Robot autonomous exploration method and system based on deep learning

Abstract

The robot autonomous exploration method and system based on deep learning disclosed in the present disclosure include: obtaining robot coordinate points and constructing the current indoor scene graph; determining the boundary of the current indoor scene graph through a deep learning network; selecting the boundary center closest to the robot coordinate point as a candidate point; generating a tree path between the robot coordinate point and the candidate point through a fast growing random tree algorithm; selecting the second to last node on the tree path as a target point; generating a global path according to the robot coordinate point and the target point; driving the robot to move along the global path at a set speed to obtain new laser point cloud data; updating the current indoor scene graph through the new laser point cloud data to complete indoor scene mapping. High-precision indoor scene mapping is achieved.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a robot autonomous exploration method and system based on deep learning. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Mobile robots have the advantages of flexible movement, easy operation, and strong robustness, and have broad application prospects in the fields of medicine, military, aerospace, logistics, etc. Mobile robot autonomous exploration means that the robot, through autonomous movement and environmental perception in an unknown environment, eventually builds a complete environmental map. Robot autonomous exploration can help people achieve map reconstruction in complex terrains, greatly reducing the adverse effects of extreme environmental conditions (such as narrow, high temperature, etc.) on people, and providing convenience for people to operate environmental maps in the later stage.

[0004] In existing technical solutions, there are mainly the following technical difficulties when robots autonomously explore and build scene maps:

[0005] (1) Low exploration efficiency. When mapping a large area, extracting map boundaries requires a lot of computing resources, and it is difficult to obtain boundaries with a large amount of information, resulting in low robot exploration efficiency.

[0006] (2) Low map completeness. When the robot performs exploratory mapping, the final environment map drawn is incomplete due to the regional limitations of the target point selection.

[0007] (3) Poor map construction flexibility. When selecting a target point, the robot is likely to select a point that is too close to an obstacle, making it difficult for the robot to reach the target point. Summary of the invention

[0008] In order to solve the above problems, the present invention proposes a robot autonomous exploration method and system based on deep learning. The boundaries of known areas and unknown areas in the map are quickly determined through the U-net deep learning network, and candidate points are selected from the boundaries. Then, a tree path is generated between the robot coordinate points and the candidate points through a fast-growing random tree algorithm, and then the target point is determined according to the tree path. When global path planning is performed through the target point, the function of real-time obstacle avoidance can be realized, and finally high-precision indoor scene mapping can be achieved.

[0009] To achieve the above objectives, the present disclosure adopts the following technical solutions:

[0010] First, a robot autonomous exploration method based on deep learning is proposed, including:

[0011] Get the robot coordinate points and build the current indoor scene graph;

[0012] Determine the boundary of the current indoor scene graph through a deep learning network;

[0013] Select the boundary center closest to the robot coordinate point as the candidate point;

[0014] Generate a tree path between the robot coordinate point and the candidate point through a fast-growing random tree algorithm;

[0015] Select the second to last node on the tree path as the target point;

[0016] Generate a global path based on the robot coordinate points and target points;

[0017] Drive the robot along the global path at a set speed to obtain new laser point cloud data;

[0018] The current indoor scene map is updated through the new laser point cloud data to complete the indoor scene mapping.

[0019] Secondly, a robot autonomous exploration system based on deep learning is proposed, including:

[0020] Data acquisition module, used to obtain the robot coordinate points and construct the current indoor scene map;

[0021] The target selection module determines the boundary of the current indoor scene graph through a deep learning network; selects the boundary center closest to the robot coordinate point as the candidate point; generates a tree path between the robot coordinate point and the candidate point through a fast-growing random tree algorithm; and selects the second-to-last node on the tree path as the target point;

[0022] Path planning module, used to generate a global path based on the robot coordinate points and target points;

[0023] The robot driving module is used to drive the robot to move along the global path at a set speed and obtain new laser point cloud data;

[0024] The mapping module is used to update the current indoor scene map through laser point cloud data to complete the indoor scene mapping.

[0025] In a third aspect, an electronic device is proposed, comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein when the computer instructions are run by the processor, the steps described in the robot autonomous exploration method based on deep learning are completed.

