Wheel odometry automatic calibration method, device, robot, system and storage medium

Abstract

The present application relates to a kind of multi-laser radar outside and wheeled odometer automatic calibration method, device, mobile robot, calibration system and computer readable storage medium.The method comprises: S1, the straightness parameter calibration of wheeled odometer, contain wheel rotation, straight line and compensation three sub-phases;S2, multi-laser radar outside parameter calibration is carried out, contain linear and rotation two sub-phases, and with cumulative displacement or angle fall into preset threshold interval trigger sampling event;S3, based on nonlinear least squares solver, with SE (2) constraint as residual, each radar outside parameter is optimized jointly;S4, by the consistency residual of secondary predetermined excitation motion detection point cloud and odometer-radar heading residual, determine whether calibration passes and output failure reason code.This method does not need external calibration tool, and realizes joint calibration and automatic quality determination by robot itself motion excitation.

Description

Technical Field

[0001] This invention relates to the field of mobile robot perception and motion control technology, specifically to an automatic calibration method, device, mobile robot, calibration system, and computer-readable storage medium for a wheeled odometer with multiple lidar external inputs. Background Technology

[0002] During the deployment of autonomous mobile robots, it is necessary to calibrate the pose parameters of the multiple lidars relative to the robot's body coordinate system, as well as the straightness parameters of the wheeled odometer, to ensure the accuracy of multi-lidar scan point cloud fusion and the accuracy of robot motion. Existing technologies typically rely on external calibration boards or manual intervention, which are complex and lack flexibility; or they use methods such as hand-eye calibration, but these often assume sufficient motion excitation, which is difficult to guarantee in the confined space of actual production lines; in addition, the parameters of multiple lidars and wheeled odometers are often calibrated separately and are not coupled, resulting in cumulative errors.

[0003] Existing technologies suffer from shortcomings such as the need for fixtures, insufficient excitation, uncoupled parameters, and lack of automated quality judgment. There is an urgent need for a solution that does not require external references, can be completed in tens of seconds with a single click, can achieve joint calibration of multiple radars and wheeled odometers, and has machine-readable quality gates. Summary of the Invention

[0004] The main objective of this application is to provide an automatic calibration method, device, mobile robot, calibration system, and computer-readable storage medium for wheeled odometers with multiple lidar external participation, in order to solve the technical problems in the prior art, such as reliance on external fixtures, insufficient excitation, cumulative errors caused by phased parameter calibration, and lack of automated quality judgment. It achieves the technical effects of no need for external reference, joint calibration, observability assurance, and automated quality gating.

[0005] To achieve the above objectives, this application adopts the following technical solution: This invention discloses an automatic calibration method for a multi-laser radar extrinsic wheel odometer, comprising: S1: Straightness parameter calibration stage, including three sub-phases: wheel rotation phase, wheel straight-line phase and wheel compensation phase. These are triggered by motion excitations of the robot rotating in place, moving straight forward, and moving straight again after correction, and output the heading ratio correction coefficient, straight-line heading deviation and displacement scale correction coefficient, respectively. S2: Multi-LiDAR extrinsic parameter calibration stage, including two sub-phases: linear motion phase and rotational motion phase, which are triggered by the robot's forward and backward straight movement and stationary forward and reverse rotation motion excitation, respectively. Each phase is triggered by the cumulative displacement or cumulative angle falling into a preset threshold range as the sampling event trigger condition. The first and last two frames of the sampling event are recorded for each LiDAR, along with the corresponding odometry pose. A minimum sampling threshold is forcibly applied to each phase. S3: Based on the ICP-Ceres joint optimization stage of extrinsic parameters, for each lidar, the extrinsic parameters of the lidar relative to the robot body are solved by the nonlinear least squares solver using the sample from S2 with SE(2) constraints as residuals and Huber loss as robust kernel. S4: In the verification phase, after applying the parameters output from S1 and S3, the consistency residual of the multi-radar point cloud measured by the secondary predetermined excitation motion and the odometry-radar heading residual are used to determine whether the calibration passes, and the corresponding failure reason code is output when the calibration fails.

[0006] Further, in step S1: the wheel rotation phase is obtained by comparing the heading change measured by the anchor lidar ICP with the heading change output by the odometer, and after multiple samplings, the heading ratio correction coefficient k_θ is obtained; the wheel straight-line phase is obtained by statistically analyzing the mean value of the displacement direction angle of the anchor lidar straight-line segment, and the straight-line heading deviation δ_heading is obtained; the wheel compensation phase is obtained by comparing the displacement amplitude of the anchor lidar straight-line segment with the displacement amplitude of the odometer after the heading ratio correction coefficient and the straight-line heading deviation have been applied, and the displacement scale correction coefficient k_d is obtained.

