A positioning mapping method and system of a hexapod robot, a storage medium and an equipment

Abstract

The application discloses a kind of hexapod robot positioning mapping method, system, storage medium and equipment, the method includes: acquisition two-dimensional point cloud data, three-dimensional point cloud data and real-time attitude data;When it is judged that any attitude angle exceeds preset stable threshold, generate gait control instruction for controlling hexapod robot;Two-dimensional point cloud data is handled using improved Hector-SLAM algorithm, generate two-dimensional grid map, and run ORB-SLAM3 algorithm that has fused YOLOv8 target detection and introduced dynamic point grading rejection mechanism to handle three-dimensional point cloud data, generate three-dimensional point cloud map;The map is weighted and fused, to obtain fused environment map;According to the path planning of fusion environment map and current positioning result, generate foot end trajectory point by forward and inverse kinematics solution, control hexapod robot to move, realize continuous positioning and mapping.The application realizes the stable, accurate autonomous positioning and mapping of hexapod robot in dynamic environment by motion control and multi-algorithm fusion.

Description

Technical Field

[0001] This invention relates to the field of localization and mapping technology, specifically to a localization and mapping method, system, storage medium, and device for a hexapod robot. Background Technology

[0002] In robot navigation and unmanned system applications, accurate positioning and mapping of mobile vehicles in unknown environments are core technologies for achieving autonomy. Currently, wheeled or tracked mobile robots have limitations in terrain adaptability, making it difficult for them to walk stably in rugged, unstructured environments. Meanwhile, traditional positioning technologies such as GPS suffer from weak or lost signals in indoor and tunnel environments, failing to meet the requirements for continuous positioning.

[0003] While visual SLAM and laser SLAM technologies can compensate for the shortcomings of GPS, they still face challenges in practical applications. For example, in dynamic environments, moving objects (such as pedestrians) can interfere with feature point extraction and matching, leading to ghosting and localization drift in mapping. Furthermore, traditional SLAM algorithms still suffer from limitations in computational efficiency and convergence, making it difficult to provide stable, real-time performance for multi-DOF robots on embedded platforms. Simultaneously, sudden posture changes in hexapod robots moving through complex terrain can cause instability, and existing systems lack the ability to adaptively adjust gait based on real-time posture feedback. Moreover, the completeness and accuracy of 3D visual mapping in complex scenes still have room for improvement. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide a localization and mapping method, system, storage medium and device for a hexapod robot, aiming to solve the above-mentioned problems described in the prior art.

[0005] The first aspect of the present invention is to provide a localization and mapping method for a hexapod robot, applied to a hexapod robot, the method comprising:

[0006] The robot's operating environment is simultaneously acquired using a pre-set lidar and depth camera to collect two-dimensional and three-dimensional point cloud data. At the same time, the robot's real-time attitude data, including roll angle, pitch angle, and yaw angle, is acquired using an IMU.

[0007] Based on the real-time posture data, when any posture angle is determined to exceed a preset stability threshold, a gait control command is generated, and the movement posture of the hexapod robot is switched from high-speed gait to high-stability gait based on the gait control command.

[0008] Based on the motion model under the high stability gait, the Hector-SLAM algorithm, which is improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy, is used to process the two-dimensional point cloud data to generate a two-dimensional raster map. The ORB-SLAM3 algorithm, which integrates YOLOv8 target detection and introduces a dynamic point hierarchical elimination mechanism, is used to process the three-dimensional point cloud data to generate a three-dimensional point cloud map.

[0009] The two-dimensional grid map and the three-dimensional point cloud map are weighted and fused to obtain a fused environment map that incorporates multi-scale information, and the fused environment map and the current positioning result of the hexapod robot are fed back in real time.

[0010] Path planning is performed based on the fused environment map and the current positioning result. Foot trajectory points are generated through forward and inverse kinematics calculations to control the movement of the hexapod robot, thereby achieving continuous localization and mapping in the environment.

[0011] According to one aspect of the above technical solution, based on the real-time posture data, when any posture angle is determined to exceed a preset stability threshold, a gait control command is generated. The step of switching the hexapod robot's motion posture from a high-speed gait to a high-stability gait based on the gait control command includes:

[0012] The real-time roll angle, pitch angle and yaw angle collected by the IMU are compared with their respective preset stability thresholds. If any angle value continues to exceed the corresponding stability threshold for a set time, it is determined that the six-legged robot is at risk of instability.

[0013] Based on the real-time posture data, the posture stability margin of the hexapod robot is calculated. Based on the stability margin, a matching high-stability gait pattern is selected from the pre-stored gait library, and a gait control command containing the target gait parameters is generated.

[0014] According to the gait control command, the current motion posture of the hexapod robot is smoothly switched from high-speed gait to high-stability gait.

[0015] According to one aspect of the above technical solution, the step of processing the two-dimensional point cloud data and generating a two-dimensional raster map using the Hector-SLAM algorithm improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy includes:

[0016] Construct a scan matching optimization problem, matching the current LiDAR scan point with an existing grid map, with the objective function being: f ( x ),in x The robot pose to be solved;

[0017] The Levenberg-Marquardt method is used for iterative solution, with an iteration step size Δ. x The calculation formula is:

[0018] ;

[0019] In the formula, J ( x ) is the objective function f ( x Jacobian matrix of ) l The damping factor, D It is the identity matrix;

[0020] After each iteration, the damping factor is adaptively updated according to the Nielsen strategy. l If the current iteration decreases the value of the objective function ( βk If the value is greater than 0, then reduce the damping factor. l To accelerate convergence, the expression is: λk = λk -1 / vk -1; otherwise, increase the damping factor. l To enhance stability, the expression is: λk = λk 1· vk -1, where vk =2· vk -1;

[0021] By iterating through loops until convergence, the optimal pose estimate is obtained and the two-dimensional grid map is updated.