[0026] In a fourth aspect, a computer-readable storage medium is proposed for storing computer instructions. When the computer instructions are executed by a processor, the steps described in the robot autonomous exploration method based on deep learning are completed.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. Based on the edge detection principle, the present invention uses the U-net segmentation network to quickly extract the boundaries between known and unknown areas in the map, which helps to improve the speed of boundary extraction, obtain the optimal target point, and effectively solve the problems of small amount of information and regional limitations of the target point.

[0029] 2. The present invention establishes a random tree between the robot coordinate points and the target points, which can simultaneously solve the problems of global path planning and target point selection.

[0030] 3. The present invention adopts the RRT algorithm to select the target point, which effectively solves the problem that the target point is too close to the obstacle, making it impossible for the robot to reach it.

[0031] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings in the specification, which constitute a part of the present application, are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0033] Figure 1 It is a framework diagram of the method disclosed in Example 1;

[0034] Figure 2 A map grid unit diagram with the center as the boundary unit disclosed in Example 1;

[0035] Figure 3 A map grid unit diagram with a center being a non-boundary unit disclosed in Example 1;

[0036] Figure 4 The U-net deep learning network structure diagram disclosed in Example 1;

[0037] Figure 5 The flowchart of the RRT algorithm disclosed in Example 1;

[0038] Figure 6 A schematic diagram of the robot selecting a target point disclosed in Example 1;

[0039] Figure 7 This is a schematic diagram of the movement of the robot at adjacent moments disclosed in Example 1;

[0040] Figure 8 This is a schematic diagram of robot speed sampling disclosed in Example 1;

[0041] Fig. 9 This is a diagram of the Cartographer-SLAM algorithm architecture disclosed in Example 1.

[0042] Among them: 1. Occupied cells, 2. Idle cells, 3. Unknown cells, 4. Boundary cells, 5. Non-boundary cells, 6. Fully connected layer, 7. Pooling layer, 8. Convolutional layer, 9. Map data stream, 10. Obstacles, 11. Expansion area of ​​obstacles, 12. Robot base, 13. Target point, 14. Candidate points, 15. Global path, 16. Robot. DETAILED DESCRIPTION

[0043] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0044] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0045] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0046] In the present disclosure, terms such as "upper", "lower", "left", "right", "front", "back", "vertical", "horizontal", "side", "bottom" and the like indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are relational words determined only for the convenience of describing the structural relationships of the various parts or elements of the present disclosure, and do not specifically refer to any part or element in the present disclosure and should not be understood as limitations on the present disclosure.

[0047] In the present disclosure, terms such as "fixed connection", "connected", "connection", etc. should be understood in a broad sense, indicating that it can be fixedly connected, integrally connected or detachably connected; it can be directly connected or indirectly connected through an intermediate medium. For relevant scientific research or technical personnel in this field, the specific meanings of the above terms in the present disclosure can be determined according to specific circumstances, and they cannot be understood as limitations on the present disclosure.

[0048] Example 1

[0049] In order to achieve high-precision indoor scene mapping, a robot autonomous exploration method based on deep learning is disclosed in this embodiment, including:

[0050] Get the robot coordinate points and build the current indoor scene graph;

[0051] Determine the boundary of the current indoor scene graph through a deep learning network;

[0052] Select the boundary center closest to the robot coordinate point as the candidate point;

[0053] Generate a tree path between the robot coordinate point and the candidate point through a fast-growing random tree algorithm;

[0054] Select the second to last node on the tree path as the target point;

[0055] Generate a global path based on the robot coordinate points and target points;

[0056] Drive the robot along the global path at a set speed to obtain new laser point cloud data;

[0057] The current indoor scene map is updated through the new laser point cloud data to complete the indoor scene mapping.

[0058] Furthermore, the specific process of constructing the current indoor scene graph is as follows:

[0059] Obtain the laser point cloud data and robot posture data at the current moment;

[0060] Construct a subgraph based on the robot’s posture data;

[0061] The constructed sub-image is updated through the laser point cloud data at the current moment to obtain the current indoor scene image.

[0062] Furthermore, the robot's position and posture data includes robot coordinate points, mileage data, posture data and acceleration.

[0063] Furthermore, the current indoor scene graph constructed is a two-dimensional occupancy grid map, and the map grid cells are divided into free cells, unknown cells and occupied cells according to the grid values. When a cell is a free cell and there is at least one unknown cell among the eight neighboring cells of the cell, the cell is a boundary cell, and all boundary cells constitute the boundary.

