An autonomous exploratory semantic map construction method and system

Abstract

The application discloses a kind of autonomous exploration type semantic map construction method and system, unknown environment is autonomously explored based on the robot autonomous exploration algorithm of improved rapidly expanding random tree, global random tree and local random tree are used respectively in exploration process to carry out global frontier point detection and local frontier point detection;After obtaining frontier point, it is clustered, selects the most profitable frontier point as optimal target point, controls robot along optimal path to reach optimal target point;Image sequence of current scene is collected in the process of robot movement and constantly updates robot position, when optimal target point changes, then re-planning path and control robot reaches new optimal target point;RGB image collected in the process of robot autonomous exploration is semantically segmented, and its corresponding depth map is combined together as the input of semantic map construction system, to complete autonomous semantic map construction task, improve the autonomy and intelligence of robot.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of indoor navigation of mobile robots, in particular to an autonomous exploration type semantic map construction method and system. BACKGROUND

[0002] Robot technology, as a representative of future high-tech, is not only the key to the upgrading of traditional manufacturing industry, but also the cornerstone of the development of emerging industries, which has important significance for promoting intelligent manufacturing, improving production efficiency and improving people's livelihood. Among them, mobile robots attract the attention of many scholars due to their high autonomy and intelligence, but the construction of target scene map is still an important research content of mobile robot autonomous navigation, and the high-precision point cloud map with semantic information has important application value for mobile robots to navigate in unknown environment.

[0003] Mobile robots move in unknown environments without any prior information, obtain the surrounding environment structure information through laser or visual sensors, complete self-positioning and map construction of unknown environment at the same time, that is, simultaneous localization and mapping (SLAM) technology. Mobile robots can accurately estimate their own position and understand the geometric structure information in the environment by using SLAM algorithm, but they cannot obtain the attribute information of objects existing in the space, so it is difficult to fully understand the scene information. With the wide application of robot scenes and the complexity of its task requirements, robots need to have scene understanding ability similar to humans, so as to replace humans to complete complex tasks. Common scene perception methods include image classification, target recognition and semantic segmentation. Semantic segmentation, as a pixel-level scene understanding method, can provide detailed interpretation of the target scene. Moreover, with the rapid development of deep learning technology in the field of image perception, especially the achievements of convolutional neural network (CNN), a large number of scholars have begun to apply deep learning to semantic segmentation and combine it with SLAM technology, so as to establish an accurate mapping relationship between object semantic information and geometric map model.

[0004] However, as the intelligence level of robots gradually improves, a map construction method is designed to enable robots to autonomously act and plan routes according to their own wishes, thereby completing the mapping task of unknown environment to replace the existing manual control method to guide the robot mapping. Therefore, the research on autonomous exploration type semantic map construction technology has important theoretical significance and broad application prospect. SUMMARY

[0005] In view of the problems in the prior art, the application provides an autonomous exploration type semantic map construction method, which applies a mobile robot autonomous exploration algorithm to a three-dimensional semantic map construction algorithm, and can achieve autonomous construction of a three-dimensional semantic map of a target scene without control.

[0006] The application is implemented by the following technical solutions:

[0007] An autonomous exploration type semantic map construction method comprises the following steps:

[0008] Step 1: performing autonomous exploration on an unknown environment by using a mobile robot autonomous exploration algorithm based on an improved rapidly-exploring random tree to obtain frontier points of a target scene;

[0009] Step 2: after the frontier points are obtained, clustering the frontier points, calculating the benefits between the positions of the frontier points and the position of the robot, and selecting a frontier point with the maximum benefit as an optimal target point;

[0010] Step 3: establishing an optimal path of the robot to the optimal target point, and continuously updating the position of the robot in the movement process of the robot along the optimal path;

[0011] Step 4: collecting a sequence of scene images in the process of the robot reaching the target point according to the optimal path;

[0012] Step 5: performing semantic segmentation on the collected images by using a trained image semantic segmentation network;

[0013] Step 6: constructing a three-dimensional semantic map of the environment according to the images with semantic information after segmentation and corresponding depth maps.