[0022] According to one aspect of the above technical solution, the step of running the ORB-SLAM3 algorithm, which integrates YOLOv8 object detection and introduces a dynamic point hierarchical removal mechanism, to process the 3D point cloud data and generate a 3D point cloud map includes:

[0023] The YOLOv8 object detection model is used to process the image data acquired by the depth camera in real time, identify dynamic objects in the image and generate corresponding detection boxes;

[0024] Based on the detection box, the feature points extracted by the ORB-SLAM3 algorithm are dynamically classified, including marking feature points located within the dynamic object detection box as high dynamic risk points, marking feature points located at the edge region of the detection box as potential dynamic points, and marking the remaining feature points as static points.

[0025] In the map point optimization and inter-frame matching stages of the SLAM process, the high-dynamic-risk points are removed, the potential dynamic points are given lower optimization weights, and the static points are given higher optimization weights. Based on the static points and the potential dynamic points, camera pose estimation and 3D map construction are performed to generate a 3D point cloud map.

[0026] According to one aspect of the above technical solution, the step of weightedly fusing the two-dimensional grid map and the three-dimensional point cloud map to obtain a fused environment map that incorporates multi-scale information, and then providing real-time data feedback of the fused environment map and the current positioning result of the hexapod robot, includes:

[0027] Spatial registration and scale normalization are performed on the two-dimensional raster map and the three-dimensional point cloud map to establish coordinate correspondence between map data.

[0028] Based on the preset confidence weight, the obstacle probability information in the registered two-dimensional grid map and the point cloud density and color texture information in the three-dimensional point cloud map are weighted and fused to generate a fused environment map that simultaneously contains two-dimensional obstacle layout and three-dimensional geometric details.

[0029] The fused environment map and the current localization result of the hexapod robot calculated by the SLAM algorithm are used together for data feedback.

[0030] According to one aspect of the above technical solution, the step of weightedly fusing the obstacle probability information in the registered two-dimensional grid map with the point cloud density and color texture information in the three-dimensional point cloud map according to a preset confidence weight, to generate a fused environment map that simultaneously contains two-dimensional obstacle layout and three-dimensional geometric details, includes:

[0031] Assign a first confidence weight based on the inverse ratio of scanning distance to the obstacle probability information in the two-dimensional grid map, and assign a second confidence weight based on the number of feature points to the point cloud density information in the three-dimensional point cloud map;

[0032] The registered 2D obstacle probability information and 3D point cloud density information are weighted and summed according to their respective confidence weights. At the same time, the color and texture information of the 3D point cloud is redundant and completed to generate a fused environmental geometry layer.

[0033] The environmental geometry layer is overlaid with the two-dimensional obstacle layout information, and the map resolution is unified through voxel meshing. Finally, a multi-scale fused environmental map containing a two-dimensional navigation layer and a three-dimensional detail layer is output.

[0034] According to one aspect of the above technical solution, the steps of performing path planning based on the fused environment map and the current positioning result, generating foot trajectory points through forward and inverse kinematics calculations, and controlling the movement of the hexapod robot to achieve continuous localization and mapping in the environment include:

[0035] Global path planning is performed based on the two-dimensional navigation layer in the fused environment map, and combined with the current positioning result, the target pose of the hexapod robot at the next moment is generated using a local obstacle avoidance algorithm.

[0036] Based on the target pose, the foot coordinates required for each leg of the hexapod robot are obtained through inverse kinematics calculation, and a smooth foot trajectory point sequence from the current foot position to the target foot position is generated.

[0037] The sequence of foot trajectory points is converted into joint angle commands for multiple servo motors through forward kinematics calculation, driving the servo motors to perform motion, so that the hexapod robot moves along the planned path, and achieves localization and mapping through continuous environmental perception and mapping updates during the movement.

[0038] A second aspect of the present invention is to provide a localization and mapping system for a hexapod robot applied to the method shown in the above-described technical solution, the system comprising:

[0039] The data acquisition module is used to simultaneously acquire two-dimensional point cloud data and three-dimensional point cloud data of the hexapod robot's operating environment through a preset lidar and depth camera, and at the same time acquire the real-time attitude data of the hexapod robot through the IMU, including roll angle, pitch angle and yaw angle.

[0040] The instruction control module is used to generate a gait control instruction based on the real-time posture data when it is determined that any posture angle exceeds a preset stability threshold, and to switch the motion posture of the hexapod robot from high-speed gait to high-stability gait based on the gait control instruction.

[0041] The map generation module is used to process the two-dimensional point cloud data based on the motion model under the high stability gait, using the Hector-SLAM algorithm improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy to generate a two-dimensional raster map, and to process the three-dimensional point cloud data using the ORB-SLAM3 algorithm which integrates YOLOv8 target detection and introduces a dynamic point hierarchical culling mechanism to generate a three-dimensional point cloud map.

[0042] The map fusion module is used to perform weighted fusion of the two-dimensional raster map and the three-dimensional point cloud map to obtain a fused environment map that incorporates multi-scale information, and to provide real-time data feedback of the fused environment map and the current positioning result of the hexapod robot.

[0043] The localization and mapping module is used to perform path planning based on the fused environment map and the current localization result, generate foot trajectory points through forward and inverse kinematics calculations, control the movement of the hexapod robot, and realize its continuous localization and mapping in the environment.

[0044] A third aspect of the present invention is to provide a readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the method described in the above-described technical solution.

[0045] A fourth aspect of the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described in the above technical solutions.

[0046] Compared with existing technologies, the advantages of using the localization and mapping method, system, storage medium, and device for the hexapod robot shown in this invention are as follows:

[0047] This invention significantly improves the robustness, accuracy, and overall system stability of hexapod robots in complex dynamic environments through multi-sensor fusion and adaptive control strategies. Firstly, by monitoring IMU attitude data in real time and dynamically switching gait, the robot proactively and smoothly transitions from a high-speed gait to a highly stable gait when facing instability risks, ensuring the stability of the data acquisition platform and reducing the interference of motion jitter on the accuracy of the SLAM algorithm from the source, achieving coordinated optimization of motion control and perception mapping. Secondly, an improved parallel processing architecture combining Hector-SLAM and ORB-SLAM3 algorithms is adopted, and LM optimization and Nielsen strategies are introduced to enhance the convergence stability of 2D mapping in unstructured terrain. Furthermore, the fusion of YOLOv8 and a dynamic point hierarchical culling mechanism effectively filters out the interference of dynamic objects on 3D localization and mapping, thus simultaneously obtaining a highly reliable 2D navigation layer and a 3D geometry layer rich in static details. Finally, a multi-scale fusion map is generated by a confidence-based weighted fusion strategy, providing a unified expression for path planning that combines drivability information with rich environmental details. Combined with map-based planning and kinematic control, a tight closed loop of perception-mapping-planning-control is formed, enabling the hexapod robot to achieve high-precision and high-reliability continuous autonomous localization and map building while ensuring its own motion stability. Attached Figure Description

[0048] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0049] Figure 1This is a flowchart illustrating the localization and mapping method for a hexapod robot provided in an embodiment of the present invention.