[0064] Furthermore, the specific process of generating a tree path between the robot coordinate point and the candidate point by using the fast growing random tree algorithm is as follows:

[0065] Randomly select a point in the current indoor scene image as a sampling point;

[0066] Select the node closest to the sampling point as the growth node;

[0067] Generate a new node from the growth node toward the sampling point;

[0068] Add new nodes that meet the conditions to the tree to generate a tree path.

[0069] Furthermore, the selection process of the robot setting speed is:

[0070] Get the robot's motion speed range;

[0071] Sampling multiple groups of speeds within the motion speed range;

[0072] Calculate the robot motion trajectory for each set of speeds;

[0073] Select the optimal motion trajectory from the robot's motion trajectory through the evaluation function;

[0074] The speed corresponding to the optimal motion trajectory is the set speed for driving the robot to move.

[0075] Furthermore, the robot motion is simplified into linear motion, and the robot motion model is constructed;

[0076] The robot motion trajectory under each set of speeds is calculated based on the robot's motion model.

[0077] The deep learning-based robot autonomous exploration method disclosed in this embodiment is discussed in detail.

[0078] Robot autonomous exploration methods based on deep learning, such as Figure 1 As shown, including:

[0079] S1: Get the robot coordinate points and build the current indoor scene graph.

[0080] In specific implementation, Fig. 9 As shown in Figure 1, Cartographer-SLAM mapping algorithm is used to build indoor scene graph. This SLAM algorithm is a SLAM algorithm based on graph optimization, and the algorithm is mainly composed of Local SLAM and GLOBAL SLAM. Wherein, Local SLAM establishes submap by receiving several adjacent frames of point cloud data after filtering. Whenever a frame of point cloud data is obtained, it is matched with the submap established recently, and the point cloud data of the current frame is inserted into the submap (i.e. scan_matching process). By constantly inserting new point cloud data frames, the updating of the submap is realized. The submap and posture estimation established by point cloud matching are reliable in a short time, but large cumulative errors will be generated for a long time, so it is necessary to further determine the global posture and global map of the robot by loop detection and back-end optimization (GLOBAL SLAM).

[0081] Among them, the specific process of constructing the current indoor scene graph is:

[0082] Obtain the laser point cloud data and the robot's position and posture data at the current moment, where the robot's position and posture data includes robot coordinate points, mileage data, posture data, acceleration and other perception data;

[0083] Construct a subgraph based on the robot’s posture data;

[0084] The constructed sub-image is updated through the laser point cloud data at the current moment to obtain the current indoor scene image.

[0085] The indoor scene graph constructed in this embodiment is a two-dimensional occupancy grid map.

[0086] S2: Determine the boundary of the current indoor scene graph, select the boundary center closest to the robot coordinate point as the candidate point; generate a tree path between the robot coordinate point and the candidate point through the fast growing random tree algorithm; select the second to last node on the tree path as the target point.

[0087] In the specific implementation, the target selection module includes a frontier detector and a global random tree planner, which are responsible for extracting the boundary points in the map and selecting an optimal target point from the boundary points. The target point is used to subsequently construct the global path of the robot.

[0088] The boundary detector is used to quickly determine the boundary of the current indoor scene map. Since the indoor scene map constructed in this embodiment adopts a two-dimensional occupancy grid map, the occupancy grid map is a grid map, and each grid has a grid value, which is the probability value of the grid being occupied. Therefore, by filling different colors in grids with different map values, a map representing environmental characteristics can be obtained, such as Figure 2 As shown, according to different grid values, the map grid cells can be divided into free cells 2 (Free), unknown cells 3 (Unknown), and occupied cells 1 (Occupied). In this embodiment, the boundary cells are defined as grid sets that meet the following conditions: Condition 1: Figure 3 As shown, the boundary cell 4 is an idle cell; Condition 2: If Figure 2 As shown, there is at least one unknown cell 3 in the eight neighboring grids of the boundary cell 4. The boundary is a set of boundary cells. The boundary in the map is quickly extracted by using the U-net deep learning network. The reason for extracting the boundary of the map is that the boundary area of ​​the map contains a lot of unknown information, and going to these unknown areas can obtain greater information gain.