[0014] Preferably, the autonomous exploration of the mobile robot autonomous exploration algorithm based on the improved rapidly-exploring random tree in step 1 is as follows:

[0015] Global and local random trees are used for global frontier point detection and local frontier point detection, the global random tree expands outward with the starting position of the robot as the root node, and will not be destroyed in the whole exploration process;

[0016] The local random tree expands outward with the current position of the robot as the root node, the whole local random tree is destroyed after a target point is found, and expands outward with the current position of the robot as the root node after the robot reaches the target point; meanwhile, if an obstacle is encountered in the growth process of the random tree, the growth strategy in the direction is continuously maintained, and a variable step length growth mode is adopted in the growth process, and the step length is gradually adjusted with the progress of exploration.

[0017] Preferably, the expression of the step length is as follows:

[0018]

[0019] wherein λ represents a threshold range, different threshold values can be selected according to different experimental environments; M area represents a region of a two-dimensional occupancy grid map detected by the laser radar.

[0020] Preferably, the expression of the benefit R is as follows:

[0021] R(x fp )=λh(x fp ,x r )I(x fp )-N(x fp )

[0022]

[0023] wherein R represents the benefit of each frontier point, the frontier point with the maximum R value is the optimal target point to be reached by the mobile robot in the next step; λ is a weight coefficient, which plays a role in making the information gain occupy a larger proportion in the calculation; x fp represents the position of the fpth frontier point after clustering; x r represents the current position of the robot; I(x) represents the information gain at the frontier point x fp , which is defined as the size of the unknown region expected to be explored at the frontier point x fp , and is calculated by taking the frontier point x fp as the center and counting the number of unknown cells in the circular region with a given radius; N(x) represents the navigation path cost, which represents the path cost consumed by the robot to move to the frontier point x fp ; h(x) is a gain coefficient set according to the distance from the frontier point x fp to the current position x r of the robot.

[0024] Preferably, the specification method of the optimal path in step 3 is as follows:

[0025] The A* algorithm is used to determine the optimal path, and the robot is controlled to reach the optimal target point along the optimal path;

[0026] In the process of movement of the robot to the optimal target point, the current position of the robot is constantly updated, and when the optimal target point changes, the planned path is updated to make the robot explore the new optimal target point.

[0027] Preferably, in step 4, the RGB-D image sequence of the scene during the movement of the robot is collected by using the depth camera carried by the robot itself, and the RGB-D image sequence includes RGB images and corresponding depth maps.

[0028] Preferably, the RGB-D image is subjected to semantic segmentation in step 5 by using a DeepLab V3+ semantic segmentation network.

[0029] Preferably, the segmented image and the corresponding depth map are used as the input of the SLAM system in step 6, the point cloud label distribution probability is updated recursively by using the recursive Bayesian, the semantic information is added to the corresponding point cloud label, and a three-dimensional semantic map of the environment is obtained.

[0030] An autonomous exploration type semantic map construction method system, which executes the autonomous exploration type semantic map construction method when running.

[0031] Compared with the prior art, the present application has the following beneficial technical effects:

[0032] The application discloses an autonomous exploration type semantic map construction method and system, and unknown environment is explored based on a mobile robot autonomous exploration algorithm of an improved rapidly-exploring random tree algorithm; the improved RRT algorithm is used in the mobile robot autonomous exploration, the speed of the mobile robot exploration is improved, and the mobile robot can efficiently explore a target scene in a complex environment; during the mobile robot autonomous exploration, environment information is collected by a depth camera carried by the mobile robot; an image semantic segmentation network trained is used to perform semantic segmentation on the collected image; the image with semantic information after segmentation and the corresponding depth map are used as the input of a SLAM system, a three-dimensional semantic map of the environment is obtained, and thus, the autonomous semantic map construction task is completed. The method combines the mobile robot autonomous exploration algorithm and the semantic map construction algorithm based on deep learning, so that the robot can autonomously complete the construction task of the semantic map of the target scene without human intervention, and the autonomy and intelligence of the mobile robot are greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 is a system overall block diagram of the application.

