Inspection robot indoor and outdoor uninterrupted navigation mode selection method constructed by multi-sensor fusion architecture

Abstract

The invention provides a method for selecting indoor and outdoor uninterrupted navigation modes of an inspection robot constructed by a multi-sensor fusion architecture, and belongs to the technical field of robot navigation. The method comprises the following steps: acquiring positioning information and environment information of an inspection robot through a sensor and a laser radar in real time; performing indoor and outdoor scene adaptive discrimination based on the obtained positioning information and environment information; selecting a path planning strategy according to a scene judgment result, performing global path planning on an outdoor scene by adopting a PRM algorithm fused with an improved A * algorithm, and performing indoor and outdoor uninterrupted navigation path planning of the inspection robot based on multi-sensor fusion by adopting the improved A * algorithm on an indoor scene; and the inspection robot moves to the target position according to the planned path. The navigation mode can flexibly adapt to different indoor and outdoor scenes, and the navigation requirements of the inspection robot in diversified application scenes are met.

Description

Technical Field

[0001] The present invention relates to a method for selecting an uninterrupted indoor and outdoor navigation mode for an inspection robot based on a multi-sensor fusion architecture, belonging to the technical field of robot navigation. Background Art

[0002] With the increasing maturity of inspection robot technology, the application of robots replacing humans for inspections is becoming more and more widespread. However, the wide application of inspection robots highly depends on the efficiency and reliability of their navigation methods. Traditional inspection robot navigation technologies are often limited to a single indoor or outdoor environment and are difficult to cope with complex scenarios combining indoor and outdoor environments. In these complex scenarios, inspection robots need to efficiently and accurately cross different environments, which poses higher requirements for navigation algorithms.

[0003] Existing inspection robot navigation algorithms are mainly divided into search-based path planning algorithms and sampling-based path planning algorithms. Search-based path planning algorithms, such as breadth-first search and depth-first search, are inefficient and prone to computational redundancy in complex scenarios because they do not consider factors such as the target location. Although the greedy best-first search algorithm improves the path search efficiency, it is prone to falling into local optimal solutions in complex environments. Although Dijkstra's algorithm can guarantee finding the shortest path, its efficiency is low in complex scenarios. The A* algorithm combines the advantages of Dijkstra's algorithm and the greedy best-first algorithm, but its path search efficiency still needs to be improved in large-scale complex scenarios combining indoor and outdoor environments.

[0004] Sampling-based path planning algorithms, such as the PRM (Probabilistic Roadmap) algorithm and the RRT (Rapidly-Exploring Random Tree) algorithm, perform well in specific environments but perform poorly in complex obstacle distributions or narrow passable areas. Asymptotically optimal algorithms such as RRT* can approximate the optimal solution, but they have large computational overhead and slow convergence speed, making it difficult to meet the requirements of navigation tasks with high real-time performance.

[0005] Therefore, how to make full use of various sensors equipped on inspection robots to achieve cross-scenario and efficient navigation control has become a key factor in improving inspection efficiency and operation effect. Summary of the Invention

[0006] The object of the present invention is to provide a method for selecting an uninterrupted indoor and outdoor navigation mode for an inspection robot based on a multi-sensor fusion architecture, so as to solve the problems of low navigation efficiency, sub-optimal paths, and easy detours, pauses, and collisions during the navigation process in the prior art in the scenario combining indoor and outdoor environments. Through multi-sensor fusion and adaptive scene recognition, high-efficiency and safe automatic navigation of the inspection robot in the scenario combining indoor and outdoor environments is realized.

