Omnidirectional mobile robot pose adjustment method based on laser radar

Abstract

The invention discloses an omni-directional mobile robot pose adjustment method based on a laser radar, and the method comprises the steps: collecting and processing multi-frame laser radar data through a system to extract environment features in a working point creation stage after a robot is precisely guided to a target pose, and carrying out the correlation storage of a working point identifier, a global pose and standard feature data after verification. And when the robot navigates to a position near a working point, triggering an accurate adjustment process: calling pre-stored standard features by a system, collecting laser radar data in real time to perform same filtering and feature extraction processing, and performing iterative calculation through a point cloud matching algorithm to obtain X and Y direction and course angle deviations between a current position and a standard position. And finally, carrying out pose correction, and carrying out real-time cyclic feedback until the deviation reaches a millimeter-level precision threshold value. According to the invention, the dependence of a traditional two-dimensional code accurate positioning scheme on illumination, cleanliness and entity marking is overcome, and the method has the advantages of high precision, strong environmental adaptability and high reliability.

Description

Technical Field

[0001] This application relates to the field of laser positioning and navigation technology, specifically to a method for adjusting the pose of an omnidirectional mobile robot based on lidar. Background Technology

[0002] In modern warehousing and logistics systems, the widespread application of automated warehousing robots has greatly improved the efficiency, accuracy, and intelligence of warehousing operations. Precise positioning and posture adjustment of robots within the warehousing environment are core prerequisites for ensuring their efficient completion of tasks such as goods handling, sorting, and storage. After a logistics robot operating in a warehousing environment reaches its work point using 2D laser navigation algorithms, its navigation accuracy is generally insufficient to meet the requirements for docking with shelving. Therefore, after navigating to the work point, it usually requires secondary precise adjustments to the robot's posture based on external sensor data to achieve docking with the shelving and complete tasks such as loading or unloading.

[0003] Currently, in the field of warehouse robot positioning, positioning solutions based on QR codes and cameras are widely used. The core principle of this type of solution is to pre-lay or set QR code tags on the warehouse floor or specific markers. The robot's camera then collects the QR code image information, calculates the robot's pose deviation, and makes adjustments. For example, in patent application number 2022104198834, entitled "A Position Adjustment Method Based on QR Code Recognition," the robot's pose deviation is calculated by taking a picture of the QR code. From a hardware perspective, this solution requires additional cameras and auxiliary lighting systems, increasing hardware costs and affecting the structural design of other systems. From an environmental adaptability perspective, this solution is highly dependent on the warehouse environment's lighting conditions, cleanliness, and the integrity of the QR code tags. When there is insufficient light intensity, direct strong light, or uneven lighting in the warehouse environment, the QR code images captured by the camera are prone to blurring and distortion, causing the image recognition algorithm to fail to accurately interpret the QR code information, thus leading to positioning failure. At the same time, dust and debris generated during warehouse operations can easily adhere to the surface of the QR code label, or the QR code label can be worn or scratched during robot operations, all of which will damage the integrity of the QR code label, making it impossible for the camera to effectively capture a clear and complete QR code image, further reducing the reliability and stability of positioning. Summary of the Invention

[0004] This application addresses the shortcomings of current QR code and camera-based warehouse robot positioning solutions in terms of environmental adaptability, positioning accuracy, flexibility, and anti-interference capabilities. It provides a 2D laser high-precision positioning and pose adjustment method for warehouse environments. This method enables stable and accurate secondary positioning and pose adjustment of robots in complex warehouse environments, meeting the demands of modern warehousing and logistics for high-precision, high-reliability, and high-flexibility robot operations. The specific technical solution is as follows: A method for adjusting the pose of an omnidirectional mobile robot based on lidar includes the following steps: Step S1: Work point information creation. Move the robot to the target work point via remote control or automatic navigation, adjust the robot pose to meet the operation requirements, and then collect and save the standard LiDAR feature data and corresponding global pose information of the work point. Step S2: Navigation and positioning. The robot plans a path according to the task and navigates to the positioning adjustment area near the target work point. Step S3: Real-time data acquisition and feature extraction. Within the positioning adjustment area, the robot's current real-time LiDAR data is acquired and processed using the same filtering and feature extraction algorithms as those used to create the working point, to obtain the current LiDAR feature data. Step S4: Pose deviation calculation. Call the standard LiDAR feature data of the target working point saved in step S1, and match the current LiDAR feature data with the standard LiDAR feature data through a point cloud matching algorithm to calculate the pose deviation between the robot's current position and the standard position of the target working point. The point cloud matching algorithm calculates the optimal alignment between two frames of point clouds to accurately measure the minute deviation between the robot's current pose and the target pose.