[0034] Figure 2 is an RRT algorithm flow block diagram of the application.

[0035] Figure 3 is an A* algorithm flow block diagram of the application

[0036] Figure 4 is a DeepLabV3+ semantic segmentation network model schematic diagram of the application.

[0037] Figure 5 is a SLAM system overall block diagram of the application

[0038] Figure 6 is a point cloud fusion flow block diagram of the application. DETAILED DESCRIPTION

[0039] The application will be further described in detail below with reference to the accompanying drawings, which are provided for explanation rather than limitation.

[0040] Referring to Figures 1-6 An autonomous exploration type semantic map construction method, comprising the following steps:

[0041] Step 1: autonomous exploration of an unknown environment based on an improved fast extended random tree mobile robot autonomous exploration algorithm to obtain frontier points of a target scene, the specific method being as follows:

[0042] S1.1) Scanning the environment using an Rplidar laser radar, and marking the explored area as 0, the unexplored area as -1, and the obstacle as 1.

[0043] S1.2) initializing a random tree and taking a random sampling point x rand , and finding the nearest point x nearest to the random sampling point in the random tree. Then, a new point x new is obtained by extending the nearest point to the random sampling point by a fixed length. Subsequently, if there is a cell with a value of -1 between x nearest and x new , the cell is taken as a frontier point and waits for further screening. If the values of all cells between x nearest and x new are 0, it is proved that there is no obstacle therein, and thus the newly grown edge and node are added to the tree; if there is a cell with a value of 1 in the connection between x nearest and x new , it is proved that the tree encounters an obstacle during the growth process, but the growth strategy in this direction is continued at this time, rather than the strategy of abandoning the growth segment adopted in the conventional RRT algorithm.

[0044] S1.3) using a global random tree and a local random tree to ensure the comprehensiveness and rapidity of exploration, the specific method being as follows:

[0045] The global random tree expands outward with the starting position of the robot as the root node, and is not destroyed during the entire exploration process, thereby ensuring the integrity of the exploration of the target scene.

[0046] The local random tree expands outward with the current position of the robot as the root node, but the entire local random tree is destroyed after a target point is found, so as to ensure that the entire algorithm does not occupy a large amount of memory; then the current position of the robot is taken as the root node to expand outward after the robot reaches the target point, thereby ensuring the rapidity of the exploration process.

[0047] S1.4) In the growth process, a variable step size growth method is adopted, and the growth step size is gradually adjusted with the progress of exploration to ensure that the random tree can detect corners and other areas in the unknown environment. The specific expression is as follows:

[0048]

[0049] wherein λ represents a threshold range, and different threshold values can be selected according to different experimental environments; M area represents the area of the two-dimensional occupancy grid map detected by the laser radar.

[0050] Step 2: After obtaining the frontier points, clustering is performed on the frontier points, the benefits between the positions of the clustered frontier points and the robot position are calculated, and the point with the maximum benefit is selected as the optimal target point.

[0051] S2.1) Since the obtained frontier points are not unique, the DBSCAN algorithm is first used to cluster the frontier points.

[0052] S2.2) After clustering all the frontier points detected by the local random tree and the global random tree, since the clustering results are not unique, the robot must select one of the clustered frontier points as the target point of the mobile robot, so as to plan a path and reach the specified target point. In the selection process, the benefits between the robot position and the positions of the frontier points are calculated, and the frontier point with the maximum benefit is selected as the target point that the robot needs to reach. The specific expression of the benefit R is as follows:

[0053] R(x fp )=λh(x fp ,x r )I(x fp )-N(x fp )

[0054]

[0055] wherein R represents the benefit of each frontier point, and the frontier point with the maximum R value is the optimal target point that the mobile robot needs to reach next; λ is a weight coefficient, which makes the information gain occupy a larger proportion in the calculation; x fp represents the position of the fpth clustered frontier point; x r represents the current position of the robot; I(x) represents the information gain at the frontier point x fp , which is defined as the size of the unknown area expected to be explored at the frontier point x fp . The calculation method is to take the point x fp as the center, and calculate the number of unknown cells in the circular area with a given radius; N(x) represents the navigation path cost, which represents the cost of moving the mobile robot to the frontier point x fpThe path cost consumed at the location; h(x) is based on the frontier point x. fp To the current position x of the mobile robot r The distance setting gain coefficient.

