Unstructured environment wheel-leg hybrid quadruped robot autonomous navigation system and method based on celestial body perception
Through a celestial perception method, the use of a visual camera to capture the sun's position and combine image processing and sensor data, the self-calibration and real-time positioning of the quadruped robot is realized, solving the positioning accuracy and reliability problems in the environment of limited GPS signal, and improving the performance of the SLAM system.
Patent Information
- Application Number
- CN202510553748.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-29
- Publication Date
- 2025-08-12
AI Technical Summary
In environments where GPS signals are limited or missing, the positioning accuracy and reliability of traditional quadruped robots are significantly affected, especially under factors such as complex lighting conditions and dynamic obstacles, the performance challenges of the SLAM method are great.
Using a method based on celestial perception, the sun's position is captured through a vision camera, combined with image processing algorithms to calculate the relative angle of the sun, and combined with the robot's attitude and motion state, the robot is fused with extended Kalman filtering to achieve self-calibration and real-time positioning of the robot.
It improves the positioning ability of the robot in areas that cannot be covered by GPS, enhances the accuracy and robustness of the SLAM system, and is suitable for complex environments.
Smart Images
Figure CN120467338A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of robot autonomous navigation, and in particular relates to an autonomous navigation system and method for a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception. Background Art
[0002] Traditional quadruped robot autonomous localization and mapping (SLAM) technology relies primarily on feature points in the environment for positioning and mapping. However, in environments where GPS signals are limited or absent, the robot cannot obtain external positioning information, significantly impacting its positioning accuracy and reliability. Furthermore, factors such as changing ambient lighting, dynamic obstacles, and complex terrain further increase the challenges of traditional SLAM methods.
[0003] In recent years, machine vision technology has garnered widespread attention in the field of robotic positioning and navigation. Using high-precision visual sensors, robots can acquire real-time image information of their surroundings, thereby identifying and locating key features in the environment. For example, deep learning-based visual SLAM systems, which utilize deep neural networks to extract environmental features, have demonstrated excellent performance under complex lighting conditions. Summary of the Invention
[0004] In order to solve the above technical problems, the present invention proposes an autonomous navigation system and method for a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception to solve the problems existing in the above-mentioned prior art.
[0005] To achieve the above objectives, the present invention provides a method for autonomous navigation of a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception, comprising:
[0006] Acquire an environmental image, obtain the initial position coordinates of the sun based on the environmental image and perform corrections to obtain the corrected position coordinates of the sun; and obtain the physical angle of the sun based on the corrected position coordinates of the sun;
[0007] Calculate the position and orientation of the robot and correct it by the physical angle of the sun to obtain an initial estimation result and perform calibration to obtain calibration position information;
[0008] The calibration position information is checked for consistency. If the consistency check is passed, the robot performs self-calibration and corrects the real-time positioning result of the robot in real time according to the calibration position information and sensor data. If the check fails, self-calibration is not performed and autonomous navigation of the robot is achieved according to the real-time positioning result.
[0009] Optionally, the process of obtaining the initial sun position coordinates and correcting them includes:
[0010] After preprocessing the environmental image, edge detection is used to obtain edge information of the environmental image. Based on the edge information, Hough circle transform is used to detect circular edges in the environmental image to obtain circular parameters. Target detection and position extraction are performed based on the circular parameters to obtain the initial position coordinates of the sun. The initial position coordinates of the sun are corrected by comparing and calculating multiple images to obtain the corrected position coordinates of the sun.
[0011] Optionally, the preprocessing includes color conversion, threshold segmentation and image denoising.
[0012] Optionally, the process of obtaining the solar physical angle based on the sun-corrected position coordinates includes:
[0013] The corrected solar position coordinates are compared with the preset model. If successfully identified, the camera's field of view angle, image resolution, and camera internal parameters are extracted. The conversion relationship between the image coordinates and the actual angle is calculated in combination with the corrected solar position coordinates to obtain the solar physical angle.
[0014] Optionally, the process of calculating the robot's position and orientation and correcting them using the sun's physical angle to obtain an initial estimation result includes:
[0015] Based on the robot's motion information, the robot's position and orientation are calculated through integration. The robot's relative displacement and rotation between two frames are calculated by analyzing the matching of feature points in consecutive image frames. When the robot starts, the robot's initial orientation relative to the sun is determined by the solar physical angle. During the robot's movement, the robot's orientation is corrected in real time by updating the solar physical angle information.