[0005] The process of achieving matching and calculating deviations can be summarized into the following three key steps: 1. Feature Comparison: Instead of directly comparing all the original data points, it compares environmental features extracted from the current real-time point cloud and the standard point cloud, such as long straight lines and right angles on walls and metal shelves. This is equivalent to comparing the "contour lines" within the point set rather than the "pixels" of the entire point set.

[0006] 2. Iterative Optimization: The algorithm performs multiple iterative calculations. Each iteration attempts to make small translations (X, Y directions) and rotations (heading angles) on the current point cloud, and then recalculates the overall difference between it and the standard point cloud features (usually measured by the sum of "point-to-line" distances). This process is similar to continuously adjusting the angle and position of a map in hand to make it perfectly match the actual environment.

[0007] 3. Output: When the algorithm finds the transformation that minimizes the difference in feature between two frame point clouds through iteration, it finds the optimal solution. This optimal transformation amount—that is, how much to move in the X direction, how much to move in the Y direction, and how much to rotate—calculates the pose deviation (Δx, Δy, Δθ). These deviation values ​​are sent to the robot's control system to drive it to perform precise pose adjustments.

[0008] Step S5: Pose adjustment. Based on the pose deviation calculated in step S4, a control command is generated using an attitude control algorithm to drive the robot to move in order to correct the pose. Steps S3 to S5 are repeated until the pose deviation is less than a preset accuracy threshold.

[0009] The attitude control algorithm receives the pose deviation (Δx, Δy, Δθ) calculated by the point cloud matching algorithm, and then decides how the robot should move (i.e., generates speed control commands) to eliminate this deviation.

[0010] The process of generating control commands includes the following three steps: 1. Deviation Input: The algorithm receives three key deviation values ​​as input: X-direction deviation (Δx): The displacement error of the robot in the forward and backward direction.

[0011] Y-direction deviation (Δy): The displacement error of the robot in the left and right directions.

[0012] Heading angle deviation (Δθ): The angular difference between the robot's current orientation and the target orientation.

[0013] 2. Control Law Calculation (Taking PID Algorithm as an Example): Proportional (P) Control: Generates a speed proportional to the magnitude of the deviation. The larger the deviation, the faster the robot moves.

[0014] Integral (I) control: Accumulates historical deviations to eliminate small, persistent steady-state errors.

[0015] Differential (D) control: It senses the trend of deviation change (i.e., the rate of change) and acts as a "damper" to prevent the robot from over-adjusting and causing oscillations.

[0016] The three deviations, Δx, Δy, and Δθ, are input into three independent PID controllers and calculated in parallel.

[0017] 3. Speed ​​Command Output: After PID calculation, the algorithm outputs the final control quantity, which is the speed command required for the robot's omnidirectional movement chassis. X-axis velocity: used to correct displacement deviation (Δx) in the forward and backward directions.

[0018] Y-axis velocity: used to correct displacement deviation (Δy) in the left and right directions.

[0019] Z-axis rotation speed: used to correct heading angle deviation (Δθ).

