the RTK signal weakens; it generates the first preset inspection path based on the A* and RRT* algorithms, combines DWA, TEB, and QZN-Planner to achieve fast local planning to obtain the second preset inspection path, and improves the obstacle avoidance confidence by fusing the lidar point cloud and the YOLOv8 detection frame through Kalman filtering. This embodiment can achieve the dynamic and accurate switching of multiple positioning modes, reduce indoor positioning drift through the millimeter-level calibration of VINS-Fusion and FAST-LIO2, realize real-time avoidance of complex obstacles by combining the dynamic window method and semantic map update, utilize the lightweight characteristics of QZN-Planner to ensure the planning efficiency, and ensure the high precision, low latency, and strong robustness of the path planning of the inspection robot in an indoor-outdoor hybrid scene.

[0095] In some embodiments, if the environmental information is an indoor environment, the positioning mode is the visual inertial navigation mode or the lidar SLAM positioning mode;

[0096] If the environmental information is an outdoor environment, the positioning mode is the RTK-GPS mode.

[0097] In this embodiment, the selection of the positioning mode is based on environmental characteristics. The indoor environment relies on local sensors to resist signal shielding, and the outdoor environment relies on satellite positioning to cover the entire area.

[0098] The visual-inertial navigation mode realizes 6DoF pose estimation without GPS through the tight coupling of an RGB-D camera and an IMU (VINS-Fusion algorithm), and is suitable for scenes with rich textures; the laser SLAM positioning mode relies on lidar point clouds (FAST-LIO2 algorithm) to construct a 3D grid map and adapts to low-light or dynamic environments. The RTK-GPS mode is adopted in outdoor environments, and centimeter-level global positioning is provided through the fusion of an RTK module and an IMU (EKF2 filter) to ensure accuracy in open areas. The dynamic weight allocation mechanism automatically switches the dominant positioning source according to the RTK signal quality to avoid path deviation caused by a single failure.

[0099] In this embodiment, through the multi-modal positioning fusion and dynamic switching mechanism, the positioning robustness of indoor and outdoor scenes is considered. The complementarity between vision and laser SLAM improves the adaptability to complex environments, and the cooperation between RTK and IMU enhances the stability of outdoor positioning, ensuring seamless connection of the inspection path planning in the event of environmental mutations and reducing the interference of positioning drift on the global path.

[0100] In some embodiments, the RTK-GPS mode is configured as:

[0101] Receiving carrier phase observations of the satellite navigation system through an RTK-GPS module, and performing real-time kinematic differential positioning with a ground reference station to obtain positioning data of the RTK-GPS module, where the positioning data includes longitude and latitude coordinates;

[0102] When the RTK-GPS module outputs a fixed solution state and the horizontal positioning accuracy is set to a preset accuracy threshold, using the positioning data as the positioning information of the inspection robot in the current environmental information and outputting it;

[0103] When the RTK-GPS module outputs a floating solution or single-point positioning state, automatically switching to the visual-inertial navigation mode or the laser SLAM positioning mode;

[0104] The visual-inertial navigation mode is configured as:

[0105] Collecting environmental color images and depth information through an RGB-D depth camera;

[0106] Combining the environmental color images and depth information with the angular velocity and acceleration data output by the inertial measurement unit, and calculating the motion state of the current inspection robot through visual odometry;

[0107] Converting the motion state into the positioning information of the inspection robot in the current environmental information and outputting it;

[0108] The laser SLAM positioning mode is configured as:

[0109] Obtaining environmental point cloud data through three-dimensional lidar scanning;

[0110] Perform voxel filtering and noise reduction on the environmental point cloud data;

[0111] Based on the filtered environmental point cloud data, calculate the matching relationship between the environmental point cloud data and the global point cloud map through the Iterative Closest Point (ICP) algorithm, and output the matching result;

[0112] Convert the matching result into the positioning information of the inspection robot in the current environmental information and output it.