[0016] The corrected robot position and orientation are fused with the relative displacement and rotation of the robot between the two frames. The error in the fusion result is corrected using the solar physical angle information as an external reference to obtain the initial estimation result.
[0017] Optionally, the process of obtaining the calibration position information includes: acquiring sensor data, performing error analysis on the initial estimation result based on the sensor data, calibrating the initial estimation result according to the error analysis result, and obtaining the calibration position information.
[0018] An optional, self-calibration process includes:
[0019] The data information of the visual camera and the inertial unit are fused, and the error is corrected using the extended Kalman filter.
[0020] The present invention also provides an autonomous navigation system for a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception, comprising a visual control module and a motion control module;
[0021] The visual control module includes the ORB-SLAM3 mapping module and the celestial body recognition and positioning module;
[0022] The ORB-SLAM3 mapping module is used to obtain the position and posture data set of the quadruped robot;
[0023] The celestial body recognition and positioning module is used to obtain the robot position and attitude errors that need to be corrected;
[0024] The motion control module includes WBC motion control and MPC model prediction;
[0025] WBC motion control is used to control the robot's body motion;
[0026] The MPC model prediction is used to predict the position and posture of the robot at the next moment, and finally output the torque parameter file required by the motor to achieve foot-end control of the quadruped robot.
[0027] The present invention also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.
[0028] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.
[0029] Compared with the prior art, the present invention has the following advantages and technical effects:
[0030] The present invention proposes to use solar position information to assist a quadruped robot in self-positioning. The position of the sun is captured in real time by a high-precision visual camera, the relative angle of the sun is calculated based on an image processing algorithm, and the current position and orientation are inferred based on the robot's initial posture and motion state. At the same time, sensor data such as gyroscopes and accelerometers are integrated to improve positioning accuracy and stability. The present invention can effectively improve the robot's positioning capability in areas that GPS cannot cover (such as the polar regions, field resource exploration, etc.), enhance the accuracy and robustness of the SLAM system, and be applicable to various complex environments. By combining solar position information and a visual system, a low-cost, highly reliable positioning solution is provided, breaking through the limitations of traditional SLAM methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:
[0032] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;
[0033] Figure 2 This is a system framework diagram of an embodiment of the present invention. DETAILED DESCRIPTION
[0034] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0035] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0036] Example 1
[0037] The sun, a stable and easily identifiable celestial body, has a highly regular and predictable position on the Earth's surface. Using machine vision technology, robots can capture the sun's position changes in real time and, combined with the robot's posture and motion information, perform self-localization. This approach not only provides reliable positioning support in environments with limited GPS signals, but also effectively improves the accuracy and robustness of SLAM systems.
[0038] This embodiment combines machine vision technology and solar position information to provide a new self-localization solution for quadruped robots. This solution improves the performance of the SLAM system while expanding the application range of robots in complex environments.
[0039] like Figure 1 As shown, this embodiment provides an autonomous navigation method for a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception, including:
[0040] Acquire an environmental image, obtain the initial position coordinates of the sun based on the environmental image and perform correction to obtain the corrected position coordinates of the sun; obtain the physical angle of the sun based on the corrected position coordinates of the sun;
[0041] As a preferred embodiment, the process of obtaining the initial position coordinates of the sun and correcting them includes:
[0042] After preprocessing the environmental image, edge detection is used to obtain the edge information of the environmental image. Based on the edge information, Hough circle transform is used to detect the circular edges in the environmental image to obtain the circular parameters. Target detection and position extraction are performed based on the circular parameters to obtain the initial position coordinates of the sun. The initial position coordinates of the sun are corrected by comparing and calculating multiple images to obtain the corrected position coordinates of the sun.
[0043] As a preferred implementation, the preprocessing includes color conversion, threshold segmentation and image denoising.
[0044] Specifically, the visual camera is initialized and the surrounding environment image is collected by the camera; the color image collected by the camera is processed, which specifically includes:
[0045] Color conversion: Switch the color gamut space to highlight the color information of the sun range in the image.
[0046] Threshold segmentation: Use color threshold to extract the sun area and filter out invalid information.
[0047] Image denoising: Remove unnecessary details from the image through Gaussian filtering and morphological operations.
[0048] Edge detection: Extract edge information from the image using the Canny edge detection algorithm.
[0049] Hough circle transform: Since the sun is circular in shape, Hough circle transform can be used to detect circular edges in the image and then determine the center position of the sun.