[0056] When the front point x fp Located at a given radius h of the robot rad Within the range, make the gain h gain Greater than 1, current edge point x fp Located at a given radius h of the robot rad When outside the range, the gain h will be... gain Setting it to 1 will make the mobile robot more inclined to explore nearby frontier points, thus avoiding repeated exploration.

[0057] Step 3: Use the A* algorithm to plan the robot's path, controlling the robot to reach the target point specified in Step 2 along the optimal path. During the movement, the robot's position and the optimal target point are continuously updated. When the optimal target point changes, the planned path is updated to allow the robot to explore the new optimal target point. This specifically includes the following steps:

[0058] S3.1) Define the cost of each movement (denoted as F), and the cost of moving along a straight line is less than the cost of moving along a diagonal.

[0059] S3.2) Define the cost of the move (denoted as H), which is the straight-line distance (Manhattan distance) from the point to be moved to the target point.

[0060] S3.3) The priority of each node is represented as follows:

[0061] G = F + H

[0062] Furthermore, the A* algorithm uses two sets, close_set and open_set, to represent the set of traversed nodes and the set of untraversed nodes, respectively. During each traversal, the node with the highest priority in the untraversed node set open_set is selected as the next node to be traversed. This process is repeated, and the selected path is the optimal path.

[0063] S3.4) Since the optimal target point will change during the exploration of the mobile robot, i.e. the optimal target point at the current time is not necessarily the optimal target point at the next time. If the robot position is not updated, i.e. the robot does not change the target point before reaching the specified target point, the robot will continue to advance to a large area of unexplored area, thereby causing the small corner in the target scene to be missed, and it is difficult to construct a complete scene map. Therefore, considering the exploration efficiency of the robot, the application updates the current position of the robot during the movement of the robot, and once the optimal target point changes, the new optimal target point is sent to the robot. At the same time, the robot gives up the previous exploration and changes to explore the new optimal target point.

[0064] Step 4, using the depth camera carried by the robot itself to collect the RGB-D image sequence of the current scene in the autonomous exploration process, and save them in two folders respectively, and generate corresponding.txt description files, which lay the foundation for subsequent construction of semantic map.

[0065] Step 5, using the trained DeepLab V3+ semantic segmentation network to perform semantic segmentation on the image sequence obtained in step 5, to obtain images with semantic information and corresponding depth maps, which specifically includes the following steps:

[0066] S5.1) Under the TensorFlow deep learning framework in the Linux operating system, build the DeepLab V3+ semantic segmentation network, mainly including the Encoder module, the Decoder module and the Xception module.

[0067] S5.2) In the Encoder module, dilated convolution with expansion rates of 1, 6, 12 and 18 is used to extract features of different scales for each layer input. At the same time, dilated convolution is combined with depth separable convolution to reduce convolution parameters and reduce computational complexity.

[0068] S5.3) In the Decoder module, low-dimensional features and high-dimensional features are combined, and 3x3 convolution is used to extract more detailed context features, and then 2 times of 4 times up-sampling are performed to output the prediction result.

[0069] S5.4) In the whole DeepLab V3+, Xception is used as the main architecture, and the original pooling layer is replaced by depth separable convolution, and BN layer and Relu activation function are added after each convolution.

[0070] Step 6, taking the image with semantic information and the corresponding depth map as the input of the SLAM system, so as to obtain the three-dimensional semantic map of the indoor environment, which specifically includes the following steps:

[0071] S6.1) recursively update the point cloud label distribution probability, so as to add semantic information to the point cloud label:

[0072]

[0073] wherein c t represents the category probability distribution of the point cloud at time t, represents the key frame set {K0, K1, …, K t}, Z represents a normalization constant, K t represents the key frame at time t.