[0113] In this embodiment, the RTK-GPS mode receives satellite carrier phase observations through the RTK-GPS module and performs real-time kinematic differential positioning with the ground reference station. The ground reference station sends correction data to eliminate errors such as ionospheric delay and satellite clock error, enabling the mobile device to obtain centimeter-level accuracy longitude and latitude coordinates. When the module outputs a fixed solution state (i.e., the carrier phase integer ambiguity is accurately solved) and the horizontal positioning accuracy reaches the preset accuracy threshold, it indicates that the positioning result is reliable and can be directly used as the global positioning information of the inspection robot. Preferably, the preset accuracy threshold is set according to the positioning requirements of the specific application scenario, usually referring to the nominal accuracy index of the device and calibrated through actual environmental tests. For example, in the inspection task, it is necessary to meet the path tracking error tolerance, and the critical value for triggering the switch is determined by comprehensively analyzing the stability of historical positioning data. If a float solution is output (indicating that the carrier phase integer ambiguity is not fully fixed) or a single-point positioning state (relying only on a single satellite without differential correction), both will cause a significant increase in positioning error, triggering the mode switching mechanism to avoid cumulative errors affecting path planning.

[0114] The visual inertial navigation mode synchronously acquires the environmental color image and depth information through the RGB-D depth camera, combines the angular velocity and acceleration data of the Inertial Measurement Unit (IMU), and uses the visual odometry algorithm (such as VINS-Fusion) for tightly coupled optimization: feature points are extracted from the RGB image, the depth information provides scale constraints, and the IMU data compensates for motion blur. The 6-degree-of-freedom pose is solved through non-linear optimization, which is applicable to indoor scenarios without GPS signals and rich textures.

[0115] The laser SLAM positioning mode uses a 3D lidar to scan the environmental point cloud data, and removes random noise and dynamic interference through voxel filtering and noise reduction processing (dividing the point cloud into uniform grids and retaining the center points); based on the filtered point cloud data, the Iterative Closest Point (ICP) algorithm performs rigid transformation solution (rotation matrix and translation vector) between the current frame of point cloud and the pre-constructed global point cloud map through frame-by-frame registration, and outputs high-precision local positioning information, which is especially applicable to low-light or low-texture environments.

[0116] This embodiment realizes positioning robustness through a mode grading and dynamic switching mechanism: RTK-GPS provides global absolute coordinates but depends on satellite signal quality. The vision and laser modes provide local relative positioning when the signal is limited, and cover complex scenarios through sensor complementarity (vision resists dynamic interference and laser resists illumination changes). The multi-source fusion of positioning information ensures that the inspection robot can still maintain continuous pose output in indoor-outdoor transition areas or occluded environments, providing a stable state input for path planning.

[0117] In some embodiments, obtaining global map information and generating positioning information of the current inspection robot according to the positioning mode includes:

[0118] When the RTK-GPS mode is adopted, convert the latitude and longitude coordinates output by the RTK-GPS module into position coordinates in the coordinate system corresponding to the global map information, denoted as the initial positioning information;

[0119] When the visual inertial navigation mode is adopted, convert the motion state output by the visual odometry algorithm into the initial positioning information corresponding to the global map information;

[0120] When the laser SLAM positioning mode is adopted, convert the matching result output by the iterative closest point algorithm into the initial positioning information corresponding to the global map information;

[0121] Real-time calculate the matching degree between the above-mentioned initial positioning information and the global map information;

[0122] When the matching degree is lower than the preset matching threshold, trigger the re-initialization of the positioning mode and update the initial positioning information, and re-calculate the matching degree between the updated initial positioning information and the global map information until the matching degree meets the preset matching threshold, and record the updated initial positioning information as the final positioning information;

[0123] Output the final positioning information, and the final positioning information includes three-dimensional position coordinates and heading angle.

[0124] In this embodiment, the RTK-GPS mode obtains latitude and longitude coordinates by receiving differential signals from satellites and ground reference stations, and maps them to the standardized spatial reference system of the global map through coordinate conversion to form initial positioning information including planar position and elevation; the visual inertial navigation mode depends on the visual odometry algorithm to track feature points in the color image sequence collected by the depth camera, combines the motion parameters of the inertial measurement unit to solve the pose transformation, and projects it onto the global map through coordinate system alignment; the laser SLAM positioning mode is based on the iterative closest point algorithm to perform frame-by-frame registration on the point cloud data scanned by the current lidar and the reference point cloud in the global map, and realizes the precise association between the local coordinate system and the global map by optimizing the rotation and translation matrix.

[0125] The matching degree evaluation is achieved by quantifying the spatial consistency between the initial positioning information and the global map. For example, in laser SLAM, the proportion of the overlapping area between the current point cloud and the global point cloud is calculated; in visual navigation, the projection deviation between the feature points and the map markers is compared; in RTK mode, the coincidence degree between the transformed coordinates and the map topological structure is examined, and the matching degree is comprehensively calculated through geometric relationships (such as point position offset, angle deviation). When the matching degree is lower than the preset matching threshold, it indicates that there is a significant conflict between the positioning data and the global map, triggering the re-initialization of the positioning mode to correct the sensor parameters or pose estimation, and making the updated positioning information meet the map constraint conditions through iterative optimization. The preset matching threshold can be set according to the system's tolerance range for positioning errors, by analyzing the typical deviation distribution of historical operation data, combining the minimum pose accuracy requirements of the path planning, and determining it after repeatedly debugging and balancing the false alarm rate and stability in the laboratory environment by simulating the positioning drift scenario.