[0007] Furthermore, in S2, the sampling event is triggered when the robot's cumulative displacement falls within a preset displacement threshold range [L_min, L_max], or when the robot's cumulative rotation angle falls within a preset angle threshold range [θ_min, θ_max]. Sampling is not triggered when the cumulative amount is less than the lower limit, and is forcibly triggered and the cumulative count is cleared when the cumulative amount is greater than the upper limit.

[0008] Furthermore, in S2, both the linear motion phase and the rotational motion phase are provided with a minimum sampling threshold and a maximum sampling limit. The solution in S3 is only entered when both phases reach the minimum sampling number. If the minimum sampling number is not reached and the timeout occurs, the insufficient excitation code in the failure reason code is recorded. When the number of samples reaches the maximum sampling limit, sampling of that phase is stopped.

[0009] Furthermore, the cost function of S3 constructs the following residual for each sampling event: , The relative pose of the i-th lidar sensor measured by the ICP algorithm at the k-th sampling point. This represents the cumulative pose of the wheel odometer within the same sampling period. Let be the extrinsic parameters of the i-th lidar to be solved; and use the Huber robust kernel function for weighted accumulation to minimize the total cost function; at the same time, perform inlier rate and residual threshold screening for each sample pair in the ICP stage, and automatically remove sample pairs that fail to pass.

[0010] Furthermore, the determination in S4 includes: (a) Apply the heading ratio correction factor, straight heading deviation and displacement scale correction factor output by S1 to the odometer; (b) Apply the extrinsic parameters of each lidar output by S3 to point cloud fusion; (c) The calibration is deemed successful when the consistency residual of the multi-radar point cloud measured by the secondary predetermined excitation motion is less than the preset translation threshold and the odometry-radar heading residual is less than the preset angle threshold; otherwise, it fails. (d) Output the failure reason code when the failure fails. The failure reason code includes at least one or more of the following: insufficient excitation code, ICP mismatch code, coefficient divergence code, multi-radar consistency residual exceeding limit code, solver non-convergence or covariance exceeding limit code.

[0011] Furthermore, S1 and S2 are completed sequentially by the same continuous motion sequence, and no stop is required between the three phases; when switching phases, the robot automatically switches only according to the sequence of motion.

[0012] Furthermore, the method estimates all parameters using the robot's own motion as the sole stimulus without relying on any external calibration board, reflective markers, or human intervention; and the failure reason codes and suggested troubleshooting actions output by the method are reported to the production line MES system as quality gates through a preset communication interface.

[0013] The present invention also provides an automatic calibration device for a wheeled odometer with multiple lidar external parameters, comprising a motion control module, a data acquisition module, a straightness parameter estimation module, an external parameter estimation module, a verification module, and a configuration writing module; each module is used to execute the method described above.

[0014] The present invention also provides a mobile robot that performs the above-described method using the multi-laser radar external participation wheeled odometer automatic calibration device as described in the second aspect of this application.

[0015] The present invention also provides a calibration system, which includes the mobile robot described above and a calibration host computer; the calibration host computer receives the status field and failure reason code output by the mobile robot in S4 through a communication interface, and reports the corresponding result to the production line MES system according to the status field.

[0016] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described above.

[0017] The present invention provides an automatic calibration method, device, mobile robot, calibration system, and computer-readable storage medium for wheeled odometers using multiple lidar extrinsic parameters. By designing a multi-stage process including straightness parameter calibration and multi-radar extrinsic parameter calibration, and by mandating that both linear and rotational phases reach the minimum number of samples, the observability of the solved equations is ensured, and solution divergence under under-excitation is avoided. By jointly solving the wheeled odometer straightness parameter and multi-radar extrinsic parameters within the same continuous motion sequence, and using the ICP-Ceres framework to construct SE(2) constraints for optimization, cumulative errors caused by staged calibration are avoided. Simultaneously, by performing residual checks based on secondary excitation motion during the verification stage and outputting a judgment result containing specific failure reason codes, machine-readable quality gating is achieved. Finally, the entire method relies solely on the robot's own motion as the sole excitation, requiring no external calibration board or manual intervention, thus achieving the beneficial effect of rapid joint calibration without external benchmarks and providing automated quality judgment. In addition, by setting threshold range trigger conditions for sampling events, static or slightly jittery data is prevented from entering the optimization process, and ICP mismatch caused by excessive single displacement is also avoided, thus improving the robustness of the calibration process.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1) This invention is entirely self-excited by the robot, requiring no fixtures or setup, and is suitable for rapid production line operations; 2) This invention solves for the straightness parameters of the wheel odometer and the external parameters of multiple radars within the same motion sequence, avoiding the accumulation of errors during staged calibration; 3) This invention uses a two-phase automatic triggering method consisting of linear and rotating phases, and applies a hard gate to the sample integrity to avoid dedivergence under underexcitation; 4) This invention: Each sample is triggered by a threshold range (minimum displacement / minimum angle, with an upper limit), which avoids static or slightly jittery data from entering the optimization, and also avoids ICP mismatch caused by a single excessive displacement; 4) This invention provides a clear determination of "calibration pass / fail + failure reason code" through secondary excitation and residual check, which can be used as a quality gate in the production line MES system; 5) The algorithm provided by this invention has no limit on the number of lidars and can calibrate N 2D lidars at the same time; it can also be used in wheeled odometers with IMU assistance. Attached Figure Description