[0074] S6.2) The system takes a surface element as a model, each surface element records the position and color information of each point, uses an OpenGL point cloud library, splices each surface element, and obtains a global point cloud map with semantic information.

[0075] The application discloses an autonomous exploration type semantic map construction method, which is based on a robot autonomous exploration algorithm of an improved rapidly-exploding random tree to autonomously explore an unknown environment and simultaneously construct a semantic map of a target scene. In the exploration process, a global random tree and a local random tree are used for global frontier point detection and local frontier point detection, respectively. If an obstacle is encountered in the growth process of the random tree, the growth strategy in the direction is continued, instead of abandoning the growth strategy in the section as in the traditional RRT algorithm, so as to ensure the rapidity of exploration and ensure the comprehensiveness of exploration in a variable step length mode. After the frontier points are obtained, the frontier points are clustered, the benefits between the positions of the frontier points and the position of the robot are calculated, and the point with the maximum benefit is selected as an optimal target point, so that the robot is more inclined to explore the nearby frontier points, thereby avoiding repeated exploration. After the optimal target point is determined, path planning is performed, the robot is controlled to reach the target point along the optimal path, the image sequence of the current scene is collected in the process of movement of the robot, and the position of the robot is updated in real time. Once the optimal target point changes, the path is re-planned and the robot is controlled to reach the new optimal target point. A deep learning-based autonomous exploration type semantic map construction system is designed, which can perform semantic segmentation on the RGB images collected in the autonomous exploration process of the robot, and combine the corresponding depth Figure 1 images of the robot to serve as the input of the semantic map construction system, so as to complete the autonomous semantic map construction task and improve the autonomy and intelligence of the robot.

[0076] The application further provides an autonomous exploration type semantic map construction system, which comprises:

[0077] The exploration module is used for exploring the unknown environment by using the improved rapidly-exploding random tree algorithm, and obtaining the boundary between the explored region and the unexplored region.

[0078] A target point selection module is configured to cluster all front edge points and select the front edge point with the highest benefit as the target point of the robot.

[0079] A path planning module is configured to use an A* algorithm to establish an optimal path for the robot to reach the target point.

[0080] An image acquisition module is configured to use a depth camera to acquire an RGB-D image sequence of the robot reaching the target point according to the optimal path.

[0081] An image segmentation module is configured to use a trained image semantic segmentation network to perform semantic segmentation on the acquired images.

[0082] A three-dimensional semantic map construction module is configured to input the segmented images with semantic information and corresponding depth maps into a SLAM system to establish a three-dimensional semantic map of the environment.

[0083] The autonomous exploration type semantic map construction system provided by the application uses an improved RRT algorithm in autonomous exploration of a mobile robot, improves the exploration speed of the mobile robot, and enables the mobile robot to efficiently explore a target scene in a complex environment. Furthermore, the autonomous exploration algorithm is combined with a semantic map construction algorithm based on deep learning, so that the robot can autonomously complete the construction task of a semantic map of the target scene without human intervention, and the autonomy of the mobile robot is greatly improved.

[0084] The division of the modules in the embodiments of the application is illustrative, and is merely a logical function division. In actual implementation, another division mode can be used. In addition, the function modules in each embodiment of the application can be integrated in one processor, or can be physically separated, or two or more modules can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software function module.

[0085] In still another embodiment of the present application, a computer device is provided, which comprises a processor and a memory, the memory being configured to store a computer program, the computer program comprising program instructions, and the processor being configured to execute the program instructions stored in the computer storage medium. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc., which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, and are particularly suitable for loading and executing one or more instructions in the computer storage medium to implement a corresponding method flow or a corresponding function; the processor in the embodiments of the present application can be used for the operation of the autonomous exploration type semantic map construction method.