[0089] The structure of the U-net deep learning network is as follows Figure 4 As shown, it includes a pooling layer 7, a convolutional layer 8 and a fully connected layer 6. The map data stream 9 flows through the pooling layer 7, the convolutional layer 8 and the fully connected layer 6 to determine the boundary of the map.

[0090] The global planner is used to select the target point that the robot is currently heading to and generate a global path from the robot's coordinate point to the target point.

[0091] The specific process of determining the target point is as follows:

[0092] Select the boundary center closest to the robot's coordinate point as the candidate point for the robot to go to;

[0093] Generate a tree path between the robot coordinate point and the candidate point through a fast-growing random tree algorithm;

[0094] Select the second to last node on the tree path as the target point.

[0095] The Rapidly Growing Random Trees Algorithm (RRT Algorithm) is a tree-like algorithm based on sampling, such as Figure 4 As shown, by randomly sampling points on the map as sampling points, the node closest to the sampling point is selected as the growth node, and then a new node is generated from the growth node toward the sampling point, and the new node that meets certain conditions is added to part of the tree.

[0096] like Figure 5 As shown in FIG. 1 , the rule of random sampling points of RRT is: randomly selecting a point in the specified space and selecting the target point as the sampling point alternately. For example, if the sampling point of this RRT is randomly obtained in a limited space, the sampling point of the next RRT will be selected at the candidate point. This is repeated iteratively to ensure that the RRT will eventually converge to the candidate point 14.

[0097] When the penultimate node on the tree path is selected as the target point 13 and the global path 15 is determined based on the target point, the robot 16 can be prevented from entering the expansion area 11 of the obstacle and the robot base 12 can be prevented from being too close to the obstacle 10.

[0098] S3: Generate a global path based on the robot coordinate points and the target point.

[0099] In the specific implementation, the global path is generated by the path planning module, and the path planning module includes a global planner (Global Planner) and a local planner (Local Planner).

[0100] Among them, after the global planner selects the final target point, it determines the global path according to the robot coordinate point and the target point. The local planning period adopts the local planning algorithm based on the dynamic window (DWA algorithm), so that the robot can move along the global path while avoiding nearby obstacles in real time. The main idea of ​​the DWA algorithm is to sample multiple sets of speeds in the robot's speed space (v, w), and then simulate the robot's motion trajectory within a certain time interval (sim_period) at these speeds. After obtaining multiple sets of motion trajectories, these trajectories are evaluated by the evaluation function, and a speed (v_best, w_best) corresponding to the optimal trajectory is selected to drive the robot to move.

[0101] The process of calculating the robot's motion trajectory is as follows: Assuming that the robot moves in a plane and is not omnidirectional, the robot's motion in adjacent moments can be simplified to a linear motion. Among them, the motion model of the robot's coordinates and yaw angle after a period of time is:

[0102]

[0103] Among them, x t ,y t ,θ t are the position coordinates and yaw angle of the robot after a period of movement, x, y, θ are the current position coordinates and yaw angle of the robot, v, w are the current linear velocity and angular velocity of the robot, Δt is the movement time interval, and α is the angle between the position vector of the robot before and after movement and the positive semi-axis of the x-axis, as shown in Figure 7 shown.

[0104] like Figure 8 As shown in the figure, during the actual movement of the robot, its speed will be subject to various constraints, such as the maximum motor speed, motor torque, maximum braking distance, and safe distance from obstacles. Based on the above constraints, the range of the robot's movement speed at that moment can be obtained, and then multiple groups of speeds are evenly sampled within this range, and the robot's speed remains unchanged within a certain time interval (sim_period).

[0105] After obtaining a set of speeds and their corresponding predicted motion trajectories through speed sampling, the following evaluation function is used to evaluate multiple motion trajectories:

[0106] G(v,w)=w1Δθ(Δt)+w2D nearest +w3v (2)

[0107] Among them, w i is the weight coefficient, Δθ(Δt) is the change of the robot’s yaw angle over a period of time, and D nearestis the distance between the robot and the nearest obstacle, and v is the current speed of the robot. Through the evaluation function, a path that maximizes G(v, w) is selected as the optimal motion path, and the corresponding speed is selected to drive the robot to complete the local path planning.

[0108] S4: The robot is driven to move along the global path at a speed corresponding to the selected optimal motion trajectory to obtain new laser point cloud data; the current indoor scene map is updated with the new laser point cloud data to complete the indoor scene map.