[0020] Step S1 also includes: S1.1: Set and verify the unique identifier information of the work point; S1.2: When the robot is stationary, acquire multiple frames of raw LiDAR data at preset intervals; S1.3: Perform point cloud filtering and feature extraction on the multi-frame raw LiDAR data to obtain standard LiDAR feature data containing environmental feature information, including contours such as long straight lines and right angles; S1.4: Calculate the pose deviation of the multi-frame data. If the deviation result is less than the static determination threshold, it is determined that the robot is in a valid static state when collecting data, and S1.5 is executed. S1.5: Store the unique identifier of the working point, the standard lidar feature data, and the corresponding global pose association in the database.

[0021] In step S1.4, a point cloud matching algorithm is used to calculate the pose deviation between the multiple frames of data. The point cloud matching algorithm is one of the PLICP algorithm, ICP algorithm, or NDT algorithm.

[0022] The ICP algorithm is an iterative nearest point algorithm. In each iteration: find the corresponding point: for each point in the current point cloud, find the nearest point in the target point cloud as the corresponding point; calculate the transformation: calculate a pose transformation (rotation and translation) that minimizes the sum of squared distances between all corresponding points; apply the transformation: apply this transformation to the current point cloud; iterate: repeat the above steps until convergence.

[0023] The PLICP algorithm is a point-to-line iterative nearest point algorithm, an improvement on ICP. Its main difference from ICP lies in the first step: instead of finding the nearest point in the current point cloud, it finds the line segment containing the point in the target point cloud and calculates the perpendicular distance from the point to that line segment.

[0024] The NDT algorithm is a normal distribution transformation algorithm. It does not directly handle the relationship between points or between points and lines. Instead, it divides the space where the target point cloud is located into multiple grids (voxels). For each grid, it calculates the statistical characteristics of the point cloud inside it and uses a multidimensional normal distribution (i.e., probability distribution) to approximate the distribution of points in the grid. The matching process is as follows: the current point cloud is substituted into this probability distribution field, and an optimization algorithm is used to find a pose transformation that maximizes the overall probability of the current point cloud appearing in the probability field of the target point cloud.

[0025] The point cloud matching algorithm in step S4 is the PLICP algorithm, and the specific steps include: Define a pose transformation model for the robot, the model including translation (tx, ty) and rotation angle θ; Model the straight line features in the standard lidar feature data to obtain the direction vector and normal vector of the straight line; Construct a distance error function from the current feature data point to the target feature line after pose transformation; The error function is minimized by the Gauss-Newton iterative algorithm to solve for the optimal pose transformation parameters, thereby obtaining the X-direction deviation Δx, Y-direction deviation Δy, and heading angle deviation Δθ.

[0026] The workflow steps of the Gauss-Newton iterative algorithm include: 1. Establish and initialize the target: The goal of the algorithm is to minimize the error function, which is the sum of squared distances from all "current points" to their "corresponding target lines". Initially, the positioning pose parameters are given an initial guess, for example, assuming no deviation, i.e., (Δx,Δy,Δθ)=(0,0,0).

[0027] 2. Iterative optimization: Linearization approximation: Near the current parameter guesses, a complex nonlinear error function is approximated by a simple "tangent" (linear model). This greatly simplifies the problem.

[0028] Solving for the correction factor: The algorithm calculates an optimal correction factor (δΔx, δΔy, δΔθ) for the current parameter guess. This correction factor is in the direction that causes the error function value to decrease the fastest.

[0029] Update parameters: Apply the correction to the current parameters to obtain a new set of more accurate parameter guesses.

[0030] New parameter = Old parameter + Correction amount 3. Convergence judgment: Check whether the new parameters are sufficiently optimized. The criteria for judgment typically include: If the correction is very small (e.g., less than 0.1 mm or 0.001°), it indicates that the parameters have basically stabilized.

[0031] Does the error value no longer decrease significantly?

[0032] Check if the number of iterations has reached a preset limit (e.g., 100 times) to prevent infinite loops.

[0033] If the convergence condition is met, the current optimal pose parameters are output as the result; otherwise, the new parameter values ​​are used to return to step 2 and continue the iteration.

[0034] The PLICP algorithm's iterative process sets an upper limit for the number of iterations and a convergence threshold. The iteration terminates when the number of iterations reaches the upper limit or the pose parameter increment norm is less than the convergence threshold.