[0050] Figure 2 This is a structural block diagram of a localization and mapping system for a hexapod robot provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be more thorough and complete.

[0052] It should be noted that when a component is said to be fixed to another component, it can be directly on the other component or there may be an intervening component. When a component is said to be connected to another component, it can be directly connected to the other component or there may be an intervening component. The terms vertical, horizontal, left and right, and similar expressions used in this document are for illustrative purposes only.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The terminology used herein includes, and / or includes, any and all combinations of one or more of the associated listed items.

[0054] Example 1

[0055] Please see Figure 1 The first embodiment of the present invention provides a localization and mapping method for a hexapod robot, applied to a hexapod robot, the method comprising steps S10-S50:

[0056] Step S10: Simultaneously collect two-dimensional point cloud data and three-dimensional point cloud data of the hexapod robot's operating environment using a preset lidar and depth camera, and at the same time collect real-time attitude data of the hexapod robot, including roll angle, pitch angle and yaw angle, using the IMU.

[0057] This invention first acquires environmental information and robot status synchronously through multiple sensors. Specifically, a pre-set LiDAR is responsible for acquiring two-dimensional point cloud slice data of the environment on a horizontal plane, while a depth camera simultaneously acquires three-dimensional point cloud data with depth information. Meanwhile, an inertial measurement unit (IMU) installed on the robot continuously measures the robot's real-time attitude angles, including roll angle (describing body roll), pitch angle (forward and backward tilt), and yaw angle (describing changes in orientation). This synchronous acquisition mechanism ensures the temporal consistency between environmental data and the robot's own state.

[0058] Step S20: Based on the real-time posture data, when any posture angle exceeds a preset stability threshold, a gait control command is generated, and the movement posture of the hexapod robot is switched from high-speed gait to high-stability gait based on the gait control command.

[0059] In this embodiment, IMU data is monitored in real time, and each attitude angle is compared with a preset stability threshold. This stability threshold is preset based on the mechanical structure and stability theory of the hexapod robot. When any attitude angle (such as excessive pitch angle due to uneven ground or rapid movement) is detected to continuously exceed its threshold, it is determined that the robot's current motion state is at risk of instability. At this time, instead of simply stopping the movement, gait control commands are actively generated to guide the robot to smoothly switch from the original high-speed gait (such as flight dynamic gait) which aims to pursue movement speed to a highly stable gait (such as triangular gait or creeping gait) that emphasizes stability.

[0060] Step S30: Based on the motion model under the high stability gait, the Hector-SLAM algorithm, which is improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy, is used to process the two-dimensional point cloud data to generate a two-dimensional raster map. The ORB-SLAM3 algorithm, which integrates YOLOv8 target detection and introduces a dynamic point hierarchical elimination mechanism, is run to process the three-dimensional point cloud data to generate a three-dimensional point cloud map.

[0061] In this embodiment, after the robot enters a highly stable gait, a dual-channel SLAM processing flow is initiated. For 2D point cloud data, an improved Hector-SLAM algorithm is employed. Traditional Hector-SLAM relies on the Gauss-Newton method for scan matching optimization; this invention replaces this with the Levenberg-Marquardt (LM) method, combined with a Nielsen damping factor update strategy. The LM method, by introducing a damping factor, can adaptively adjust between gradient descent and the Newton method, providing stronger robustness when the optimization problem has poor condition numbers (such as sparse environmental features). The Nielsen strategy intelligently amplifies or reduces the damping factor based on the increase or decrease of the objective function value after each iteration, thereby achieving a better balance between convergence speed and stability. This improvement enables the algorithm to reliably generate accurate 2D grid maps even in complex vibration environments that a hexapod robot may encounter.

[0062] In parallel, an improved ORB-SLAM3 algorithm is run on the 3D point cloud data. The improvement lies in integrating the YOLOv8 object detection model and introducing a dynamic point grading and elimination mechanism. YOLOv8 is responsible for quickly and accurately identifying dynamic objects such as pedestrians and vehicles from depth camera images. The feature points extracted by ORB-SLAM3 are then graded: those completely within the dynamic object detection bounding box are considered high-dynamic-risk points, which are highly likely to interfere with localization; those located at the edge of the box are potential dynamic points and require careful handling; the rest are reliable static points. In subsequent pose estimation and map optimization, high-dynamic-risk points are eliminated, and different optimization weights are assigned to points of different grades. This mechanism significantly improves the localization accuracy in dynamic environments and the static consistency of the constructed 3D point cloud map.

[0063] Step S40: The two-dimensional grid map and the three-dimensional point cloud map are weighted and fused to obtain a fused environment map that incorporates multi-scale information, and the fused environment map and the current positioning result of the hexapod robot are fed back in real time.

[0064] In this embodiment, the generated 2D grid map (excelling at representing obstacle outlines and passable areas) and 3D point cloud map (rich in geometric details and color textures) are spatially registered and then fused using a weighted algorithm. The fusion process is not a simple superposition, but rather involves information complementarity based on their respective confidence levels (e.g., the 2D map is based on scan matching scores, and the 3D map on feature point quality), ultimately generating a multi-scale fused environmental map. This map simultaneously possesses a 2D layer conducive to path planning and a structural layer containing rich 3D information. This map, along with the robot's current localization result (pose) calculated in real-time by the algorithm, is fed into a feedback loop, providing the robot with global environmental awareness and its own state information for autonomous decision-making and motion control.