[0050] Target detection and position extraction: Identify the position of the sun in the image and extract the sun's position information. Use morphological operations (closing operations) to calculate and calibrate the center or center of mass of the sun as the position coordinates of the sun in the image.
[0051] Error correction: Eliminate errors caused by robot movement, camera shake, and light by comparing and calculating multiple images.
[0052] As a preferred embodiment, the process of obtaining the solar physical angle based on the corrected solar position coordinates includes:
[0053] The corrected sun position coordinates are compared with the preset model. If successfully identified, the camera's field of view angle, image resolution, and camera internal parameters are extracted. Combined with the corrected sun position coordinates, the conversion relationship between the image coordinates and the actual angle is calculated to obtain the sun's physical angle.
[0054] Specifically, the position coordinates of the sun are extracted and compared with the models in the database. If no match is found with any model, the process returns to S1. If the recognition is successful, the process proceeds to S4 to calculate the sun angle.
[0055] Based on the determined sun position coordinates, the camera's field of view (FOV), and the image resolution, the conversion relationship between the image coordinates and the actual angle is calculated. Based on the sun's position in the image and the camera's intrinsic parameters (such as focal length, sensor size, etc.), the angle of the sun in physical space is calculated using trigonometric functions:
[0056]
[0057] where x is the x-coordinate of the sun in the image, c x is the center coordinate of the image, f x is the focal length of the camera.
[0058] Calculate the robot's position and orientation and correct it using the sun's physical angle to obtain an initial estimate and perform calibration to obtain calibrated position information.
[0059] As a preferred embodiment, the process of calculating the robot's position and orientation and correcting them by the sun's physical angle to obtain an initial estimation result includes:
[0060] Based on the robot's motion information, the robot's position and orientation are calculated through integration. The robot's relative displacement and rotation between two frames are calculated by analyzing the matching of feature points in consecutive image frames. When the robot starts, the robot's initial orientation relative to the sun is determined by the solar physical angle. During the robot's movement, the robot's orientation is corrected in real time by updating the solar physical angle information.
[0061] The corrected robot position and orientation are fused with the relative displacement and rotation of the robot between the two frames. The error in the fusion result is corrected using the solar physical angle information as an external reference to obtain the initial estimation result.
[0062] As a preferred embodiment, the process of obtaining the calibration position information includes: acquiring sensor data, performing error analysis on the initial estimation result based on the sensor data, calibrating the initial estimation result according to the error analysis result, and obtaining the calibration position information.
[0063] Specifically, the odometry method and visual odometry are used to calculate the position and orientation of the positioning robot. The specific calculation method is as follows:
[0064] Odometry is a method that uses the robot's motion information (gyroscope) to estimate its position and orientation. By knowing the robot's initial position and its speed, acceleration, rotation angle, and other information at each moment, the robot's current position and orientation can be continuously updated.
[0065] The robot's position is inferred using the robot's velocity at each time step:
[0066] x t+1 =x t +v·Δt·cos(θ t );
[0067] y t+1 =y t +v·Δt·sin(θ t );
[0068] Where v is the robot speed, θt is the current heading angle, Δt is the time step, x t 、y t is the current position, x t+1 、y t+1 is the next position after the update.
[0069] Use the gyroscope sensor (IMU) to obtain the orientation change angle and update the robot's heading angle.
[0070] θ t+1 =θ t +Δθ t ;
[0071] Where Δθ t is the change in angle during the current time step.
[0072] Visual odometry uses image sequences captured by a camera, combined with image processing algorithms (such as SIFT, SURF, and ORB) to extract feature points. It then matches features between consecutive image frames and calculates the robot's relative displacement between frames. Using these matched feature points, combined with the camera's motion model, the robot's three-dimensional trajectory is estimated, thereby inferring its position and rotation within the environment, enabling autonomous positioning and navigation.
[0073] The current position and orientation calculated by the positioning algorithm are combined with the data fed back by the sensors to optimize and calibrate the robot's current position error and orientation error.
[0074] If the robot has a known target position or expected trajectory (for example, from path planning or through reference positions provided by other sensors), it can be determined whether the localization error is too large by calculating the error between the current estimated position and the target position:
[0075]
[0076] The position error represents the difference between the robot's estimated position and the target position, measured in Euclidean distance, x estimated Indicates the x coordinate of the robot's current position, y estimated Indicates the y coordinate of the robot's current position, x target The x coordinate of the target position (real position or reference position), y target The y coordinate of the target position (real position or reference position).