[0019] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments and to explain the operating principles of the embodiments in conjunction with the relevant descriptions in the specification. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention. Components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0020] Figure 1 This is an overall flowchart of the method provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the motion sequence during the straightness calibration stage of an embodiment of the present invention; Figure 3 This is a schematic diagram of the motion sequence and sampling events during the multi-radar extrinsic parameter calibration stage (Linear / Angular) according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the factor graph for ICP-Ceres joint optimization in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] The method provided by this invention is mainly applied to scenarios where mobile robots (such as AGV / AMR) are automatically calibrated together at production line workstations or deployment sites, using the external parameters of multiple two-dimensional lidars and the straightness parameters of wheeled odometers.

[0024] Before describing the specific implementation methods, let's first explain some key terms used in this document: Anchor lidar: This refers to the lidar designated to provide a unique observation reference when performing straightness parameter calibration for S1. This lidar is typically selected for its good field of view coverage and stable echo.

[0025] ICP stands for Iterative Closest Point Algorithm, a point cloud registration method used to align two sets of point clouds and solve for their relative poses.

[0026] SE(2): refers to a special two-dimensional Euclidean group used to represent rigid body motion (translation and rotation) in a plane.

[0027] Huber loss: A robust kernel function used in nonlinear optimization to impose a small penalty on outliers, thereby improving the robustness of the solution.

[0028] Failure reason code: refers to a machine-readable identifier that is output to indicate the specific failure type when the calibration process fails for a specific reason, such as "insufficient stimulus" or "ICP mismatch".

[0029] like Figure 1 As shown in the embodiment of this application, an automatic calibration method for a wheeled odometer with multiple lidar external participation is provided. The method includes: S1: Straightness parameter calibration stage, including three sub-phases: wheel rotation phase, wheel straight-line phase, and wheel compensation phase; S2: Multi-LiDAR extrinsic parameter calibration stage, including two sub-phases: linear motion phase and rotational motion phase; S3: External parameter joint optimization stage based on ICP-Ceres; S4: Verification phase.

[0030] In this embodiment, the above stages work together. Stage S1 directly excites and solves the odometer's own heading ratio, heading deviation, and displacement scale error through specific robot movements (rotation in place, straight movement, and corrected straight movement). Stage S2 triggers sampling events for each lidar through another set of alternating linear and rotational movements, accumulating data constraints for solving extrinsic parameters. Stage S3 uses the data collected in S2 to construct the SE(2) constraint equation between the observation pose of each lidar and the odometer pose, and solves the extrinsic parameters of all lidars through nonlinear optimization. Stage S4 applies the results of S1 and S3, verifies the convergence and consistency of the calibration results through secondary excitation movements, and finally gives a clear pass judgment or a failure report containing specific reasons. Together, they solve the problems of relying on external benchmarks, parameter calibration separation leading to cumulative errors, and lack of automated quality judgment in the prior art, achieving a comprehensive effect of no external benchmark required, rapid joint calibration, ensuring observability, and realizing automated quality gating.

[0031] Taking a specific application scenario as an example, suppose a mobile robot equipped with four 2D LiDARs is parked at a production line station. After starting the calibration program, the robot first executes stage S1: turning 90 degrees to the left in place, recording the heading change Δθ_lidar measured by the anchor LiDAR ICP and the odometer output Δθ_odom; then turning 90 degrees to the right, and recording again; repeating this process until the number of samples is sufficient, and calculating the heading scale correction coefficient k_θ. Next, the robot moves forward 1 meter, obtaining the displacement direction and magnitude through the anchor LiDAR ICP, and calculating the mean value of the direction angle to obtain the straight-line heading deviation δ_heading. Subsequently, after applying k_θ and δ_heading, the robot moves forward 1 meter again, comparing the displacement amplitude, and obtaining the displacement scale correction coefficient k_d. Following this, the robot seamlessly enters phase S2, alternating between forward and backward linear motion and left and right rotational motion. Whenever the cumulative displacement or cumulative angle falls within a preset range (e.g., 0.5m to 2.0m, or 45° to 180°), sampling of all four LiDARs is triggered, recording the point cloud and corresponding odometry pose of each LiDAR at the beginning and end of the event. Once at least four valid samples have been collected for both linear and rotational motions, phase S3 begins, where an optimization problem is constructed for each LiDAR, using all sample pairs to solve for its extrinsic parameters (x, y, θ). Finally, in phase S4, the obtained k_θ, δ_heading, k_d, and the four extrinsic parameters are written to a configuration file. The robot performs a brief verification motion to check the consistency of the point cloud fusion. If the residuals meet the requirements, the calibration is reported as successful; otherwise, a specific failure reason code is output.