[0065] Step S50: Path planning is performed based on the fused environment map and the current positioning result. Foot trajectory points are generated through forward and inverse kinematics calculations to control the movement of the hexapod robot, thereby achieving continuous localization and mapping in the environment.

[0066] It should be noted that, based on the latest fused map and self-localization, a path planning algorithm is used to calculate global or local target points. Through inverse kinematics, the target motion of the robot body is decomposed into a sequence of specific trajectory points at the ends of its six legs. Then, through forward kinematics, the foot trajectories are converted into rotation angle commands for each joint servo motor, driving the robot to move precisely. During movement, sensors continuously collect new data, and the SLAM algorithm updates the map and localization accordingly, forming a continuous closed loop of perception-mapping-planning-control-re-perception, thereby achieving stable and reliable autonomous localization and map building in unknown or dynamic environments.

[0067] Compared with existing technologies, the localization and mapping method for the hexapod robot shown in this embodiment has the following advantages:

[0068] This embodiment significantly improves the robustness, accuracy, and overall system stability of a hexapod robot's localization and mapping in complex dynamic environments through multi-sensor fusion and adaptive control strategies. Firstly, by monitoring IMU attitude data in real time and dynamically switching gait, the robot proactively and smoothly transitions from a high-speed gait to a highly stable gait when facing instability risks, ensuring the stability of the data acquisition platform and reducing the interference of motion jitter on the accuracy of the SLAM algorithm from the source, achieving coordinated optimization of motion control and perception mapping. Secondly, an improved parallel processing architecture of Hector-SLAM and ORB-SLAM3 algorithms is adopted, and LM optimization and Nielsen strategies are introduced to enhance the convergence stability of 2D mapping in unstructured terrain. Furthermore, the fusion of YOLOv8 and a dynamic point hierarchical culling mechanism effectively filters out the interference of dynamic objects on 3D localization and mapping, thereby simultaneously obtaining a highly reliable 2D navigation layer and a 3D geometry layer rich in static details. Finally, a multi-scale fusion map is generated by a confidence-based weighted fusion strategy, providing a unified expression for path planning that combines drivability information with rich environmental details. Combined with map-based planning and kinematic control, a tight closed loop of perception-mapping-planning-control is formed, enabling the hexapod robot to achieve high-precision and high-reliability continuous autonomous localization and map building while ensuring its own motion stability.

[0069] Example 2

[0070] The second embodiment of the present invention also provides a localization and mapping method for a hexapod robot. The method shown in this embodiment is basically similar to the method shown in the first embodiment, except that:

[0071] In this embodiment, based on the real-time attitude data, when any attitude angle exceeds a preset stability threshold, a gait control command is generated. The step of switching the hexapod robot's motion posture from a high-speed gait to a high-stability gait based on the gait control command includes:

[0072] The real-time roll angle, pitch angle and yaw angle collected by the IMU are compared with their respective preset stability thresholds. If any angle value continues to exceed the corresponding stability threshold for a set time, it is determined that the six-legged robot is at risk of instability.

[0073] Based on the real-time posture data, the posture stability margin of the hexapod robot is calculated. Based on the stability margin, a matching high-stability gait pattern is selected from the pre-stored gait library, and a gait control command containing the target gait parameters is generated.

[0074] According to the gait control command, the current motion posture of the hexapod robot is smoothly switched from high-speed gait to high-stability gait.

[0075] It should be noted that this step refines the triggering conditions for gait switching. Specifically, it does not switch immediately when the attitude angle momentarily exceeds the threshold, but requires it to continue for a set duration.

[0076] Upon determining a risk of instability, the system doesn't simply switch to a single preset stable gait. Instead, it introduces the calculation of a stability margin. This stability margin is a quantitative indicator used to measure the distance or safety of the current posture from the theoretical stability boundary. Based on this margin value, the system selects the mode that best matches the current stability margin requirement from a pre-stored gait library containing various highly stable gait modes (such as triangular gaits with different stride amplitudes and leg lift heights). Subsequently, the generated gait control commands not only include the target gait mode command but also the detailed parameters required to achieve that gait, such as stride length, stride height, and cycle, ensuring the executability of the commands.

[0077] Furthermore, the transition from a high-speed gait to a highly stable gait must be smooth. This means that the robot's speed or foot trajectory cannot change abruptly, otherwise it may trigger secondary instability. Smooth transitions are usually achieved through trajectory interpolation or transitional gait sequences to ensure that the robot's center of gravity shift and foot landing are continuous and stable, thereby improving stability while ensuring the stability of the sensor platform required for mapping.

[0078] In this embodiment, the step of using the Hector-SLAM algorithm, improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy, to process the two-dimensional point cloud data and generate a two-dimensional raster map includes:

[0079] Construct a scan matching optimization problem, matching the current LiDAR scan point with an existing grid map, with the objective function being: f ( x ),in x The robot pose to be solved;

[0080] The Levenberg-Marquardt method is used for iterative solution, with an iteration step size Δ. x The calculation formula is:

[0081] ;

[0082] In the formula, J ( x ) is the objective function f ( x Jacobian matrix of ) l The damping factor, D It is the identity matrix;

[0083] After each iteration, the damping factor is adaptively updated according to the Nielsen strategy. l If the current iteration decreases the value of the objective function ( βk If the value is greater than 0, then reduce the damping factor. l To accelerate convergence, the expression is: λk = λk -1 / vk -1; otherwise, increase the damping factor. l To enhance stability, the expression is: λk = λk 1· vk -1, where vk =2· vk -1;

[0084] By iterating through loops until convergence, the optimal pose estimate is obtained and the two-dimensional grid map is updated.

[0085] It's important to note that the core of Hector-SLAM is scan matching, which involves finding a robot pose transformation that maximizes the overall matching accuracy when the current frame's laser point cloud data is projected onto the existing map. This matching accuracy is quantified by an objective function f(x), typically a function of the probability that a point occupies a grid cell on the map. Solving for the optimal pose x* is essentially the process of minimizing f(x).