[0126] The finally output positioning information needs to include three-dimensional position coordinates and heading angle, ensuring that the inspection path planning has a dual benchmark of spatial position and attitude, and supporting the collaborative decision-making of the global path and local obstacle avoidance.

[0127] In this embodiment, through the collaborative application of multi-modal positioning technologies, the positioning robustness and accuracy of the inspection robot in complex environments are effectively improved. The RTK-GPS mode realizes wide-area high-precision geographical coordinate mapping with the help of differential signals. The visual inertial navigation mode fuses visual feature tracking and inertial motion parameter calculation of local poses. The laser SLAM mode relies on point cloud registration to establish precise spatial associations. The three complement each other to overcome the limitations of single sensors. The matching degree evaluation dynamically verifies the positioning reliability based on geometric consistency, and the preset threshold balances the system's fault tolerance and stability, ensuring that the positioning information is iteratively optimized through re-initialization in abnormal states. The finally output three-dimensional position coordinates and heading angle provide a spatial benchmark for the global path planning, and at the same time support the real-time attitude adjustment of local obstacle avoidance, forming a multi-scale collaborative decision-making ability, and ensuring the continuous and reliable execution of the inspection task under satellite signal occlusion, dynamic light changes or complex terrain interference.

[0128] In some embodiments, generating the second preset inspection path according to the first preset inspection path and the local environment map includes:

[0129] Read the path point sequence of the first preset inspection path from the global map information, and convert the path point sequence to the coordinate system corresponding to the local environment map, denoted as the first path point sequence;

[0130] Perform density-based spatial clustering processing on the point cloud data in the local environment map to identify discrete point cloud clusters;

[0131] Calculate the minimum bounding cube for each point cloud cluster to obtain the three-dimensional position information and size information of the obstacles;

[0132] Mark the three-dimensional position information and size information of the obstacle in the local environment map;

[0133] When an obstacle is detected on the first path point sequence, it is recorded as a marked obstacle, and a local obstacle avoidance path is generated using the dynamic window method, including:

[0134] Taking the current motion state of the inspection robot as the initial condition, sample multiple groups of velocity combinations in the velocity space, where the velocity combination includes linear velocity and angular velocity;

[0135] Generate a motion trajectory within a preset time period for each group of velocity combinations;

[0136] Evaluate the collision risk and execution efficiency of each motion trajectory with the marked obstacle;

[0137] Select the motion trajectory with zero collision risk and optimal efficiency as the local obstacle avoidance path;

[0138] Convert the local obstacle avoidance path into a second path point sequence and splice it with the obstacle-free section of the first path point sequence;

[0139] Generate a continuous executable third path point sequence;

[0140] Connect the third path point sequence to form a second preset inspection path.

[0141] In this embodiment, after the path point sequence of the first preset inspection path is imported from the global map, it needs to be mapped to the independent coordinate reference frame of the local environment map through coordinate system conversion to form a first path point sequence aligned with the real-time sensor data space.

[0142] Density-based spatial clustering processing divides the local environment point cloud collected by the lidar according to the point cloud distribution density, and merges adjacent point sets with density meeting the threshold into point cloud clusters to represent the spatial contour of discrete obstacles.

[0143] Minimum bounding cube calculation extracts the three-dimensional bounding box parameters for each point cloud cluster, including the center point coordinates, length, width, height dimensions, and attitude angles, so as to abstractly describe the physical occupancy information of the obstacle.

[0144] The core of the dynamic window method lies in constructing a velocity-trajectory search space: taking the current linear velocity, angular velocity and kinematic constraints of the robot as the boundaries, discretely sample multiple groups of linear velocity and angular velocity combinations within the feasible velocity range, and each group of velocities corresponds to a predicted motion trajectory generated within a preset time period, and the trajectory shape is deduced from the robot kinematic model.