[0007] To achieve the above object, the present invention is realized through the following technical solutions: A method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot based on a multi-sensor fusion architecture, comprising the following steps: Step 1: Obtain the positioning information and environmental information of the inspection robot in real time through sensors and lidar. The positioning information includes linear velocity, angular velocity, and position transformation information, and the environmental information includes depth maps and lidar point cloud information; Step 2: Perform adaptive discrimination of indoor and outdoor scenes based on the obtained positioning information and environmental information, including: Calculate the depth mean and depth difference in the forward direction of the inspection robot for preliminary scene judgment. If the preset inequality group is satisfied, proceed to the next judgment; otherwise, determine it as an indoor scene. The preset inequality group is as follows: , where is the depth mean of the recent frames in the forward direction of the inspection robot, is the maximum difference in pixel depth of the recent frames of the image in the forward direction of the inspection robot, is the maximum measurable depth of the sensor, is the minimum measurable depth of the sensor; If it is preliminarily determined as an outdoor scene, divide the depth map into intervals and extract Harris 3D feature points. Fit a plane through the RANSAC algorithm and count the maximum number of inliers. If the maximum number of inliers satisfies the following inequality, proceed to the next judgment; otherwise, determine it as an indoor scene. The inequality is as follows: ; Take the plane of the inspection robot coordinate system with the front as the axis, divide a region on each side by degrees, and count the number of point clouds with the left observation distance value being the farthest detection distance threshold of the lidar and the number of point clouds with the right observation distance value being the farthest detection distance threshold of the lidar. When is satisfied, it is finally determined as an outdoor scene; otherwise, it is determined as an indoor scene. is the lidar scanning frequency, is the lidar scanning resolution; Step 3: Select a path planning strategy according to the scene discrimination result. For the outdoor scene, use the PRM algorithm to fuse and improve the A* algorithm for global path planning. For the indoor scene, use the improved A* algorithm for path planning; Step 4: The inspection robot moves to the target position according to the planned path.

[0008] Preferably, the recent The average frame depth and the inspection robot's forward direction are close The maximum difference in pixel depth of a frame image is calculated as follows: , in, is the pixel width of the depth map, is the pixel height of the depth map, Represents the depth information of each pixel in the depth map, Indicates the maximum depth value that can be described by all pixels in the depth map. Indicates the minimum depth value that can be described by all pixels in the depth map. Indicates the linear speed of the inspection robot. represents the sampling period of linear velocity and angular velocity, represents the angular velocity of the inspection robot, Indicates the horizontal coordinate of the inspection robot's position transformation, Indicates the vertical coordinate of the inspection robot's position transformation, Indicates the position change information sampling period.

[0009] Preferably, the sensors of the inspection robot include internal sensors and external sensors; the internal sensors include inertial units and wheel encoders, which are used to calculate linear velocity and angular velocity through extended Kalman filtering; the external sensors include RGBD cameras, which are used to obtain position transformation information and depth maps based on changes in landmark positions between adjacent frames.

[0010] Preferably, the improved A* algorithm includes: The cost function is dynamically weighted and improved as follows: , in, is the actual cost from the starting node to the current node n, is the estimated cost from the current node to the target node, is the Euclidean distance from the current node to the target node, is the Euclidean distance from the starting node to the target node; When a node is expanded, child nodes are filtered based on the direction of the target point, and nodes with an angle greater than 90 degrees to the main direction are closed; Perform a secondary screening of the remaining child nodes and close the nodes located at the obstacles.

[0011] Preferably, the path planning of the PRM fusion improved A* algorithm specifically includes: Through the PRM algorithm, the density Scatter points and remove points in obstacle areas; in, is a minimum value greater than 0. Use the filtered points to construct map nodes. After the node construction is completed, traverse all nodes and construct edges. Based on the constructed edges, optimize the edge path through the improved A* algorithm to generate a smooth local path.

[0012] Preferably, the method for screening child nodes based on the direction of the target point is as follows: During the node expansion process of the A* algorithm, compare the direction of the line connecting the current node and the target node with the angles of the current surrounding 8 directions, select the direction with the smallest angle as the main direction, and close the potential child nodes in the three directions with an angle greater than 90 degrees with the main direction.

[0013] Preferably, the method for performing secondary screening on the remaining child nodes and closing the nodes located in the obstacle is as follows: Among the remaining child nodes after the initial screening, perform secondary screening on the child nodes in the four directions of up, down, left, and right. If the child node is in the obstacle, close the child node and stop the subsequent expansion of child nodes.

[0014] Preferably, during the movement of the inspection robot, local dynamic obstacle detection is performed through the fusion of a lidar and an RGBD camera. The RGBD camera supplements the observation blind area in front of the lidar with a depth range of ; In the formula, is the vertical distance from the center point of the rigid body of the lidar carried by the inspection robot to the ground, is the pitch observation angle range of the lidar, is the maximum radial distance of the nearest observation blind area of the lidar.

[0015] The advantages of the present invention are as follows: Through multi-sensor fusion and adaptive scene recognition, the inspection robot can efficiently search for better paths in the combined indoor and outdoor scenarios, meeting the real-time navigation application requirements in complex large-scale scenarios.