[0035] The attitude control algorithm in step S5 is a PID control algorithm. The pose deviation is used as the input of the PID control algorithm, and the output is the robot's X-axis speed, Y-axis speed and Z-axis rotation speed control commands.

[0036] In step S3, when collecting real-time lidar data, the lidar's scanning angle range and scanning frequency parameters are consistent with the parameters when collecting standard data in step S1.

[0037] The accuracy thresholds include an X-direction deviation threshold, a Y-direction deviation threshold, and a heading angle deviation threshold. The X-direction deviation threshold and the Y-direction deviation threshold are both 5 mm, and the heading angle deviation threshold is 0.5°.

[0038] In step S1.2, the specific method for obtaining multiple frames of raw LiDAR data is as follows: first, obtain one frame of LiDAR data, wait for 1 second, and then obtain the next frame of LiDAR data.

[0039] The specific implementation involves initiating the robot's global localization program. The robot is moved to the vicinity of the target work point automatically or remotely. The remote controller precisely adjusts the robot's working posture until it meets the docking requirements with the shelf. The program then saves the current work point's posture on the global map. Two frames of raw 2D LiDAR data are acquired at one-second intervals, containing spatial information such as distances and angles of objects in the surrounding environment. The raw data is then processed using filtering and feature extraction algorithms to extract features commonly found in warehouse environments, such as long straight lines and right-angled contours. The two frames of LiDAR data after feature extraction are used to calculate the posture deviation using a point cloud matching algorithm. If the result is less than a set threshold, it indicates that the robot is stationary and the obtained LiDAR feature data is reliable. Finally, the robot's control system associates the extracted 2D LiDAR data with the corresponding work point identification information and stores it in the robot's local storage module or a cloud database connected to the robot for subsequent local localization.

[0040] According to the warehouse operation task plan, the robot's navigation system generates a navigation path from its current position to the target work point. The robot moves autonomously along the navigation path, gradually approaching the target work point. When the robot navigates to the vicinity of the target work point and determines that it has entered the positioning adjustment area, it executes a precise pose adjustment program. First, it retrieves the 2D LiDAR feature data corresponding to the target work point from the storage module or cloud database. Then, it collects the robot's current real-time 2D LiDAR data and uses the same algorithm for feature extraction. During the acquisition process, it ensures that the LiDAR scanning parameters (such as scanning angle range, scanning frequency, etc.) are consistent with the parameters when acquiring the original data to ensure the comparability of the two acquisitions and reduce positioning errors caused by parameter differences. The current real-time 2D LiDAR data after feature extraction and the original data after feature extraction are input into a pre-built point cloud matching algorithm model. The point cloud matching algorithm performs iterative calculations to optimally match the point cloud in the current LiDAR data with the point cloud in the original LiDAR data. The algorithm calculates the positional deviation between the robot's current position and the standard position of the target work point, including the x-axis deviation, y-axis deviation, and heading angle deviation. Based on the deviation calculation results, a speed control command is generated using an attitude control algorithm to move the robot for attitude correction. Finally, it is determined whether the robot's pose deviation meets the accuracy requirements. If not, the pose deviation is recalculated and adjusted in real time; if it meets the requirements, the pose adjustment task ends.

[0041] This application has the following beneficial effects: 1. Strong environmental adaptability: This invention is based on 2D LiDAR for data acquisition and positioning. The LiDAR is not affected by the lighting conditions or cleanliness of the warehouse environment. Even in dim lighting, direct sunlight, or environments with a lot of dust and debris, it can still stably acquire clear and accurate environmental data, effectively avoiding the positioning failure problem caused by camera images being blurry or distorted. At the same time, it does not rely on QR code tags, and there is no situation where tag wear or damage affects positioning, which greatly improves adaptability in complex warehouse environments.

[0042] 2. High positioning accuracy: First, features are extracted from the laser data, and then a point cloud matching algorithm is used to match the laser data. After processing by the feature extraction algorithm, laser point cloud noise can be effectively filtered out, improving the data matching accuracy and thus achieving millimeter-level positioning accuracy.