[0109] The method disclosed in this embodiment can quickly extract the boundary between the known area and the unknown area in the map through the deep learning method, select the candidate point, and then generate a tree path between the robot coordinate point and the candidate point through the fast growing random tree algorithm, further select the second to last node on the tree path as the target point, and finally generate a global path between the robot and the target point, effectively solving the problem that the target point is too close to the obstacle, which makes the robot unable to reach it. By establishing a random tree between the robot coordinates and the target point, the global path planning and target point selection problems can be solved at the same time. According to the edge detection principle, the boundary between the known area and the unknown area in the map can be extracted, which helps to obtain the optimal target point, and effectively solves the problem of small information amount and regional limitation of the target point. The adopted Cartographer-SLAM mapping algorithm can realize high-precision indoor scene mapping by matching the laser point cloud with the sub-map for loop detection.

[0110] Example 2

[0111] In this embodiment, a robot autonomous exploration system based on deep learning is disclosed, including:

[0112] Data acquisition module, used to obtain the robot coordinate points and construct the current indoor scene map;

[0113] The target selection module determines the boundary of the current indoor scene graph through a deep learning network; selects the boundary center closest to the robot coordinate point as the candidate point; generates a tree path between the robot coordinate point and the candidate point through a fast-growing random tree algorithm; and selects the second-to-last node on the tree path as the target point;

[0114] Path planning module, used to generate a global path based on the robot coordinate points and target points;

[0115] The robot driving module is used to drive the robot to move along the global path at a set speed and obtain new laser point cloud data;

[0116] The mapping module is used to update the current indoor scene map through new laser point cloud data to complete the indoor scene mapping.

[0117] Example 3

[0118] In this embodiment, an electronic device is disclosed, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, the steps described in the deep learning-based robot autonomous exploration method disclosed in Example 1 are completed.

[0119] Example 4

[0120] In this embodiment, a computer-readable storage medium is disclosed for storing computer instructions. When the computer instructions are executed by a processor, the steps described in the deep learning-based robot autonomous exploration method disclosed in Example 1 are completed.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A robot autonomous exploration method based on deep learning, characterized in that: include: Get the robot coordinate points and build the current indoor scene graph; Use Cartographer-SLAM mapping algorithm to build indoor scene maps; Determine the boundary of the current indoor scene graph through a deep learning network. Specifically, based on the edge detection principle, quickly extract the boundary between the known area and the unknown area in the map through a U-net segmentation network; the U-net segmentation network includes a pooling layer, a convolution layer and a fully connected layer, and the map data flow flows through the pooling layer, the convolution layer and the fully connected layer to determine the boundary of the map; Select the boundary center closest to the robot coordinate point as the candidate point; A tree path is generated between the robot coordinate point and the candidate point through a fast-growing random tree algorithm, specifically: a point is randomly selected as a sampling point in the current indoor scene graph; the node closest to the sampling point is selected as a growth node; a new node is generated from the growth node toward the sampling point; the new node that meets the conditions is added to the tree to generate a tree path; Select the second to last node on the tree path as the target point; Generate a global path based on the robot coordinate points and target points; Drive the robot along the global path at a set speed to obtain new laser point cloud data; Update the current indoor scene map with new laser point cloud data, and finally achieve high-precision indoor scene mapping; The current indoor scene graph constructed is a two-dimensional occupancy grid map. The map grid cells are divided into free cells, unknown cells and occupied cells according to the grid values. When a cell is a known cell and there is at least one unknown cell among the eight neighboring cells of the cell, the cell is a boundary cell, and all boundary cells constitute the boundary. The process of selecting the robot's set speed is as follows: Get the robot's movement speed range; Sampling multiple groups of speeds within the motion speed range; Calculate the robot motion trajectory for each set of speeds; Select the optimal motion trajectory from the robot's motion trajectory through the evaluation function; The speed corresponding to the optimal motion trajectory is the set speed for driving the robot to move; Simplify the robot's motion into linear motion and build the robot's motion model; Calculate the robot motion trajectory at each set of speeds based on the robot's motion model; The process of calculating the robot's motion trajectory is as follows: Assuming that the robot moves in a plane and is not omnidirectional, the robot's motion in adjacent moments can be simplified to a linear motion, where the motion model of the robot's coordinates and yaw angle after a period of time is: in, , , are the position coordinates and yaw angle of the robot after it moves for a period of time. , , θ are the current position coordinates and yaw angle of the robot respectively, , are the current linear velocity and angular velocity of the robot, respectively. is the exercise time interval, is the robot's position vector before and after movement and The angle between the positive half axis and the positive half axis; After obtaining a set of speeds and their corresponding predicted motion trajectories through speed sampling, the following evaluation function is used to evaluate multiple motion trajectories: in, is the weight coefficient, is the change in the robot’s yaw angle over a period of time, is the distance between the robot and the nearest obstacle, and v is the current speed of the robot. The evaluation function is used to select a path that maximizes G(v,w), which is the optimal motion path. The corresponding speed is selected to drive the robot to complete the local path planning.