[0145] The collision risk assessment is achieved by detecting the spatial intersection of the predicted trajectory and the marked obstacle cube, and the execution efficiency is comprehensively quantified based on the trajectory length, speed stability, and heading deviation. During the path stitching process, the second path point sequence of the obstacle avoidance path needs to be end-point matched and smoothly connected with the unaffected section of the first path point sequence, and finally a continuous third path point sequence including obstacle avoidance correction is generated. The third path point sequence is transformed into a second preset inspection path that meets the motion continuity through curve fitting methods such as cubic spline interpolation.

[0146] In this embodiment, through multi-modal data processing and dynamic path optimization, the autonomous obstacle avoidance ability and path planning robustness of the inspection robot in complex environments are effectively improved. The density-based spatial clustering processing accurately identifies the obstacle contours in the local environmental point cloud, and combines the minimum bounding cube calculation to abstract the discrete point cloud into a structured obstacle model, enhancing the real-time performance and interpretability of environmental perception. The dynamic window method generates a dynamic obstacle avoidance trajectory through velocity space sampling and kinematic model calculation, taking into account both collision risk avoidance and execution efficiency optimization to ensure the feasibility and smoothness of path correction. The path stitching mechanism uses end-point matching and curve fitting technologies to achieve seamless connection between the local obstacle avoidance path and the global path, ensuring the continuous executability of the third path point sequence. The generation process of the second preset inspection path integrates global planning constraints and local environmental dynamic changes, and forms a hierarchical decision-making architecture with both global goal orientation and local real-time response through multi-stage collaboration of coordinate transformation, obstacle modeling, trajectory evaluation, and path fusion, significantly improving the adaptability and task continuity of the inspection system in sudden obstacle scenarios.

[0147] In some embodiments, the environmental feature information includes environmental color images and point cloud data. Collecting the environmental feature information in the outdoor environment and extracting the environmental features includes:

[0148] Performing a deep learning-based object detection algorithm on the environmental color image to identify environmental objects under preset categories, denoted as the object detection result;

[0149] Performing plane feature extraction on the point cloud data to obtain geometric features, where the geometric features include the geometric information of the ground and the geometric information of the obstacles;

[0150] Spatially aligning and fusing the object detection result and the geometric features to obtain the environmental features.

[0151] In this embodiment, the environmental color image data represents the visual semantic attributes of the outdoor scene, and the point cloud data represents the three-dimensional geometric structure of the outdoor scene. The environmental color image is a two-dimensional pixel array in RGB format, captured by an optical sensor and used to record the texture, color, and lighting information of the scene; the point cloud data is generated by a lidar or a depth camera, containing a set of coordinates of discrete three-dimensional space points, reflecting the surface morphology and spatial distribution of the environment.

[0152] The object detection algorithm based on deep learning uses a convolutional neural network model to perform per-pixel analysis on environmental color images, identify and frame environmental targets under preset categories. Preferably, the preset categories may include enumerable semantic object categories such as vehicles, pedestrians, vegetation, etc. The object detection results are output as the bounding box coordinates, class labels, and confidence levels of each target.

[0153] Plane feature extraction performs surface fitting on point cloud data through the random sample consensus algorithm or the region growing method, and segments the ground plane equation parameters and the convex hull boundary of obstacles. The geometric information of the ground includes the plane height, normal vector, and undulation degree. The geometric information of the obstacles covers the outer contour size, spatial orientation, and occupied volume.

[0154] Spatial alignment unifies the two-dimensional pixel coordinate system of the object detection results and the three-dimensional world coordinate system of the point cloud to the same reference system through the sensor joint calibration parameters, realizing the precise matching of semantic labels and geometric boundaries. The fusion process superimposes the semantic attributes of the object detection results and the topological structure of the geometric features to form a multi-dimensional environmental feature expression with object categories, spatial positions, and contour descriptions, providing a composite environmental representation for subsequent path planning.

[0155] In this embodiment, through multi-modal data fusion and structured feature extraction, the comprehensiveness and accuracy of outdoor environment perception are significantly improved. The environmental color image and the point cloud data respectively capture the visual semantic attributes and three-dimensional geometric structures of the scene, forming complementary data sources. The object detection algorithm based on deep learning accurately identifies environmental targets in preset categories, endowing the system with semantic understanding ability; plane feature extraction segments the ground plane parameters and the convex hull boundary of obstacles from the point cloud data, establishing an accurate geometric topology model. The spatial alignment mechanism uses the sensor calibration parameters to achieve the precise registration of two-dimensional semantic labels and three-dimensional geometric boundaries, ensuring the physical consistency of semantic attributes and spatial information. The fusion process multi-dimensionally correlates the class labels, confidence levels of the object detection results with the outer contour size and occupied volume of the geometric features, constructing an environmental feature expression with object recognition, pose estimation, and spatial occupancy description. This composite environmental representation provides a comprehensive decision-making basis for path planning by integrating semantic passable area recognition and obstacle physical constraint analysis, effectively enhancing the environmental adaptability and decision-making reliability of the autonomous navigation system in complex outdoor scenes.