[0043] 3. High flexibility: This invention does not require the installation of QR code labels in the storage area. When the layout of the storage racks is adjusted, it is only necessary to re-collect the original 2D LiDAR data of each work point after the adjustment. There is no need to remove and re-install the labels, which reduces operating costs and workload, greatly improves the flexibility of the storage area layout adjustment, and adapts to the dynamic changes in the storage operation scenario.

[0044] 4. Strong anti-interference capability: LiDAR detects the environment by emitting laser beams, and is not affected by visual interference factors such as shelf reflections and product packaging colors in the warehouse environment. It can collect environmental data stably and accurately, ensuring the stability and reliability of the positioning process and avoiding the problem of increased positioning error or positioning failure caused by visual interference.

[0045] 5. Good real-time performance: Compared with image recognition algorithms, the computational complexity of feature extraction and point cloud matching algorithms is lower and the data processing speed is faster. It can process the data collected by 2D LiDAR in real time, quickly calculate the position deviation and complete the pose adjustment, effectively avoiding the problem of lagging positioning results and meeting the real-time requirements of high-speed operation of warehouse robots. Attached Figure Description

[0046] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0047] Figure 1 This is a schematic diagram of the overall implementation process of the method provided in the embodiments of this application.

[0048] Figure 2 This is a flowchart illustrating the procedure for creating a working point as described in the embodiments of this application.

[0049] Figure 3 This is a flowchart illustrating the precise pose adjustment procedure in the embodiments of this application. Detailed Implementation

[0050] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0051] The scheme of this application will be further described below with reference to the accompanying drawings.

[0052] This embodiment uses an automated handling robot in a warehouse as an example. This warehouse has densely packed shelves and a wide variety of goods. The working environment presents problems such as uneven lighting, glare from shelf surfaces, and dust generation during operations. Traditional positioning solutions based on QR codes and cameras often suffer from positioning failures, insufficient accuracy, and slow adjustment speeds, affecting operational efficiency. This embodiment employs the high-precision pose adjustment method for omnidirectional mobile robots based on 2D LiDAR provided by this invention, achieving precise pose adjustment of the robot in this warehouse environment.

[0053] refer to Figure 1 This invention provides a method for adjusting the pose of an omnidirectional mobile robot based on lidar, the method comprising the following steps: Step S1: Workpoint Creation Process.

[0054] First, a global warehouse map needs to be created, and a global positioning algorithm based on LiDAR, IMU, and wheel odometry needs to be run. The robot is moved to the vicinity of the work point via automatic navigation or remote control. Then, the remote control speed is switched to the lowest level to precisely adjust the robot's pose until it meets the docking requirements with the shelf. The robot work point addition program is run to add a new work point and record the current global pose and LiDAR feature data.

[0055] The wheeled odometer directly calculates the robot's relative displacement and turning angle (heading angle change) over a short period by measuring the number of wheel rotations. The IMU (Inertial Measurement Unit) measures the robot's angular velocity rate and acceleration in three-dimensional space, and obtains attitude angle and velocity changes through integration. The lidar measures precise distance information of the robot's surrounding environment by emitting laser beams and receiving their reflections, generating a point cloud map. The data from these three sources are fused. LiDAR positioning provides high-precision absolute pose as a calibration benchmark. When laser matching is successful, the algorithm uses this as a reference to reset the accumulated errors of the wheeled odometer and IMU, preventing positioning "drift." During the interval between two lidar scans, the algorithm relies on the IMU and wheeled odometer to provide high-frequency, continuous pose prediction and updates. The IMU provides accurate rotational information, compensating for the wheeled odometer's shortcomings in measuring steering.

[0056] like Figure 2 As shown, the robot workpoint creation procedure described in this embodiment of the invention includes the following steps: Step S1.1: Set the work point ID and name.

[0057] Enter the current work point ID and name. The system checks if the work point ID and name already exist in the database. If they do, the user is prompted to re-enter them; otherwise, proceed to step S1.2.