[0086] Specifically, this embodiment uses the LM method for iterative solution. Its formula for calculating the iteration step size Δx* combines the advantages of gradient descent (ensuring the descent direction) and the Gauss-Newton method (accelerating convergence). The (J^TJ+λD) term in the formula is crucial, where J is the Jacobian matrix of the objective function f(x) with respect to pose x, representing the local gradient information of f, λ is the damping factor, and D is usually taken as the identity matrix or a matrix composed of the diagonal elements of J^TJ. When λ is large, the equation approximates gradient descent, with a small but stable step size; while when λ is small, the equation approximates the Gauss-Newton method, with a large step size and fast convergence.

[0087] It should be noted that this embodiment abandons the fixed or simple heuristic update of λ and adopts the Nielsen strategy. This strategy dynamically adjusts λ based on the actual effect of each iteration (whether the objective function value decreases). Specifically, if the current iteration successfully reduces the objective function value (βk>0), it indicates that the model fits well, and the confidence level can be increased. By reducing λ (dividing by the factor v), the next iteration can be made closer to the Gauss-Newton method, thereby accelerating the convergence speed. Conversely, if the iteration increases the objective function value, it indicates that the current model may be unreliable or has approached the extreme point, and λ needs to be increased (multiplied by the factor v) to enhance the stability of the algorithm and prevent divergence. Here, v is a factor greater than 1, and the update method vk=2·vk-1 allows it to quickly respond to changes in the optimization process. Through this adaptive mechanism, the algorithm can still achieve fast, stable, and accurate pose optimization under the drastic pose changes that may be caused by the movement of the hexapod robot and the noise of the scan data, thereby constructing a high-quality two-dimensional grid map.

[0088] In this embodiment, the step of running the ORB-SLAM3 algorithm, which integrates YOLOv8 object detection and introduces a dynamic point hierarchical removal mechanism, to process the 3D point cloud data and generate a 3D point cloud map includes:

[0089] The YOLOv8 object detection model is used to process the image data acquired by the depth camera in real time, identify dynamic objects in the image and generate corresponding detection boxes;

[0090] Based on the detection box, the feature points extracted by the ORB-SLAM3 algorithm are dynamically classified, including marking feature points located within the dynamic object detection box as high dynamic risk points, marking feature points located at the edge region of the detection box as potential dynamic points, and marking the remaining feature points as static points.

[0091] In the map point optimization and inter-frame matching stages of the SLAM process, the high-dynamic-risk points are removed, the potential dynamic points are given lower optimization weights, and the static points are given higher optimization weights. Based on the static points and the potential dynamic points, camera pose estimation and 3D map construction are performed to generate a 3D point cloud map.

[0092] It should be noted that this embodiment deeply integrates the cutting-edge deep learning object detection model YOLOv8 with the traditional geometric SLAM framework ORB-SLAM3. The YOLOv8 model is known for its fast forward inference speed and high detection accuracy. It can process RGB images from depth cameras in real time, accurately identify and select dynamic objects such as pedestrians, vehicles, and animals in the image, and provide semantic-level prior information for subsequent SLAM feature point selection.

[0093] The ORB-SLAM3 algorithm extracts a large number of ORB feature points from the image. Traditional SLAM treats all these points equally for computation, but this embodiment introduces a dynamic point hierarchical elimination mechanism. First, using the detection boxes generated by YOLOv8, ORB feature points completely within the boxes are marked as high-dynamic-risk points. These points are highly likely to belong to the dynamic objects themselves, and their spatial positions change rapidly over time. If used for pose calculation and mapping, they will introduce serious errors. Second, feature points located at the edges of the detection boxes are marked as potential dynamic points. These points may belong to dynamic objects, or they may belong to their shadows, reflections, or static backgrounds in contact with them, exhibiting uncertainty. Finally, feature points outside the detection boxes are marked as reliable static points.

[0094] In the two core SLAM steps of map point optimization and inter-frame matching, differentiated strategies are adopted for three types of points: 1. High-dynamic-risk points are directly eliminated to fundamentally prevent them from polluting the system. 2. Potential dynamic points are downweighted, given lower weights in the optimization objective function to reduce their impact on the final result. 3. Static points are given high optimization weights and full trust.

[0095] By differentiating feature points, the ORB-SLAM3 algorithm primarily relies on reliable static features when estimating camera pose and constructing 3D map points (point clouds). This not only significantly improves the robot's localization accuracy in dynamic environments but also ensures that the generated 3D point cloud map is mainly composed of static environmental elements, thus significantly enhancing the long-term usability of the map.

[0096] In this embodiment, the step of weightedly fusing the two-dimensional grid map and the three-dimensional point cloud map to obtain a fused environment map that incorporates multi-scale information, and then providing real-time data feedback of the fused environment map and the current positioning result of the hexapod robot, includes:

[0097] Spatial registration and scale normalization are performed on the two-dimensional raster map and the three-dimensional point cloud map to establish coordinate correspondence between map data.

[0098] Based on the preset confidence weight, the obstacle probability information in the registered two-dimensional grid map and the point cloud density and color texture information in the three-dimensional point cloud map are weighted and fused to generate a fused environment map that simultaneously contains two-dimensional obstacle layout and three-dimensional geometric details.

[0099] The fused environment map and the current localization result of the hexapod robot calculated by the SLAM algorithm are used together for data feedback.

[0100] It should be noted that direct fusion is not feasible because 2D raster maps and 3D point cloud maps originate from different sensors and have different coordinate systems, data structures, and scales. Therefore, the first step is to perform spatial registration and scale normalization. This typically relies on extrinsic parameter calibration between sensors, projecting the 3D point cloud data onto the 2D map plane, or establishing a unified world coordinate system for both to ensure that the same physical location in the map has a consistent coordinate correspondence in both types of map data.