[0077] The orientation error is determined by calculating the angular difference between the robot's current orientation and the expected orientation:
[0078] Δθ=|θ estimated -θ target |;
[0079] Among them, Δθ represents the robot orientation error, usually expressed as an angle difference, θ estimated The robot's estimated heading angle, usually refers to the robot's current heading or direction, θ target Indicates the target orientation angle, usually referring to the reference direction or target direction that the robot should face.
[0080] The calibration position information is checked for consistency. If the consistency check passes, the robot will perform self-calibration and correct the real-time positioning result of the robot in real time according to the calibration position information and sensor data. If the check fails, the robot will not perform self-calibration and realize autonomous navigation of the robot according to the real-time positioning result.
[0081] As a preferred embodiment, the self-calibration process includes:
[0082] The data information of the visual camera and the inertial unit are fused, and the error is corrected using the extended Kalman filter.
[0083] Specifically, if the consistency check fails, that is, by comparing the current robot position and the sensor data, the error between the two is large, it means that there is a large error between the current position and the expected position, and it is necessary to recalculate the calibration position information after self-calibration; if the consistency check passes, no calibration is required and the next step can be directly performed.
[0084] The robot performs self-calibration, fuses the data information of the visual camera and the inertial unit (IMU), uses the extended Kalman filter (EKF) to correct errors, and locates the current position and orientation of the robot.
[0085] The Extended Kalman Filter (EKF) predicts the current position and orientation based on the position and velocity at the previous moment:
[0086]
[0087] in, is the predicted robot state (position and orientation), u t is the current control input (speed, steering angle, etc.).
[0088] Then update the predicted state through visual information:
[0089] K t =P t|t-1 H T (HP t|t-1 H T +R) -1 ;
[0090] Among them, P t|t-1is the predicted covariance matrix, H is the measurement matrix, R is the measurement noise matrix, K t is the Kalman gain.
[0091] The robot's position, orientation, and multiple sensor data are corrected in real time to ensure that the robot always maintains high-precision positioning during operation.
[0092] The positioning information after positioning, sensor data fusion and self-calibration is output to the control system to support the robot's decision-making and navigation.
[0093] According to the output positioning information and task objectives, the actual motion instructions of the robot are executed to ensure that the robot can complete the task autonomously (such as path planning, obstacle avoidance, target navigation, etc.).
[0094] like Figure 2 As shown, this embodiment also provides an unstructured environment wheel-leg hybrid quadruped robot autonomous navigation system based on celestial body perception, including a visual control module and a motion control module.
[0095] The visual control module includes the ORB-SLAM3 mapping module and the celestial body recognition and positioning module.
[0096] The ORB-SLAM3 mapping module includes key frame extraction, construction of local maps and formation of prior map models. It extracts key nodes from the data stream collected by the quadruped vision camera, builds an environmental map model and fuses the local map model, and finally outputs the position and posture data set of the quadruped robot.
[0097] The ORB-SLAM3 mapping module performs feature point detection and tracking in the following steps:
[0098] The image sequence captured by the visual camera is input into the ORB-SLAM3 system; the acceleration and angular velocity data are obtained from the IMU sensor, and preprocessed such as filtering denoising and zero bias correction are performed, and time synchronization is performed with the image input data; ORB feature points are extracted for each frame of the image. These feature points are representative and distinguishing points in the image, such as corner points, edge points, etc.; the feature points of the current frame are matched with the feature points of the previous frames to find the corresponding positions of the same feature points in different frames; according to the preset rules, key frames are selected from continuous frames. Key frames are important references for building maps and performing pose estimation; based on the matched feature point pairs, the triangulation principle and motion estimation algorithm are used to calculate the pose changes of the robot in three-dimensional space, that is, position and posture; in the initial stage, when enough feature points are tracked and matched, an initial map is constructed and some feature points are added to the map as map points.
[0099] The backend receives keyframes and their corresponding feature points, pose estimation information, and IMU preprocessing data from the frontend for pose estimation and map construction. The process includes:
[0100] (1) Posture optimization: Combined with IMU data, using optimization algorithms such as Bundle Adjustment, the robot's posture estimation is optimized based on the correspondence between feature points between key frames and the motion constraints of the IMU to improve positioning accuracy.
[0101] (2) Map construction: Add new feature points to the map to build and update the map representation of the environment.