[0156] In some embodiments, updating the local environmental map according to environmental features includes:

[0157] Associatively storing the environmental features and the positioning data;

[0158] And establishing a spatio-temporal index of the environmental features based on the timestamp and the positioning data;

[0159] When the environmental features at the same position are repeatedly detected, update the confidence score of the environmental features at the same position;

[0160] And output the environmental features extracted at the current moment to the local environmental map.

[0161] In this embodiment, the process of the environmental features updating the local environmental map realizes the continuous optimization of the map data through multi-dimensional information association and dynamic confidence evaluation.

[0162] The positioning data refers to the real-time pose information obtained by the mobile carrier through the inertial measurement unit, the global positioning system or the visual odometer, including the position coordinates and the attitude angle parameters, and is used to bind the environmental features to the absolute space coordinate system.

[0163] The spatio-temporal index constructs a two-way retrieval structure of the environmental features in the time dimension and the space dimension based on the acquisition time sequence recorded by the timestamp and the spatial coordinate grid determined by the positioning data, and supports the rapid query of historical feature data according to the time interval or the geographical area.

[0164] The confidence score reflects the credibility of the environmental features when they are repeatedly observed at the same spatial position. The initial detection assigns a basic score, and the subsequent repeated detection improves the score value through probability accumulation or Bayesian update rules, and is used to distinguish stable environmental features from instantaneous interference noise.

[0165] The local environmental map is a dynamically updated rasterized or vectorized spatial database, which stores in real time the environmental features extracted at the current moment and their associated positioning data, spatio-temporal index and confidence score. When outputting, it preferentially retains the high-confidence features and eliminates the low-score entries. The update mechanism ensures the spatial continuity of the environmental features in the map through spatio-temporal consistency verification, eliminates the instantaneous abnormal data caused by sensor errors or dynamic obstacles, and maintains the geometric accuracy and semantic reliability of the local environmental map.

[0166] In this embodiment, the continuous optimization of the local environment map is achieved through multi-dimensional information association and dynamic confidence evaluation. Its process can be understood as follows: binding and storing environmental features with positioning data, constructing a spatio-temporal index, updating the confidence scores of duplicate detection features, and outputting high-confidence features to the map. The positioning data maps environmental features to the absolute space coordinate system through real-time pose information to ensure geographical space consistency. The spatio-temporal index establishes a two-way retrieval structure by combining timestamps and spatial coordinate grids to improve the query efficiency of historical feature data. The confidence scores are dynamically adjusted based on probability accumulation or Bayesian rules to effectively distinguish stable environmental features from instantaneous noise. The local environment map stores through a rasterization or vectorization mechanism, preferentially retains high-scoring features and eliminates low-confidence entries, and combines spatio-temporal consistency verification to eliminate sensor errors and dynamic interference. This method significantly improves the geometric accuracy and semantic reliability of map data, enhances the robustness of environmental representation by continuously fusing multi-period observation information, provides a real-time and stable spatial cognition basis for the autonomous navigation system, and ensures the accuracy and environmental adaptability of path planning decisions in dynamic scenarios.

[0167] In some embodiments, generating a local obstacle avoidance path according to environmental features and updating the local obstacle avoidance path to a second preset inspection path includes:

[0168] Identifying local obstacle information in the current traveling direction according to environmental features, where the local obstacle information includes the geometric features and confidence scores of local obstacles in the current local environment map;

[0169] Calculating the avoidance priorities of each local obstacle according to the local obstacle information;

[0170] Using an improved RRT* algorithm to generate multiple candidate obstacle avoidance paths around local obstacles;

[0171] Performing a safety assessment on the candidate obstacle avoidance paths, and excluding candidate obstacle avoidance paths with a distance less than a preset safety threshold from local obstacles;

[0172] Performing a smoothness assessment on the remaining candidate obstacle avoidance paths, and selecting the candidate obstacle avoidance path with the smallest curvature change, denoted as the local obstacle avoidance path;

[0173] Extracting continuous path segments in the second preset inspection path that are not affected by local obstacles;

[0174] Smoothingly connecting the local obstacle avoidance path and the continuous path segments with a B-spline curve to obtain a fused obstacle avoidance path;

[0175] Performing equidistant sampling on the fused obstacle avoidance path to generate an updated second preset inspection path.