[0016] The fusion of the RGBD camera improves the dynamic obstacle avoidance ability during navigation. By supplementing the detection blind area of the lidar, collisions of the inspection robot during navigation are avoided.

[0017] The combined use of the improved A algorithm and the PRM fusion improved A algorithm improves the path search efficiency and the quality of the navigation path, reducing the phenomena of invalid inflection points and path extension.

[0018] The navigation mode of the present invention can flexibly adapt to different indoor and outdoor scenarios, meeting the navigation requirements of the inspection robot in diverse application scenarios. Brief Description of the Drawings

[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation on the present invention.

[0020] Figure 1 It is a schematic flowchart of the method of the present invention; Figure 2 It is a schematic flowchart of the adaptive selection of the navigation algorithm for the example of the present invention. Detailed implementation manners

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] Embodiment 1 As Figure 1 - Figure 2 shown, a method for constructing an uninterrupted indoor and outdoor navigation mode selection for an inspection robot based on a multi-sensor fusion architecture solves the technical problem that when the inspection robot navigates in an indoor and outdoor combined scenario, due to the huge differences in the structures of the indoor and outdoor scenarios, the conventional algorithms cannot ensure both the path search efficiency and the relatively optimal navigation path mileage in such scenario areas, and there are often problems of detouring and pausing in the middle and collisions during the navigation process. The following technical solutions are specifically adopted: Multi-sensor data fusion: The inspection robot obtains its own positioning information and environmental observation information through internal sensors (such as inertial units and wheel encoders) and external sensors (such as lidar and RGBD cameras).

[0023] Adaptive discrimination of indoor and outdoor scenarios: Adaptive discrimination of indoor and outdoor scenarios is carried out by fusing lidar with an RGBD camera, and the current scenario is comprehensively judged using depth information, light intensity, and geometric structure information.

[0024] Multi-strategy path planning: Different static global path planning algorithms are selected according to the scenario multi-strategies. The PRM fusion improved A algorithm is used for path planning in the outdoor scenario, and only the improved A algorithm is used for the indoor and outdoor uninterrupted navigation path planning of the inspection robot based on multi-sensor fusion in the indoor scenario.

[0025] Improved A* algorithm: The cost function of the original A* algorithm is designed with dynamic distance weighting to improve the node search efficiency; during the node expansion process, the child nodes are initially and secondarily screened based on the position information of the target point, and the child nodes with a high danger coefficient are closed.

[0026] PRM Fusion Improved A* Algorithm: First, use the PRM algorithm to perform a rough path search, generate a guiding path, and construct a sampling corridor area in the direction of this path; subsequently, the PRM algorithm scatters points in the scene, constructs map nodes after removing obstacle points, and optimizes the paths of each edge through the improved A* algorithm.

[0027] Specifically, this embodiment includes the following steps: S1: Obtain the positioning information and environmental information of the inspection robot in real time through sensors and lidar. The positioning information includes linear velocity, angular velocity, and position transformation information, and the environmental information includes depth maps and lidar point cloud information; As a refinement of the above embodiment, the sensors of the inspection robot include internal sensors and external sensors; the internal sensors include an inertial unit and a wheel encoder with a sampling period of for calculating the linear velocity and angular velocity through an extended Kalman filter; the external sensors include an RGBD camera for obtaining position transformation information and depth maps based on the change in the position of road signs between adjacent frames; the sampling period of the RGBD device is and the measurable depth range is .

[0028] S2: Perform indoor and outdoor scene adaptive discrimination based on the obtained positioning information and environmental information, including: S201: Calculate the depth mean and depth difference in the forward direction of the inspection robot for a preliminary scene judgment. If the preset inequality group is satisfied, proceed to the next judgment; otherwise, determine it as an indoor scene. The preset inequality group is as follows: , where is the depth mean of the inspection robot in the forward direction for the recent frames, is the maximum difference in the pixel depth of the inspection robot in the forward direction for the recent frames, is the maximum measurable depth, is the minimum measurable depth, represents rounding down, and in this embodiment takes the value of 10.

[0029] As a refinement of the above embodiment, when all the expressions in the inequality group are satisfied, it is considered that through the analysis of the scene depth information, it is preliminarily judged that the inspection robot is currently in an outdoor scene.