[0102] (3) Loop detection: By comparing the similarity between the current keyframe and the historical keyframes in the map, combined with IMU data to assist in judgment, check whether the robot has returned to the location it has visited before.
[0103] (4) Solar celestial body positioning data fusion: The position information provided by the solar celestial body positioning module is used as a global position reference and integrated with the robot posture in the back-end map to further correct the robot's position and posture in the global coordinate system and reduce positioning deviations caused by environmental changes or sensor errors.
[0104] (5) Global optimization: If a loop is detected, global optimization is performed in combination with IMU information to correct the accumulated errors in the map and ensure the consistency and accuracy of the map.
[0105] (6) Map update and maintenance: Continuously update and maintain maps, remove unreliable feature points, and optimize the structure and representation of maps.
[0106] The celestial body recognition and positioning module includes image processing, sun position recognition, sun angle calculation, combination of visual odometry and IMU, and correction of robot position estimation. It pre-processes the collected images, uses Hough circle transform to locate the position of the sun in the image, calculates the sun altitude angle through trigonometric functions, and outputs data set files of visual odometry and IMU, corrects the robot's position estimation, and finally outputs the robot position and attitude error that needs to be corrected.
[0107] The process of integrating solar object positioning into the orb-slam3 control framework involves:
[0108] (1) Acquisition and conversion of solar celestial body positioning data
[0109] Sun position recognition: Obtain the two-dimensional position information of the sun in the image from the front-end visual camera module.
[0110] Three-dimensional coordinate conversion: Using the camera's intrinsic and extrinsic parameters, the two-dimensional image coordinates of the sun are converted into three-dimensional space coordinates to obtain the position of the sun in the world coordinate system.
[0111] (2) Correlation between solar celestial body positioning data and robot posture
[0112] Time synchronization: Ensure that the solar celestial positioning data is synchronized with other sensor data (visual camera and IMU) processed by the ORB-SLAM3 backend.
[0113] Coordinate system alignment: Align the coordinate system of the solar celestial positioning data with the coordinate system used by the robot to facilitate subsequent fusion calculations.
[0114] (3) Integrate solar celestial body positioning data for global optimization
[0115] Optimization problem construction: In the global optimization of the ORB-SLAM3 backend, the poses of the image keyframes recognized by the mechanical camera and the positions of the map points are used as optimization variables. At the same time, the solar celestial position data is introduced as an additional observation value into the optimization problem.
[0116] Construct an error function: Define an error function to measure the difference between the robot's pose estimate and the solar celestial body positioning data. Specifically, the Euclidean distance and angular difference between the position of the sun in the robot's pose estimate and the actual position of the sun in the solar celestial body positioning data are calculated as the error term. The specific formula is as follows:
[0117]
[0118] Where: A and B coordinates are (x1, y1, z1) and (x2, y2, z2) respectively, A is the center of gravity of the camera, and B is the solar celestial body positioning coordinate.
[0119] By minimizing these errors, the robot's pose estimate is adjusted to be consistent with the solar celestial positioning data.
[0120] Fusion Optimization Framework: Combine the error terms of solar positioning data with the original error terms of ORB-SLAM3 (such as the reprojection error of visual feature points and the pre-integration error of the IMU) and optimize them as a fused error term. Use a graph optimization framework (such as g2o or Sophus) to implement this process, using various error terms as edges and optimization variables as vertices to build a graph model for optimization.
[0121] Optimization algorithm selection: Use the Gauss-Newton optimization algorithm to minimize the constructed error function, which iteratively adjusts the optimization variables to minimize the error function.
[0122] Pose Update: Through the optimization process, the robot's pose estimate is adjusted to make it more consistent with the solar position data and other sensor data. At the same time, the position of the map points is updated to ensure the accuracy and consistency of the map.
[0123] (4) Update map and pose estimation
[0124] Map point adjustment: Adjust the positions of feature points in the map based on the fused and optimized pose information. This includes removing unreliable feature points, adding new feature points, and optimizing the map structure and representation.
[0125] Pose correction: Use the optimized pose estimation to correct the robot's position and posture in the global coordinate system, reduce positioning deviations caused by environmental changes or sensor errors, and improve the accuracy and reliability of the robot's pose estimation.
[0126] (5) Continuous integration and updating
[0127] Loop processing: As the robot continues to move and collect new data, the above steps are repeated continuously, integrating the new solar celestial body positioning data into the pose estimation and map update, ensuring that the robot can maintain high-precision positioning and map construction even during long-term operation and in complex environments.