[0058] Step S1.2: Acquire lidar data.

[0059] Acquire one frame of LiDAR data, wait 1 second, and then acquire another frame of LiDAR data.

[0060] Step S1.3: Point cloud filtering and feature extraction.

[0061] The two frames of LiDAR data are processed using a feature extraction algorithm. In this embodiment, sparse laser points are first filtered out, and then a feature extraction algorithm is used to extract feature data such as long straight lines and right angles from the point cloud.

[0062] Step S1.4: Calculate pose deviation.

[0063] A point cloud matching algorithm is used to calculate the pose deviation of the processed two frames of data. Specific parameters include heading angle deviation, X-direction deviation, and Y-direction deviation. If all parameter results are less than a set threshold, the robot is considered to be currently stationary, and the obtained radar feature data is valid; otherwise, proceed to step S1.2. In this embodiment, the point cloud matching algorithm uses a pre-constructed PLICP (Point-to-Line Iterative Closest Point) matching algorithm to optimally match the point cloud in the current laser data with the point cloud in the original laser data. Utilizing the point-to-line distance metric, the algorithm effectively reduces the impact of laser point cloud noise on the matching results, improving matching accuracy. The point cloud matching algorithm can be any matching algorithm based on ICP, NDT, or an improved version thereof.

[0064] Step S1.5: Save the working point data.

[0065] Associate the current working point ID, name, global pose, and LiDAR feature data, and store them in a local or server database.

[0066] Step S2: Navigate to the designated work point.

[0067] The warehouse management system issues a work task, requiring the robot to move from its current docking point (initial position) to the work point to perform the goods handling task. The robot's navigation system is based on a pre-built warehouse environment map and uses A / B navigation. A path planning algorithm generates a navigation path. The robot moves autonomously according to the navigation path. During the movement, it calculates its global position information in real time by fusing data from LiDAR, IMU, and wheel odometry to achieve path tracking, and finally reaches the positioning adjustment area near the working point (a circular area with a radius of 10cm centered on the working point). At this point, the robot's initial position deviates from the standard position of the working point. (The A...) The path planning algorithm steps include initialization: the algorithm uses the robot's starting position and the target working point position as the starting point and target point of the search.

[0068] Map processing: The algorithm abstracts the environment into a network of nodes (i.e., grids) based on a pre-built warehouse environment map (usually a raster map in which obstacles such as shelves and walls are marked as impassable areas).

[0069] Path search: The algorithm starts from the starting point and calculates the f(n) value of all its traversable adjacent nodes. It always selects the node with the smallest f(n) value as the next step and adds it to a "candidate list of optimal paths" (called the open list).

[0070] Repeat this process, expanding outwards from the starting point like a ripple, and continuously updating the shortest path cost of each node until the search ripple reaches the target point.

[0071] Path generation: Once the target point is reached, the algorithm can reconstruct the shortest, collision-free continuous path from the starting point to the end point by tracing back the parent node of each node from the target point.

[0072] Step S3: Collect and process the lidar data at the current location.

[0073] like Figure 3 As shown, the latest frame of LiDAR data is obtained, and the same feature extraction algorithm is used to process the LiDAR data.

[0074] Step S4: Calculate the positional deviation using a point cloud matching algorithm.

[0075] like Figure 3 As shown, the 2D LiDAR feature data corresponding to the current working point is retrieved from the database, and the current feature data obtained in step S3 and the original feature data are input into the PLICP matching algorithm model. This algorithm model runs in the robot controller. In this embodiment, the specific parameters are set as follows: the maximum number of iterations is 100, the convergence threshold (position deviation change) is 0.1 mm, and the point cloud search radius is 0.1 m. Iterative calculations are performed to obtain the deviation between the current pose and the original pose. If the results are all less than the preset accuracy thresholds (e.g., x-axis and y-axis deviation thresholds of 5 mm, heading angle deviation threshold of 0.5°), the attitude adjustment task ends; otherwise, step S5 is executed.