[0086] In still another embodiment of the present application, the present application further provides a storage medium, specifically a computer readable storage medium (Memory), which is a memory device in a computer device, and is configured to store programs and data. It can be understood that the computer readable storage medium herein can include an internal storage medium in the computer device, and of course can also include an expansion storage medium supported by the computer device. The computer readable storage medium provides a storage space, which stores an operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer readable storage medium herein can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory. One or more instructions stored in the computer readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the autonomous exploration type semantic map construction method in the above embodiments.

[0087] Those skilled in the art will appreciate that embodiments of the application can be devised for a method, a system, or a computer program product. Accordingly, the present application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0088] The present application is described in reference to the flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0089] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0091] Finally, it should be noted that the above-mentioned embodiments are merely intended for describing and illustrating, not limiting the technical solutions of the present application. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered in the protection scope of the claims of the present application.

Claims

1. A method for constructing an autonomous exploratory semantic map, characterized in that, Includes the following steps: Step 1: The mobile robot autonomous exploration algorithm based on the improved fast expanding random tree explores the unknown environment autonomously to obtain the frontier points of the target scene; The autonomous exploration of a mobile robot based on an improved fast expanding random tree algorithm is as follows: Global and local random trees are used for global and local frontier point detection. The global random tree expands outward from the robot's starting position and is not destroyed during the entire exploration process. The local random tree expands outward from the robot's current position as the root node. Once a target point is found, the entire local random tree is destroyed, and it expands outward again from the robot's current position as the root node after the robot reaches the target point. At the same time, if an obstacle is encountered during the growth of the random tree, the growth strategy in that direction is maintained. A variable step size growth method is adopted during the growth process, and the growth step size is gradually adjusted as the exploration progresses. Step 2: After obtaining the leading edge points, cluster them, calculate the benefit between each leading edge point position and the robot position, and select the leading edge point with the largest benefit as the optimal target point. Step 3: Establish the optimal path for the robot to reach the optimal target point, and continuously update the robot's position as the robot moves along the optimal path; The method for updating the optimal path is as follows: The A* algorithm is used to determine the optimal path, and the robot is controlled to reach the optimal target point along the optimal path. As the robot moves toward the optimal target point, its current position is continuously updated. When the optimal target point changes, the planned path is updated to enable the robot to explore the new optimal target point. Step 4: Collect scene image sequences during the process of the robot reaching the target point according to the optimal path; Step 5: Use the trained image semantic segmentation network to perform semantic segmentation on the acquired images; Step 6: Use the segmented image and the corresponding depth map as input to the SLAM system, use recursive Bayesian to update the point cloud label distribution probability, add semantic information to the corresponding point cloud labels, and obtain a 3D semantic map of the environment.

2. The autonomous exploratory semantic map construction method according to claim 1, characterized in that, The expression for the step size is as follows: Where λ represents the threshold range; M area This indicates the area of ​​the two-dimensional grid map detected by the lidar.

3. The autonomous exploratory semantic map construction method according to claim 1, characterized in that, The expression for the profit R is as follows: R(x fp )=λh(x fp ,x r )I(x fp )-N(x fp ) Where R represents the payoff at each frontier point, λ is the weighting coefficient, and x fp Indicates the position of the fp-th frontier point after clustering; x r I(x) represents the current position of the robot; I(x) represents the leading edge point x. fp The information gain at point x, N(x) represents the navigation path cost, and h(x) is the information gain based on the leading edge point x. fp To the robot's current position x r The distance setting gain coefficient.

4. The autonomous exploratory semantic map construction method according to claim 1, characterized in that, In step 4, the robot's own depth camera is used to acquire RGB-D image sequences of the scene during the robot's movement. The RGB-D image sequences include RGB images and corresponding depth maps.

5. The autonomous exploratory semantic map construction method according to claim 1, characterized in that, In step 5, the DeepLab V3+ semantic segmentation network is used to perform semantic segmentation on the RGB-D image.

6. A system for an autonomous exploratory semantic map construction method, characterized in that, When the system is running, it executes the autonomous exploratory semantic map construction method as described in any one of claims 1-5.

Images