[0101] After registration, the complementary information from the two maps is weighted and fused. The advantage of the 2D raster map lies in its clear binary (occupied / vacant) or probabilistic representation of obstacles, making it ideal for navigation and obstacle avoidance planning. The 3D point cloud map, on the other hand, contains the vertical structure of the environment, surface geometric details, and color and texture information. During the actual fusion, a preset confidence weight is assigned to each information source. For example, for close-range areas with high scan quality, the 3D point cloud may be given a higher weight; while for distant areas or areas lacking texture, 2D laser data may be more reliable. The result of this weighted fusion is a fused environmental map that retains the clear division of traversable areas from the 2D layer while overlaying the rich geometric and texture details of the 3D layer, achieving multi-scale complementarity of information.

[0102] Furthermore, it should be noted that the generated fused environment map, together with the robot's current localization result (i.e., position and orientation) calculated in real time by the SLAM algorithm within the fused map coordinate system, constitutes a complete environmental perception and state estimation package. This data package is fed back to the upper-level control system in real time, enabling the robot not only to see the real-time environment but also to know its exact position on the map, providing comprehensive input for subsequent autonomous decision-making and motion control.

[0103] In this embodiment, the step of weightedly fusing the obstacle probability information in the registered two-dimensional grid map with the point cloud density and color texture information in the three-dimensional point cloud map according to a preset confidence weight, to generate a fused environment map that simultaneously includes two-dimensional obstacle layout and three-dimensional geometric details, includes:

[0104] Assign a first confidence weight based on the inverse ratio of scanning distance to the obstacle probability information in the two-dimensional grid map, and assign a second confidence weight based on the number of feature points to the point cloud density information in the three-dimensional point cloud map;

[0105] The registered 2D obstacle probability information and 3D point cloud density information are weighted and summed according to their respective confidence weights. At the same time, the color and texture information of the 3D point cloud is redundant and completed to generate a fused environmental geometry layer.

[0106] The environmental geometry layer is overlaid with the two-dimensional obstacle layout information, and the map resolution is unified through voxel meshing. Finally, a multi-scale fused environmental map containing a two-dimensional navigation layer and a three-dimensional detail layer is output.

[0107] It's important to note that for obstacle probability information in 2D raster maps, a first confidence weight is assigned based on the inverse relationship between the scanning distance and the data. This is because the measurement error of LiDAR typically increases with distance; closer scanning points are more reliable, and the confidence level of the derived obstacle probability information should be higher. For point cloud density information in 3D point cloud maps, a second confidence weight is assigned based on the number of feature points. In ORB-SLAM3, regions rich in feature points generally indicate rich texture and high-quality 3D reconstruction, resulting in a correspondingly higher confidence level for their point cloud density information. This dynamic weight allocation mechanism makes the fusion process more intelligent, allowing for adaptive adjustments based on the quality of the data itself.

[0108] After determining the weights, the registered 2D obstacle probability grid and the 3D point cloud density field (which can be regarded as an occupancy probability representation in 3D space) are weighted and summed. This operation essentially fuses the occupancy grid information from the two sensors within a probabilistic framework, generating a more robust and accurate environmental geometry layer. Simultaneously, the RGB color texture information inherent in the 3D point cloud is subjected to redundancy removal (removing duplicate or incorrect colors) and completion processing (appropriately interpolating or filling areas with missing texture due to occlusion).

[0109] Finally, the detailed environmental geometry layer generated above is overlaid with the two-dimensional obstacle layout information mainly used for navigation (which can be regarded as a binary or probabilistic navigation layer). Then, voxel meshing is used to downsample or normalize the entire map (whether it's a two-dimensional layer or a three-dimensional point cloud) to a uniform spatial resolution (voxel size) to reduce data volume and ensure data format consistency. The final output is a structured, multi-scale fused environmental map containing a two-dimensional navigation layer and a three-dimensional detail layer.

[0110] In this embodiment, the steps of performing path planning based on the fused environment map and the current positioning result, generating foot trajectory points through forward and inverse kinematics calculations, and controlling the movement of the hexapod robot to achieve continuous localization and mapping in the environment include:

[0111] Global path planning is performed based on the two-dimensional navigation layer in the fused environment map, and combined with the current positioning result, the target pose of the hexapod robot at the next moment is generated using a local obstacle avoidance algorithm.

[0112] Based on the target pose, the foot coordinates required for each leg of the hexapod robot are obtained through inverse kinematics calculation, and a smooth foot trajectory point sequence from the current foot position to the target foot position is generated.

[0113] The sequence of foot trajectory points is converted into joint angle commands for multiple servo motors through forward kinematics calculation, driving the servo motors to perform motion, so that the hexapod robot moves along the planned path, and achieves localization and mapping through continuous environmental perception and mapping updates during the movement.

[0114] It's important to note that path planning is divided into two levels: global and local. First, based on the 2D navigation layer in the fused map, global path planning is performed using algorithms such as A and D to obtain a coarse path from the starting point to the target point. Then, combined with the frequently updated current localization results provided by SLAM, and guided by the global path, local obstacle avoidance algorithms such as Dynamic Window (DWA) and artificial potential field methods are employed to consider new obstacles detected by the sensors that may not yet appear in the global map, calculating the safe target pose that the robot needs to reach in the next moment, including both position and orientation.

[0115] After obtaining the robot's target pose, the problem transforms into how to control the movement of its six legs to achieve that pose. At this point, based on the robot's body dimensions, leg lengths (link parameters), and the target pose, the three-dimensional spatial coordinates that each leg's foot should reach in the next instant to support and drive the robot body are calculated. From the current position of one foot to the next target position, a smooth sequence of trajectory points needs to be generated to ensure continuous, shock-free foot movement, thereby guaranteeing the stability of the robot body.

[0116] The sequence of three-dimensional trajectory points at the foot needs to be converted into rotation angles for each servo motor (joint motor) on each leg of the robot. The control system sends these joint angle commands to each servo motor, driving it to rotate precisely, ultimately enabling the foot to move along the planned trajectory, thus moving the entire robot body to the target pose.