[0176] In this embodiment, local obstacle avoidance path generation and path update achieve dynamic obstacle avoidance through multi-level evaluation and path optimization. Local obstacle information refers to the geometric features of obstacles and their confidence score extracted from the current local environment map. The geometric features include the outer contour size, occupied volume, and spatial distribution pattern of the obstacles. The confidence score is derived from the probability accumulation result of historical repeated detections and is used to represent the reliability of the existence of obstacles.

[0177] The avoidance priority is calculated by weighting the spatial occupancy range of the obstacle geometric features and the confidence score. Preferably, obstacles with a large occupied volume and high confidence obtain a higher priority. The improved RRT* algorithm improves the path search efficiency by introducing dynamic step size adjustment and heuristic sampling strategies, and generates candidate obstacle avoidance paths that satisfy kinematic constraints around the obstacles. The safety evaluation compares the minimum Euclidean distance between the candidate path and the obstacle with a preset safety threshold, which can be dynamically set according to the physical size and motion speed of the carrier to ensure that the path has no collision risk.

[0178] The smoothness evaluation quantifies the path continuity by calculating the change in the second derivative of the path curvature, and selects the path with the smallest curvature change to reduce the motion control complexity. The continuous path segment is a sequence of adjacent nodes in the second preset inspection path that is not affected by the current obstacle, and its spatial coordinates are quickly extracted through grid indexing. The B-spline curve smooth connection uses the control point interpolation method to achieve the geometric continuity transition between the local obstacle avoidance path and the continuous path segment, and eliminates the sharp corners at the path turning points. The equidistant sampling discretizes the fused obstacle avoidance path at a fixed interval to generate an updated second preset inspection path that meets the control accuracy of the motion execution mechanism.

[0179] This embodiment ensures that the inspection path maintains global task continuity while avoiding sudden obstacles through dynamic obstacle priority determination, multi-path optimization screening, and geometric smoothing processing, improving the real-time response ability and motion stability of the autonomous navigation system in complex environments.

[0180] In a second aspect, this embodiment also provides a computer-readable storage medium, on which computer program instructions are stored, and the computer program instructions, when executed by a processor, implement the method described in the first aspect.

[0181] The computer program involved in this embodiment can be stored in a computer-readable storage medium. The computer-readable storage medium includes, but is not limited to, magnetic disks, magnetic tapes, magnetic cards, floppy disks, flash memories, optical discs, optical cards, read-only memories (ROMs), random access memories (RAMs), erasable programmable ROMs (EPROMs), and electrically erasable programmable ROMs (EEPROMs), etc. It also includes other biological, physical, or chemical structures that can achieve the same or equivalent functions as the above-listed storage media, such as units with information storage capabilities like DNA, RNA, proteins, etc. In a specific embodiment, the storage medium involved can be one of the above medium types or a combination of the above medium types. In different embodiments, the computer program involved in the embodiment can be centrally stored in a single medium or distributedly stored in multiple media. The memory containing the computer-readable storage medium can be a non-volatile memory or a random access memory. These computer-readable storage media can be built into the device or can be connected to the device involved in the embodiment as an external device or a part of an external device. In some embodiments, the memory with the computer-readable storage medium is deployed locally; in other embodiments, a scheme of deploying the memory away from the processor can also be adopted, such as a network-attached memory accessed via an RF circuit or an external port and a communication network, where the communication network can be the Internet, one or more internal networks, a local area network (LAN), a wide area wireless network (WLAN), a storage area network (SAN), etc., or a suitable combination thereof, as long as the computer device can access the memory. In addition, the computer program involved in the embodiment can be stored in plaintext / ciphertext form or can be designed as training data and be integrally reorganized and implicitly stored in the parameter states of a deep neural network or other machine learning models through model training.

[0182] Please refer to Figure 2 , in a third aspect, this embodiment also provides an electronic device 1, including a memory 11 and a processor 12. The memory 11 is used to store one or more computer program instructions. Among them, the one or more computer program instructions are executed by the processor 12 to implement the method described in the first aspect.

[0183] The processor 12 described in this embodiment can be implemented by hardware, firmware, software, or a combination thereof. It can use circuits, one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), central processing units (CPUs), controllers, microcontrollers, microprocessors, or at least one of the like. It also includes other physical, biological, or chemical structures that can achieve functions similar to or equivalent to those of the above-listed processors, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some steps, all steps, or any combination of the steps mentioned in the computer programs or methods involved in the various embodiments of this application.