[0030] The calculation methods for the depth mean of the inspection robot in the forward direction for the recent frames and the maximum difference in the pixel depth of the inspection robot in the forward direction for the recent frames are as follows: , Among them, is the pixel width of the depth map, is the pixel height of the depth map, represents the depth information at each pixel position of the depth map, represents the maximum depth value that all pixel points of the depth map can describe, represents the minimum depth value that all pixel points of the depth map can describe, represents the linear velocity of the inspection robot, represents the sampling period of the linear velocity and the angular velocity, represents the angular velocity of the inspection robot, represents the abscissa of the position transformation of the inspection robot, represents the ordinate of the position transformation of the inspection robot, represents the sampling period of the position transformation information, that is, the sampling period of the RGBD camera.

[0031] S202: If it is initially determined to be an outdoor scene, then divide the depth map into intervals and extract Harris3D feature points. Fit a plane through the RANSAC algorithm and count the maximum number of inliers. If the maximum number of inliers satisfies the following inequality, then proceed to the next judgment, otherwise determine it as an indoor scene; As a refinement of the above embodiment, first evenly divide the depth map into the number of intervals equal to the pixel value of the short side of the depth map through the scene geometric structure analysis method. Extract and retain one feature point for each interval using the Harris3D feature extraction algorithm, and a total of feature points are extracted. After extracting the feature points, use the RANSAC algorithm to fit a plane to obtain the maximum number of inliers N, and finally perform scene analysis and judgment. The inequality is as follows: , S203: Divide the plane of the inspection robot coordinate system into one region on each side, left and right, with the front as the axis by degrees, and count the number of point clouds with the left observation distance value being the maximum detection distance threshold of the lidar and the number of point clouds with the right observation distance value being the maximum detection distance threshold of the lidar. When is satisfied, then finally determine it as an outdoor scene, otherwise determine it as an indoor scene. is the lidar scanning frequency, is the lidar scanning resolution; S3: Select a path planning strategy according to the scene discrimination result. For the outdoor scene, the PRM algorithm is used to fuse and improve the A* algorithm for global path planning. For the indoor scene, the improved A* algorithm is used for the indoor and outdoor seamless navigation path planning of the inspection robot based on multi-sensor fusion.

[0032] As a refinement of the above embodiment, the improved A* algorithm includes: S301: Perform distance dynamic weighting design on the cost function of the original A* algorithm path search.

[0033] Specifically, performing distance dynamic weighting design on the cost function of the original A* algorithm path search means that in order to improve the node search efficiency of the original A* algorithm, dynamic weight allocation is performed on the actual cost from the initial position to the current position and the estimated cost from the current position to the target end position. The cost function in the original A* algorithm is: , In the above formula, is the actual cost from the start node to the current node n, is the estimated cost from the current node to the target node. The improved cost function of the improved A* algorithm is as follows: , In the above formula, is the Euclidean distance from the current node to the target node, is the Euclidean distance from the start node to the target node.

[0034] In the initial stage of navigation it is more biased towards the estimated cost , making the path search more conducive to quickly expanding towards the target position. As the path search progresses, will dynamically increase the weight proportion to improve the quality of the path search.

[0035] S302: During the node expansion process, initially screen the child nodes based on the target point position information.

[0036] Specifically, initially screening the child nodes based on the target point position information means that during the node expansion process of the A* algorithm, the angle between the connection direction of the current node and the target node and the eight surrounding directions of the current node is compared, and the direction with the smallest angle is selected as the main direction, and the potential child nodes in the three directions with an angle greater than 90 degrees with the main direction are closed.

[0037] S303: After the initial screening of the child nodes, perform secondary screening and close the child nodes with high risk factors.

[0038] Specifically, after the initial screening of the child nodes, secondary screening is carried out and the child nodes with high risk coefficients are closed, which means that among the remaining 5 child nodes after the initial screening, the child nodes in the four directions of up, down, left, and right are screened again. If the child node is in an obstacle, the child node is closed and the subsequent expansion of the child nodes is stopped.