[0117] Throughout the movement, the LiDAR, depth camera, and IMU continuously work to collect new environmental and attitude data. The SLAM algorithm continuously updates the fused map and robot localization results based on the new data, forming a complete autonomous motion closed loop of perception-planning-control-reperception. This enables the hexapod robot to continuously explore the unknown environment while moving, while building and updating the environmental map and ensuring its own accurate positioning in the map, thereby achieving synchronous localization and map building.

[0118] Example 3

[0119] Please see Figure 2 The third embodiment of the present invention provides a localization and mapping system for a hexapod robot, applying the method shown in any of the above embodiments, specifically to a hexapod robot. The system includes:

[0120] The data acquisition module 10 is used to simultaneously acquire two-dimensional point cloud data and three-dimensional point cloud data of the hexapod robot's operating environment through a preset lidar and depth camera, and at the same time acquire the real-time attitude data of the hexapod robot through the IMU, including roll angle, pitch angle and yaw angle.

[0121] The instruction control module 20 is used to generate a gait control instruction based on the real-time posture data when it is determined that any posture angle exceeds a preset stability threshold, and to switch the motion posture of the hexapod robot from high-speed gait to high-stability gait based on the gait control instruction.

[0122] The map generation module 30 is used to process the two-dimensional point cloud data based on the motion model under the high stability gait, using the Hector-SLAM algorithm improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy to generate a two-dimensional raster map, and to process the three-dimensional point cloud data by running the ORB-SLAM3 algorithm which integrates YOLOv8 target detection and introduces a dynamic point hierarchical elimination mechanism to generate a three-dimensional point cloud map.

[0123] The map fusion module 40 is used to perform weighted fusion of the two-dimensional grid map and the three-dimensional point cloud map to obtain a fused environment map that incorporates multi-scale information, and to provide real-time data feedback of the fused environment map and the current positioning result of the hexapod robot.

[0124] The localization and mapping module 50 is used to perform path planning based on the fused environment map and the current localization result, generate foot trajectory points through forward and inverse kinematics calculations, control the movement of the hexapod robot, and realize its continuous localization and mapping in the environment.

[0125] Compared with existing technologies, the localization and mapping system for the hexapod robot shown in this embodiment has the following advantages:

[0126] This embodiment significantly improves the robustness, accuracy, and overall system stability of a hexapod robot's localization and mapping in complex dynamic environments through multi-sensor fusion and adaptive control strategies. Firstly, by monitoring IMU attitude data in real time and dynamically switching gait, the robot proactively and smoothly transitions from a high-speed gait to a highly stable gait when facing instability risks, ensuring the stability of the data acquisition platform and reducing the interference of motion jitter on the accuracy of the SLAM algorithm from the source, achieving coordinated optimization of motion control and perception mapping. Secondly, an improved parallel processing architecture of Hector-SLAM and ORB-SLAM3 algorithms is adopted, and LM optimization and Nielsen strategies are introduced to enhance the convergence stability of 2D mapping in unstructured terrain. Furthermore, the fusion of YOLOv8 and a dynamic point hierarchical culling mechanism effectively filters out the interference of dynamic objects on 3D localization and mapping, thereby simultaneously obtaining a highly reliable 2D navigation layer and a 3D geometry layer rich in static details. Finally, a multi-scale fusion map is generated by a confidence-based weighted fusion strategy, providing a unified expression for path planning that combines drivability information with rich environmental details. Combined with map-based planning and kinematic control, a tight closed loop of perception-mapping-planning-control is formed, enabling the hexapod robot to achieve high-precision and high-reliability continuous autonomous localization and map building while ensuring its own motion stability.

[0127] Example 4

[0128] A fourth embodiment of the present invention provides a readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the methods described in the above embodiments.

[0129] Example 5

[0130] A fifth embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the methods described in the above embodiments.

[0131] In the description of this specification, references to the terms "an embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0132] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for localization and mapping of a hexapod robot, characterized in that, Applied to a hexapod robot, the method includes: The robot's operating environment is simultaneously acquired using a pre-set lidar and depth camera to collect two-dimensional and three-dimensional point cloud data. At the same time, the robot's real-time attitude data, including roll angle, pitch angle, and yaw angle, is acquired using an IMU. Based on the real-time posture data, when any posture angle is determined to exceed a preset stability threshold, a gait control command is generated, and the movement posture of the hexapod robot is switched from high-speed gait to high-stability gait based on the gait control command. Based on the motion model under the high stability gait, the Hector-SLAM algorithm, which is improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy, is used to process the two-dimensional point cloud data to generate a two-dimensional raster map. The ORB-SLAM3 algorithm, which integrates YOLOv8 target detection and introduces a dynamic point hierarchical elimination mechanism, is used to process the three-dimensional point cloud data to generate a three-dimensional point cloud map. The two-dimensional grid map and the three-dimensional point cloud map are weighted and fused to obtain a fused environment map that incorporates multi-scale information, and the fused environment map and the current positioning result of the hexapod robot are fed back in real time. Path planning is performed based on the fused environment map and the current positioning result. Foot trajectory points are generated through forward and inverse kinematics calculations to control the movement of the hexapod robot, thereby achieving continuous localization and mapping in the environment. The step of weightedly fusing the two-dimensional grid map and the three-dimensional point cloud map to obtain a fused environment map that incorporates multi-scale information, and then providing real-time data feedback of the fused environment map and the current positioning result of the hexapod robot, includes: Spatial registration and scale normalization are performed on the two-dimensional raster map and the three-dimensional point cloud map to establish coordinate correspondence between map data. Based on the preset confidence weight, the obstacle probability information in the registered two-dimensional grid map and the point cloud density and color texture information in the three-dimensional point cloud map are weighted and fused to generate a fused environment map that simultaneously contains two-dimensional obstacle layout and three-dimensional geometric details. The fused environment map and the current localization result of the hexapod robot calculated by the SLAM algorithm are used together for data feedback; The step of weightedly fusing obstacle probability information in the registered 2D grid map with point cloud density and color texture information in the 3D point cloud map according to preset confidence weights to generate a fused environment map that simultaneously includes 2D obstacle layout and 3D geometric details includes: Assign a first confidence weight based on the inverse ratio of scanning distance to the obstacle probability information in the two-dimensional grid map, and assign a second confidence weight based on the number of feature points to the point cloud density information in the three-dimensional point cloud map; The registered 2D obstacle probability information and 3D point cloud density information are weighted and summed according to their respective confidence weights. At the same time, the color and texture information of the 3D point cloud is redundant and completed to generate a fused environmental geometry layer. The environmental geometry layer is overlaid with the two-dimensional obstacle layout information, and the map resolution is unified through voxel meshing. Finally, a multi-scale fused environmental map containing a two-dimensional navigation layer and a three-dimensional detail layer is output.