[0184] Adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are as follows:

[0185] Through the multi-modal positioning fusion and dynamic weight allocation mechanism, the present invention combines the complementary cooperation of the RTK-GPS mode, visual inertial navigation mode, and laser SLAM positioning mode to achieve centimeter-level positioning accuracy and seamless switching ability in complex indoor and outdoor scenarios. Based on the A* algorithm and RRT* optimization, a global first preset inspection path is generated. Combining the dynamic window method, improved RRT* algorithm, and QZN-Planner lightweight planner, a local obstacle avoidance path is quickly generated. The path continuity and execution accuracy are ensured through B-spline curve smoothing connection and equidistant sampling. Environmental features are constructed by fusing object detection, point cloud clustering, and Kalman filtering to form multi-dimensional representations. Combining confidence scoring and spatio-temporal indexing, the local environmental map is dynamically optimized to effectively distinguish stable obstacles from instantaneous noise. The cooperative processing of multi-source data and spatio-temporal consistency verification enhance the anti-interference ability of the system. The fusion of laser point cloud and visual semantic information improves the confidence of obstacle avoidance. The dynamic weight allocation mechanism ensures the positioning robustness when the RTK signal weakens. This method takes into account both the global path optimality and local obstacle avoidance real-time performance, reduces the computational load through lightweight planning and multi-modal perception, and significantly improves the motion stability, environmental adaptability, and task continuity of the inspection robot in mixed scenarios.

[0186] The above are only some embodiments of the present invention, and thus do not limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall similarly be included in the patent protection scope of the present invention.

Claims

1. A dynamic programming method for inspection paths, characterized in that, Applicable to inspection robots, the method includes: Obtain environmental information, and select a positioning mode according to the environmental information. The positioning modes include RTK-GPS mode, visual inertial navigation mode, and laser SLAM positioning mode. The environmental information includes indoor environment and outdoor environment; Obtain global map information, and generate positioning information of the current inspection robot according to the positioning mode; Generate a local environmental map within a preset range according to the positioning information and the global map information; Moreover, obtain a first preset inspection path associated with the global map information, and generate a second preset inspection path according to the first preset inspection path and the local environmental map; If the environmental information is an outdoor environment, the method further includes: Collect environmental feature information in the outdoor environment and extract environmental features; Update the local environmental map according to the environmental features, generate a local obstacle avoidance path, and update the local obstacle avoidance path to the second preset inspection path.

2. The dynamic planning method for the inspection path according to claim 1, wherein If the environmental information is an indoor environment, the positioning mode is visual inertial navigation mode or laser SLAM positioning mode; If the environmental information is an outdoor environment, the positioning mode is RTK-GPS mode.

3. The dynamic planning method for inspection path according to claim 2, characterized in that The RTK-GPS mode is configured as: Receive carrier phase observations of the satellite navigation system through the RTK-GPS module, and perform real-time kinematic differential positioning with a ground reference station to obtain positioning data of the RTK-GPS module. The positioning data includes longitude and latitude coordinates; When the RTK-GPS module outputs a fixed solution state and the horizontal positioning accuracy is within a preset accuracy threshold, use the positioning data as the positioning information of the inspection robot in the current environmental information and output it; When the RTK-GPS module outputs a floating solution or single-point positioning state, automatically switch to visual inertial navigation mode or laser SLAM positioning mode; The visual inertial navigation mode is configured as: Collect environmental color images and depth information through an RGB-D depth camera; Combine the environmental color images and depth information with the angular velocity and acceleration data output by the inertial measurement unit, and calculate the motion state of the current inspection robot through visual odometry; Convert the motion state into the positioning information of the inspection robot in the current environmental information and output it; The laser SLAM positioning mode is configured as: Obtain environmental point cloud data by scanning with a 3D lidar; Perform voxel filtering and noise reduction processing on the environmental point cloud data; Based on the filtered environmental point cloud data, calculate the matching relationship between the environmental point cloud data and the global point cloud map through the iterative closest point algorithm, and output the matching result; Convert the matching result into the positioning information of the inspection robot in the current environmental information and output it.