[0039] As a refinement of the above embodiment, the path planning of the PRM fusion improved A* algorithm specifically includes: Scattering points with density by the PRM algorithm, and removing the points in the obstacle area; , wherein, is the scattering point density, is the estimated cost function of the improved A* algorithm, is a very small value greater than 0. If the grid map information at the position of the point is 1, it means an obstacle (the grid map is composed of countless small grids, and each small grid is marked with 0 and 1 values. 0 is empty and the visualization color is white; 1 is occupied and the visualization color is black. According to the position information of the scattered points, check whether the grid information at the corresponding position is 1 or 0 to determine whether there is an obstacle). After removing the points on the obstacle, the remaining points are used to construct the nodes of the map. After the node construction is completed, all nodes are traversed and edges are constructed, and then the improved A* algorithm is guided by each edge to optimize the paths of each edge, generating a smooth local path.

[0040] It should be noted that: when the inspection robot is navigating and determines that it has switched between indoor and outdoor, it will perform a static global path search again. For example: starting outdoors, the PRM fusion improved A* algorithm is started. During navigation, if it is determined to be indoors, from the current position to the end point, the improved A* algorithm is used. The reverse is true for indoor -> outdoor. Generally, there are more obstacles indoors and the navigation distance is longer outdoors. The advantage of the PRM compared to the A* algorithm is that in the case of not too many obstacles and a long distance, the search efficiency and computational complexity are much higher than those of the A* algorithm. The disadvantage is that in an environment with many obstacles and a small passage space, the searched path is often not optimal. In a large scene, the local incremental static global path planning strategy is used. For example, in the A* algorithm, a static global path is first searched, and when a certain distance is reached, a new section is searched and extended. It will not be a one-step solution because the search time is long and the computational complexity increases exponentially, occupying the processor resources. The improved A* algorithm used in this patent is also a local incremental method and will not search for a path across the starting and ending points when indoors (assuming the starting and ending points exceed the search section distance threshold).

[0041] S4: The inspection robot moves to the target position according to the planned path.

[0042] As a refinement of the above embodiments, during the movement of the inspection robot, local dynamic obstacle detection is performed by fusing a lidar and an RGBD camera, and the lidar environmental information is supplemented by fusing the RGBD observation information with a depth range of to avoid collisions when the inspection robot navigates to indoor scenarios such as the corners of long corridor structures.

[0043] , wherein, is the vertical distance from the center point of the rigid body of the lidar carried by the inspection robot to the ground, is the pitch observation angle range of the lidar, is the maximum radial distance of the blind area for the lidar's nearest observation.

[0044] When the inspection robot performs path planning, due to its hardware characteristics, the lidar inevitably has a detection blind area in the X-Z plane in the form of a right triangle with a height of h and an acute angle of . Therefore, by fusing the depth observation information of the RGBD camera in this area for environmental observation supplementation, it can effectively prevent the robot from colliding with dynamic obstacles when moving or turning in place at indoor corners.

[0045] It should be noted that: The present invention performs complementary fusion of the advantages of various sensors such as the lidar and RGBD camera commonly carried by the inspection robot, and particularly designs a strategy for adaptive scene recognition and selection of different path methods according to the structural and requirement differences between indoor and outdoor scenarios. In addition, to improve the path search efficiency and navigation safety, the present invention particularly designs an improved A* algorithm and RGBD dynamic obstacle observation constraints to meet the high-efficiency and safe automatic navigation requirements of the inspection robot in the combined indoor and outdoor scenarios.

[0046] Finally, it should be noted that: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot based on a multi-sensor fusion architecture, characterized in that, It includes the following steps: Step 1: Obtain the positioning information and environmental information of the inspection robot in real time through sensors and lidar. The positioning information includes linear velocity, angular velocity, and position transformation information, and the environmental information includes depth map and laser point cloud information; Step 2: Perform indoor and outdoor scene adaptive discrimination based on the obtained positioning information and environmental information, including: Calculate the depth mean and depth difference in the forward direction of the inspection robot for preliminary scene judgment. If the preset inequality group is satisfied, proceed to the next judgment; otherwise, determine it as an indoor scene. The preset inequality group is as follows: ， Among them, is the mean depth of the near frame in the forward direction of the inspection robot, is the maximum difference in the pixel depth of the near frame image in the forward direction of the inspection robot, is the maximum measurable depth of the sensor, is the minimum measurable depth of the sensor; If it is initially determined to be an outdoor scene, the depth map is divided into intervals and Harris 3D feature points are extracted. The plane is fitted by the RANSAC algorithm and the maximum number of inliers is counted. If the maximum number of inliers satisfies the following inequality, the next step of judgment is carried out; otherwise, it is determined to be an indoor scene. The inequality is as follows: ; The plane of the inspection robot's coordinate system is axially centered on the front, with left and right degrees each divided into a region, and the number of point clouds with the left observation distance value being the maximum detection distance threshold of the lidar is counted and the number of point clouds with the right observation distance value being the maximum detection distance threshold of the lidar . When is satisfied, it is finally determined as an outdoor scene; otherwise, it is determined as an indoor scene. is the scanning frequency of the lidar, is the scanning resolution of the lidar; Step 3: Select a path planning strategy according to the scene discrimination result. For the outdoor scene, use the PRM algorithm to fuse and improve the A* algorithm for global path planning, and for the indoor scene, use the improved A* algorithm for path planning; Step 4: The inspection robot moves to the target position according to the planned path.