2. The robot autonomous exploration method based on deep learning as claimed in claim 1, characterized in that: The specific process of constructing the current indoor scene graph is: Obtain the laser point cloud data and robot posture data at the current moment; Construct a subgraph based on the robot’s posture data; The constructed sub-image is updated through the laser point cloud data at the current moment to obtain the current indoor scene image.

3. The robot autonomous exploration method based on deep learning as claimed in claim 2, characterized in that: The robot's position and posture data includes robot coordinate points, mileage data, posture data and acceleration.

4. A robot autonomous exploration system based on deep learning, characterized in that: include: Data acquisition module, used to obtain the robot coordinate points and construct the current indoor scene map; The target selection module determines the boundary of the current indoor scene map through a deep learning network. Specifically, based on the edge detection principle, the boundary between the known area and the unknown area in the map is quickly extracted through the U-net segmentation network; the U-net segmentation network includes a pooling layer, a convolution layer and a fully connected layer, and the map data flow flows through the pooling layer, the convolution layer and the fully connected layer to determine the boundary of the map; Select the boundary center closest to the robot coordinate point as the candidate point; generate a tree path between the robot coordinate point and the candidate point through the fast growing random tree algorithm, specifically: randomly select a point in the current indoor scene graph as the sampling point; select the node closest to the sampling point as the growth node; generate a new node from the growth node toward the sampling point; add the new node that meets the conditions to the tree to generate a tree path; Select the second to last node on the tree path as the target point; Path planning module, used to generate a global path based on the robot coordinate points and target points; The robot driving module is used to drive the robot to move along the global path at a set speed and obtain new laser point cloud data; The mapping module is used to update the current indoor scene map with new laser point cloud data, and finally achieve high-precision indoor scene mapping; The process of selecting the robot's set speed is as follows: Get the robot's movement speed range; Sampling multiple groups of speeds within the motion speed range; Calculate the robot motion trajectory for each set of speeds; Select the optimal motion trajectory from the robot's motion trajectory through the evaluation function; The speed corresponding to the optimal motion trajectory is the set speed for driving the robot to move; Simplify the robot's motion into linear motion and build the robot's motion model; Calculate the robot motion trajectory at each set of speeds based on the robot's motion model; The process of calculating the robot's motion trajectory is as follows: Assuming that the robot moves in a plane and is not omnidirectional, the robot's motion in adjacent moments can be simplified to a linear motion, where the motion model of the robot's coordinates and yaw angle after a period of time is: in, , , are the position coordinates and yaw angle of the robot after it moves for a period of time. , , θ are the current position coordinates and yaw angle of the robot respectively, , are the current linear velocity and angular velocity of the robot, respectively. is the exercise time interval, is the robot's position vector before and after movement and The angle between the positive half axis and the positive half axis; After obtaining a set of speeds and their corresponding predicted motion trajectories through speed sampling, the following evaluation function is used to evaluate multiple motion trajectories: in, is the weight coefficient, is the change in the robot’s yaw angle over a period of time, is the distance between the robot and the nearest obstacle, and v is the current speed of the robot. The evaluation function is used to select a path that maximizes G(v,w), which is the optimal motion path. The corresponding speed is selected to drive the robot to complete the local path planning.

5. An electronic device, characterized in that: It includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the deep learning-based robot autonomous exploration method described in any one of claims 1 to 3 are completed.

6. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of the deep learning-based robot autonomous exploration method described in any one of claims 1 to 3.

Images