[0076] Step S5: Robot pose adjustment.

[0077] like Figure 3 As shown, the deviation data obtained in step S4 is input into the attitude control algorithm to calculate the speed parameters for robot attitude adjustment and send speed control commands. In this embodiment, a PID control algorithm is used to construct the attitude control algorithm. The input parameters are the pose deviation (x-axis deviation, y-axis deviation, yaw angle deviation) between the current position and the original position, and the output parameters are the robot speed control variables (X-axis speed, Y-axis speed, Z-axis rotation speed). The attitude control algorithm can be any closed-loop control algorithm based on pose deviation.

[0078] This embodiment also provides a specific calculation method for the PLICP matching algorithm as follows: 1. Definition of Pose Transformation Model. Assume that the pose transformation of the robot in the 2D plane is determined by translation. and rotation angle The corresponding homogeneous transformation matrix is ​​formed. Represented as: For any point in the current point cloud set After transformation, we obtain ,Right now: 2. Linear Feature Modeling of Target Point Cloud. The linear features extracted from the target point cloud (original laser feature data) are defined using a two-point method: Let the linear feature be... Two adjacent feature points in the target point cloud and Determined, straight line Direction vector and normal vector They are respectively: straight line The implicit equation is: 3. Point-to-line distance error function. The error function of PLICP is defined as "the distance from the current point to the line after the transformation". to the target line The sum of squared distances, that is: Will Substitute the coordinates and expand, then combine the properties of the implicit equation of a straight line. It can be simplified to: Where n is the total number of feature points in the current laser point cloud being matched. unit normal vector The amount.

[0079] 4. Least squares optimization solution. This is achieved by minimizing the error function. Solving for optimal pose parameters Iterative optimization using the Gauss-Newton method: Step 1: Initial value setting: During the first iteration, the initial values ​​of the pose parameters are set to the result of the previous iteration (set to the initial value during the initial iteration). ); Step 2: Jacobian matrix of error with respect to parameters: for the error function about (Rotational Increment) (x-direction translation increment) Calculate the partial derivative of the (y-direction translation increment) and construct the Jacobian matrix. Due to the rotation angle The deviation is relatively small (it gradually converges during the iteration process), so a small angle approximation is used. , The elements of the Jacobian matrix are: in The partial derivatives are calculated as follows: Step 3: Solve the incremental equation: According to the Gauss-Newton method, the incremental equation... Satisfies the normal equation: in This represents the error vector. The increment is obtained by solving the above linear equations using Cholesky decomposition or QR decomposition. ; Step 4: Iterative Update and Convergence Judgment: Update pose parameters , , Recalculate the error function The iteration converges if any of the following conditions are met: Error change (In this embodiment) ),in, This is the total error value recalculated after updating the pose parameters in this iteration. ΔE represents the total error value of the previous iteration, and ΔE represents the absolute value of the change in total error between two adjacent iterations. Incremental norm (In this example) ); The number of iterations has reached the upper limit (set to 100 in this example).

[0080] Final pose deviation calculation. After iterative convergence, based on the optimal transformation matrix... Calculate the pose deviation of the current point cloud relative to the target point cloud: X-direction deviation: Y direction deviation: Heading angle deviation: In this embodiment, if <5mm <5mm and If the deviation is less than 0.5∘, then the pose deviation is determined to meet the accuracy requirements.

[0081] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the article or device that includes that element.