2. The method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot constructed with a multi-sensor fusion architecture according to claim 1, wherein, The near direction of the inspection robot's forward movement The average depth of the near frame and the near The calculation method for the maximum difference between the pixel depths of the frame images is as follows: ， Among them, is the pixel width of the depth map, is the pixel height of the depth map, represents the depth information of each pixel of the depth map, represents the maximum depth value that all pixel points of the depth map can describe, represents the minimum depth value that all pixel points of the depth map can describe, represents the linear velocity of the inspection robot, represents the sampling period of the linear velocity and angular velocity, represents the angular velocity of the inspection robot, represents the abscissa of the position transformation of the inspection robot, represents the ordinate of the position transformation of the inspection robot, represents the sampling period of the position transformation information.

3. The method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot constructed with a multi-sensor fusion architecture according to claim 1, wherein The sensors of the inspection robot include internal sensors and external sensors; the internal sensors include an inertial unit and a wheel encoder, which are used to calculate the linear velocity and angular velocity through extended Kalman filtering; the external sensors include an RGBD camera, which is used to obtain the position transformation information and depth map based on the change of landmark positions between adjacent frames.

4. The method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot constructed with a multi-sensor fusion architecture according to claim 1, wherein The improved A* algorithm includes: Perform dynamic weighting on the cost function, and after improvement, it is: ， Among them, is the actual cost from the starting node to the current node n, is the estimated cost from the current node to the target node, is the Euclidean distance from the current node to the target node, is the Euclidean distance from the starting node to the target node; When expanding nodes, screen sub-nodes based on the direction of the target point, and close the nodes with an angle greater than 90 degrees with the main direction; Perform secondary screening on the remaining sub-nodes and close the nodes located in the obstacles.

5. The method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot constructed with a multi-sensor fusion architecture according to claim 4, characterized in that, The path planning of the PRM fusion improved A* algorithm specifically includes: Disperse points with density by the PRM algorithm and remove the points in the obstacle area; Among them, is a very small value greater than 0. Map nodes are constructed using the filtered points. After the nodes are constructed, all nodes are traversed and edges are constructed. Based on the constructed edges, the edge path is optimized by improving the A* algorithm to generate a smooth local path.

6. The method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot constructed with a multi-sensor fusion architecture according to claim 4, wherein The specific method of screening sub-nodes based on the direction of the target point is as follows: During the node expansion process of the A* algorithm, compare the angle between the connection direction of the current node and the target node with the angles of the current surrounding 8 directions, select the direction with the smallest angle as the main direction, and close the potential sub-nodes in the three directions with an angle greater than 90 degrees with the main direction.

7. The method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot constructed with a multi-sensor fusion architecture according to claim 6, characterized in that, The specific method of performing secondary screening on the remaining sub-nodes and closing the nodes located in the obstacles is as follows: Among the remaining sub-nodes after the initial screening, perform secondary screening on the sub-nodes in the four directions of up, down, left, and right. If the sub-node is in an obstacle, close the sub-node and stop the expansion of subsequent sub-nodes.

8. The method for selecting an uninterrupted indoor and outdoor navigation mode of an inspection robot constructed with a multi-sensor fusion architecture according to claim 1, characterized in that During the movement of the inspection robot, local dynamic obstacle detection is carried out by fusing a lidar and an RGBD camera, and the RGBD camera is used to supplement the observation blind area in front of the lidar with a depth range of . In the formula, is the vertical distance from the center point of the rigid body of the lidar carried by the inspection robot to the ground, is the pitch observation angle range of the lidar, is the maximum radial distance of the blind area for the lidar's near observation.

Images