[0128] Dynamic Adjustment: The fusion algorithm parameters are dynamically adjusted based on environmental changes and sensor data quality. For example, in situations with significant lighting changes, the weight of solar positioning data can be increased; in environments with many dynamic obstacles, the influence of IMU data can be increased to maintain positioning accuracy and robustness.
[0129] The motion control module includes WBC motion control and MPC model prediction, which are used to control the robot body movement and predict the robot's position and posture at the next moment, respectively, and finally output the torque parameter file required by the motor to achieve foot-end control of the quadruped robot.
[0130] This embodiment further provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.
[0131] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.
[0132] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.
Claims
1. A method for autonomous navigation of a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception, characterized in that: The following steps are involved: Acquire an environmental image, obtain the initial position coordinates of the sun based on the environmental image and perform corrections to obtain the corrected position coordinates of the sun; and obtain the physical angle of the sun based on the corrected position coordinates of the sun; Calculate the position and orientation of the robot and correct it by the physical angle of the sun to obtain an initial estimation result and perform calibration to obtain calibration position information; The calibration position information is checked for consistency. If the consistency check is passed, the robot performs self-calibration and corrects the real-time positioning result of the robot in real time according to the calibration position information and sensor data. If the check fails, self-calibration is not performed and autonomous navigation of the robot is achieved according to the real-time positioning result.
2. The autonomous navigation method of a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception according to claim 1 is characterized in that: The process of obtaining the initial sun position coordinates and making corrections includes: After preprocessing the environmental image, edge detection is used to obtain edge information of the environmental image. Based on the edge information, Hough circle transform is used to detect circular edges in the environmental image to obtain circular parameters. Target detection and position extraction are performed based on the circular parameters to obtain the initial position coordinates of the sun. The initial position coordinates of the sun are corrected by comparing and calculating multiple images to obtain the corrected position coordinates of the sun.
3. The autonomous navigation method of a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception according to claim 2 is characterized in that: The preprocessing includes color conversion, threshold segmentation and image denoising.
4. The autonomous navigation method of a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception according to claim 1 is characterized in that: The process of obtaining the solar physical angle based on the sun-corrected position coordinates includes: The corrected solar position coordinates are compared with the preset model. If successfully identified, the camera's field of view angle, image resolution, and camera internal parameters are extracted. The conversion relationship between the image coordinates and the actual angle is calculated in combination with the corrected solar position coordinates to obtain the solar physical angle.
5. The autonomous navigation method of a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception according to claim 1 is characterized in that: The process of calculating the robot's position and orientation and correcting them by the sun's physical angle to obtain an initial estimate includes: Based on the robot's motion information, the robot's position and orientation are calculated through integration. The robot's relative displacement and rotation between two frames are calculated by analyzing the matching of feature points in consecutive image frames. When the robot starts, the robot's initial orientation relative to the sun is determined by the solar physical angle. During the robot's movement, the robot's orientation is corrected in real time by updating the solar physical angle information. The corrected robot position and orientation are fused with the relative displacement and rotation of the robot between the two frames. The error in the fusion result is corrected using the solar physical angle information as an external reference to obtain the initial estimation result.
6. The autonomous navigation method of a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception according to claim 1 is characterized in that: The process of obtaining the calibration position information includes: acquiring sensor data, performing error analysis on the initial estimation result based on the sensor data, calibrating the initial estimation result according to the error analysis result, and obtaining the calibration position information.
7. The autonomous navigation method of a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception according to claim 1 is characterized in that: The self-calibration process includes: The data information of the visual camera and the inertial unit are fused, and the error is corrected using the extended Kalman filter.
8. An autonomous navigation system for a wheel-leg hybrid quadruped robot in an unstructured environment based on celestial body perception, characterized in that: Including visual control module and motion control module; The visual control module includes the ORB-SLAM3 mapping module and the celestial body recognition and positioning module; The ORB-SLAM3 mapping module is used to obtain the position and posture data set of the quadruped robot; The celestial body recognition and positioning module is used to obtain the robot position and attitude errors that need to be corrected; The motion control module includes WBC motion control and MPC model prediction; WBC motion control is used to control the robot's body motion; The MPC model prediction is used to predict the position and posture of the robot at the next moment, and finally output the torque parameter file required by the motor to achieve foot-end control of the quadruped robot.
9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.