2. The localization and mapping method for a hexapod robot according to claim 1, characterized in that, Based on the real-time attitude data, when any attitude angle exceeds a preset stability threshold, a gait control command is generated. The step of switching the hexapod robot's motion posture from a high-speed gait to a high-stability gait based on the gait control command includes: The real-time roll angle, pitch angle and yaw angle collected by the IMU are compared with their respective preset stability thresholds. If any angle value continues to exceed the corresponding stability threshold for a set time, it is determined that the six-legged robot is at risk of instability. Based on the real-time posture data, the posture stability margin of the hexapod robot is calculated. Based on the stability margin, a matching high-stability gait pattern is selected from the pre-stored gait library, and a gait control command containing the target gait parameters is generated. According to the gait control command, the current motion posture of the hexapod robot is smoothly switched from high-speed gait to high-stability gait.

3. The localization and mapping method for a hexapod robot according to claim 2, characterized in that, The steps for processing the two-dimensional point cloud data and generating a two-dimensional raster map using the Hector-SLAM algorithm, which is improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy, include: Construct a scan matching optimization problem, matching the current LiDAR scan point with an existing grid map, with the objective function being: f ( x ),in x The robot pose to be solved; The Levenberg-Marquardt method is used for iterative solution, with an iteration step size Δ. x The calculation formula is: ； In the formula, J ( x ) is the objective function f ( x Jacobian matrix of ) λ The damping factor, D It is the identity matrix; After each iteration, the damping factor is adaptively updated according to the Nielsen strategy. λ If the current iteration decreases the value of the objective function ( βk If the value is greater than 0, then reduce the damping factor. λ To accelerate convergence, the expression is: λk = λk -1 / vk -1; otherwise, increase the damping factor. λ To enhance stability, the expression is: λk = λk 1· vk -1, where vk =2· vk -1; By iterating through loops until convergence, the optimal pose estimate is obtained and the two-dimensional grid map is updated.

4. The localization and mapping method for a hexapod robot according to claim 3, characterized in that, The steps for processing the 3D point cloud data and generating a 3D point cloud map by running the ORB-SLAM3 algorithm, which integrates YOLOv8 object detection and introduces a dynamic point hierarchical culling mechanism, include: The YOLOv8 object detection model is used to process the image data acquired by the depth camera in real time, identify dynamic objects in the image and generate corresponding detection boxes; Based on the detection box, the feature points extracted by the ORB-SLAM3 algorithm are dynamically classified, including marking feature points located within the dynamic object detection box as high dynamic risk points, marking feature points located at the edge region of the detection box as potential dynamic points, and marking the remaining feature points as static points. In the map point optimization and inter-frame matching stages of the SLAM process, the high-dynamic-risk points are removed, the potential dynamic points are given lower optimization weights, and the static points are given higher optimization weights. Based on the static points and the potential dynamic points, camera pose estimation and 3D map construction are performed to generate a 3D point cloud map.

5. The localization and mapping method for a hexapod robot according to claim 1, characterized in that, Based on the fused environment map and the current positioning result, path planning is performed, foot trajectory points are generated through forward and inverse kinematics calculations, and the movement of the hexapod robot is controlled to achieve continuous localization and mapping in the environment. The steps include: Global path planning is performed based on the two-dimensional navigation layer in the fused environment map, and combined with the current positioning result, the target pose of the hexapod robot at the next moment is generated using a local obstacle avoidance algorithm. Based on the target pose, the foot coordinates required for each leg of the hexapod robot are obtained through inverse kinematics calculation, and a smooth foot trajectory point sequence from the current foot position to the target foot position is generated. The sequence of foot trajectory points is converted into joint angle commands for multiple servo motors through forward kinematics calculation, driving the servo motors to perform motion, so that the hexapod robot moves along the planned path, and achieves localization and mapping through continuous environmental perception and mapping updates during the movement.

6. A localization and mapping system for a hexapod robot applied to the method described in any one of claims 1-5, characterized in that, The system, applied to a hexapod robot, includes: The data acquisition module is used to simultaneously acquire two-dimensional point cloud data and three-dimensional point cloud data of the hexapod robot's operating environment through a preset lidar and depth camera, and at the same time acquire the real-time attitude data of the hexapod robot through the IMU, including roll angle, pitch angle and yaw angle. The instruction control module is used to generate a gait control instruction based on the real-time posture data when it is determined that any posture angle exceeds a preset stability threshold, and to switch the motion posture of the hexapod robot from high-speed gait to high-stability gait based on the gait control instruction. The map generation module is used to process the two-dimensional point cloud data based on the motion model under the high stability gait, using the Hector-SLAM algorithm improved by the Levenberg-Marquardt nonlinear optimization method and the Nielsen damping factor update strategy to generate a two-dimensional raster map, and to process the three-dimensional point cloud data using the ORB-SLAM3 algorithm which integrates YOLOv8 target detection and introduces a dynamic point hierarchical culling mechanism to generate a three-dimensional point cloud map. The map fusion module is used to perform weighted fusion of the two-dimensional raster map and the three-dimensional point cloud map to obtain a fused environment map that incorporates multi-scale information, and to provide real-time data feedback of the fused environment map and the current positioning result of the hexapod robot. The localization and mapping module is used to perform path planning based on the fused environment map and the current localization result, generate foot trajectory points through forward and inverse kinematics calculations, control the movement of the hexapod robot, and realize its continuous localization and mapping in the environment.

7. A readable storage medium having computer instructions stored thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-5.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1-5.

Images