[0082] The above provides a detailed description of a lidar-based omnidirectional mobile robot pose adjustment method. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the lidar-based omnidirectional mobile robot pose adjustment method and its core ideas. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A laser radar-based omnidirectional mobile robot pose adjustment method, characterized by, The method comprises the following steps: Step S1: work point information creation, moving the robot to a target work point by remote control or automatic navigation, adjusting the robot pose to meet the work requirements, collecting and saving the standard laser radar feature data and the corresponding global pose information of the work point; Step S2: navigation positioning, the robot plans a path according to the work task and navigates to a positioning adjustment area near the target work point; Step S3: real-time data acquisition and feature extraction, in the positioning adjustment area, the current real-time laser radar data of the robot is collected, and the same filtering and feature extraction algorithm as the work point creation is used for processing to obtain the laser radar feature data of the current position; Step S4: pose deviation calculation, calling the standard laser radar feature data of the target work point saved in step S1, matching the current laser radar feature data with the standard laser radar feature data through a point cloud matching algorithm, and calculating the pose deviation between the current position of the robot and the standard position of the target work point; Step S5: pose adjustment, according to the pose deviation calculated in step S4, generating control instructions by using a pose control algorithm to drive the robot to move to correct the pose, and repeating steps S3 to S5 until the pose deviation is less than a preset precision threshold.

2. The laser radar-based omnidirectional mobile robot pose adjustment method according to claim 1, characterized in that, The step S1 further comprises: S1.1: setting and verifying the unique identification information of the work point; S1.2: acquiring multiple frames of laser radar original data at intervals of a preset time when the robot is in a stationary state; S1.3: performing point cloud filtering and feature extraction on the multiple frames of laser radar original data to obtain standard laser radar feature data containing environmental feature information, and the environmental features include long straight lines and right angle contour features of walls and metal shelves; S1.4: calculating the pose deviation of the multiple frames of data, if the deviation result is less than a stationary judgment threshold, it is judged that the robot is in an effective stationary state when collecting data, and S1.5 is executed; S1.5: storing the unique identification information of the work point, the standard laser radar feature data and the corresponding global pose information into a database. 3.The laser radar-based omnidirectional mobile robot pose adjustment method of claim 2, wherein, In the step S1.4, the point cloud matching algorithm is used to calculate the pose deviation between the multiple frames of data, and the point cloud matching algorithm is one of PLICP algorithm, ICP algorithm or NDT algorithm.

4. The laser radar-based omnidirectional mobile robot pose adjustment method according to claim 1, characterized in that, The point cloud matching algorithm in step S4 is PLICP algorithm, and the specific steps comprise: defining a pose transformation model of the robot, the model comprising a translation amount (tx, ty) and a rotation angle θ; modeling the straight line features in the standard laser radar feature data to obtain the direction vector and normal vector of the straight line; constructing a distance error function of the current feature data points after pose transformation to the target feature straight line; minimizing the error function by using Gauss-Newton iterative algorithm to solve the optimal pose transformation parameters, thereby obtaining the X direction deviation Δx, Y direction deviation Δy and heading angle deviation Δθ.

5. The laser radar-based omnidirectional mobile robot pose adjustment method according to claim 4, characterized in that, The iteration process of the PLICP algorithm sets an upper limit of the number of iterations and a convergence threshold, and the iteration is terminated when the number of iterations reaches the upper limit or the pose parameter increment norm is less than the convergence threshold.

6. The laser radar-based omnidirectional mobile robot pose adjustment method according to claim 1, wherein, The pose control algorithm in the step S5 is a PID control algorithm, the pose deviation is taken as an input of the PID control algorithm, and outputs are X-axis speed, Y-axis speed and Z-axis rotation speed control instructions of the robot.

7. The laser radar-based omnidirectional mobile robot pose adjustment method according to claim 1, characterized by, In the step S3, when collecting real-time laser radar data, the scanning angle range and scanning frequency parameters of the laser radar are consistent with the parameters when collecting standard data in the step S1. 8.The laser radar-based omnidirectional mobile robot pose adjustment method of claim 1, wherein, The precision threshold value includes an X-direction deviation threshold value, a Y-direction deviation threshold value and a heading angle deviation threshold value, the X-direction deviation threshold value and the Y-direction deviation threshold value are 5 mm, and the heading angle deviation threshold value is 0.5°. 9.The laser radar-based omnidirectional mobile robot pose adjustment method of claim 2, wherein, In the step S1.2, a specific manner of acquiring the multiple frames of laser radar original data is that one frame of laser radar data is acquired first, and then the next frame of laser radar data is acquired after waiting for 1 second.

Images