4. The dynamic planning method for the inspection path according to claim 3, characterized in that Obtain global map information, and generating the positioning information of the current inspection robot according to the positioning mode includes: When using the RTK-GPS mode, convert the longitude and latitude coordinates output by the RTK-GPS module into position coordinates in the coordinate system corresponding to the global map information, denoted as initial positioning information; When the visual inertial navigation mode is adopted, convert the motion state output by the visual odometry algorithm into initial positioning information corresponding to the global map information; When the laser SLAM positioning mode is adopted, convert the matching result output by the iterative closest point algorithm into initial positioning information corresponding to the global map information; Calculate the matching degree between the initial positioning information and the global map information shown above in real time; When the matching degree is lower than the preset matching threshold, trigger the re-initialization of the positioning mode and update the initial positioning information, recalculate the matching degree between the updated initial positioning information and the global map information until the matching degree meets the preset matching threshold, and record the updated initial positioning information as the final positioning information; Output the final positioning information, which includes three-dimensional position coordinates and heading angle.

5. The dynamic planning method for the inspection path according to claim 1, characterized in that Generating the second preset inspection path according to the first preset inspection path and the local environment map includes: Read the path point sequence of the first preset inspection path from the global map information, and convert the path point sequence into the coordinate system corresponding to the local environment map, denoted as the first path point sequence; Perform density-based spatial clustering processing on the point cloud data in the local environment map to identify discrete point cloud clusters; Calculate the minimum circumscribed cube for each point cloud cluster to obtain the three-dimensional position information and size information of the obstacle; Mark the three-dimensional position information and size information of the obstacle in the local environment map; When it is detected that there is an obstacle on the first path point sequence, it is denoted as a marked obstacle, and the local obstacle avoidance path is generated by using the dynamic window method, including: Taking the current motion state of the inspection robot as the initial condition, sampling multiple groups of velocity combinations in the velocity space, and the velocity combination includes linear velocity and angular velocity; Generate a motion trajectory within a preset time period for each velocity combination; Evaluate the collision risk and execution efficiency of each motion trajectory with the marked obstacle; Select the motion trajectory with zero collision risk and optimal efficiency as the local obstacle avoidance path; Convert the local obstacle avoidance path into a second path point sequence, and splice it with the obstacle-free section of the first path point sequence; Generate the continuous and executable third path point sequence; Connect the third path point sequence to form the second preset inspection path.

6. The dynamic planning method for inspection path according to claim 1, characterized in that The environmental feature information includes environmental color images and point cloud data. Collecting environmental feature information in the outdoor environment and extracting environmental features includes: Perform a deep learning-based object detection algorithm on the environmental color image to identify environmental targets under a preset category, denoted as the object detection result; Extract plane features from the point cloud data to obtain geometric features, and the geometric features include geometric information of the ground and geometric information of obstacles; Spatially align and fuse the object detection result and the geometric features to obtain the environmental features.

7. The dynamic planning method for the inspection path according to claim 6, characterized in that Updating the local environment map according to the environmental features includes: Associatively store the environmental features with the positioning data; And establish a spatio-temporal index of the environmental features based on the time stamp and the positioning data; When the environmental features at the same position are repeatedly detected, update the confidence score of the environmental features at the same position; And output the environmental features extracted at the current moment to the local environmental map.

8. The dynamic planning method for the inspection path according to claim 7, characterized in that Generating a local obstacle avoidance path according to the environmental features and updating the local obstacle avoidance path to the second preset inspection path includes: Identifying local obstacle information in the current traveling direction according to the environmental features, where the local obstacle information includes the geometric features and confidence scores of local obstacles in the current local environmental map; Calculating the avoidance priorities of each local obstacle according to the local obstacle information; Using an improved RRT* algorithm to generate multiple candidate obstacle avoidance paths around the local obstacles; Performing a safety assessment on the candidate obstacle avoidance paths, and excluding the candidate obstacle avoidance paths with a distance from the local obstacles less than a preset safety threshold; Performing a smoothness assessment on the remaining candidate obstacle avoidance paths, and selecting the candidate obstacle avoidance path with the smallest curvature change as the local obstacle avoidance path; Extracting the continuous path segments in the second preset inspection path that are not affected by the local obstacles; Smoothingly connecting the local obstacle avoidance path and the continuous path segments with a B-spline curve to obtain a fused obstacle avoidance path; Performing equidistant sampling on the fused obstacle avoidance path to generate an updated second preset inspection path.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, The computer program instructions, when executed by a processor, implement the method according to any one of claims 1 to 8.

10. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store one or more computer program instructions, where the one or more computer program instructions are executed by the processor to implement the method according to any one of claims 1 to 8.

Images