[0032] Based on the above embodiments, regarding the aforementioned S1, S1 specifically includes: The wheel rotation phase is obtained by comparing the heading change measured by the anchor lidar ICP with the odometer output heading change, and after multiple samplings, a heading proportional correction coefficient k_θ is calculated; the wheel straight-line phase is obtained by statistically analyzing the average displacement direction angle of the anchor lidar's straight-line segment, and the straight-line heading deviation δ_heading is calculated; the wheel compensation phase, after the heading proportional correction coefficient and the straight-line heading deviation have been applied, is obtained by comparing the displacement amplitude of the anchor lidar's straight-line segment with the odometer's displacement amplitude, and the displacement scale correction coefficient k_d is calculated. Furthermore, combining the above principles, this method can bring the unique and beneficial effect of directly using the robot's own motion to correct the odometer system error, providing a more accurate odometer pose input for subsequent external parameter calibration, and reducing coupling errors from the source. As a specific implementation method, the formula for calculating k_θ is: k_θ=mean(Δθ_odom / Δθ_lidar), where mean represents the average value; δ_heading is the average value of the displacement direction angle θ_lidar obtained by the anchor lidar ICP during multiple straight movements; the formula for calculating k_d is: k_d=mean(Δd_lidar / Δd_odom).

[0033] Based on the above embodiments, regarding S2 mentioned above, the triggering condition for this sampling event is that the robot's cumulative displacement falls within a preset displacement threshold range [L_min, L_max], or the robot's cumulative rotation angle falls within a preset angle threshold range [θ_min, θ_max]. Sampling is not triggered when the cumulative amount is less than the lower limit, and is forcibly triggered and the cumulative count is reset when the cumulative amount is greater than the upper limit. Furthermore, combining the above principles, this threshold range triggering mechanism can provide robustness benefits by avoiding invalid data generated due to the robot being stationary or experiencing minor vibrations, and by preventing ICP matching failure due to excessively large single-movement displacement or angle. As a specific implementation, L_min can be set to 0.5 meters, L_max can be set to 2.0 meters; θ_min can be set to π / 4 (i.e., 45 degrees), and θ_max can be set to π (i.e., 180 degrees).

[0034] Based on the above embodiments, for S2 mentioned above, both the linear motion phase and the rotational motion phase have a minimum sampling threshold and a maximum sampling limit. The solution process for S3 only proceeds when both phases reach the minimum sampling count. If the minimum sampling count is not reached and a timeout occurs, the insufficient excitation code in the failure reason code is recorded. Sampling for that phase stops when the sample count reaches the maximum sampling limit. Furthermore, combining the above principles, using this mandatory minimum sampling threshold can bring unique and beneficial effects, such as strictly ensuring the observability of the external parameter solution equation system and effectively avoiding solution divergence or unreliable results under under-excitation environments. As a specific implementation, the minimum sampling threshold min_size can be set to 4, and the maximum sampling limit max_size can be set to 8.

[0035] Based on the above embodiments, for S3 mentioned above, the cost function of S3 constructs the following residual for each sampling event: , The relative pose of the i-th lidar sensor measured by the ICP algorithm at the k-th sampling point. This represents the cumulative pose of the wheel odometer within the same sampling period. Let be the extrinsic parameters of the i-th lidar to be solved; and use the Huber robust kernel function for weighted accumulation to minimize the total cost function; at the same time, perform inlier rate and residual threshold screening for each sample pair in the ICP stage, and automatically remove sample pairs that fail. Furthermore, combined with the above principles, the residual form based on SE(2) constraints and the Huber robust kernel function can bring the unique beneficial effect of robustly solving the extrinsic parameters even when there is observation noise and outliers; while the inlier rate screening in the ICP stage further ensures the data quality of the input optimization problem. As a specific implementation method, the threshold parameter δ of the Huber loss can be set to 0.05 m; the ICP inlier rate threshold can be set to 0.5.

[0036] Based on the above embodiments, regarding S4 mentioned above, the determination of S4 includes: (a) applying the heading ratio correction coefficient, straight heading deviation, and displacement scale correction coefficient output by S1 to the odometer; (b) applying the extrinsic parameters of each lidar output by S3 to point cloud fusion; (c) determining that the calibration passes when the consistency residual of the multi-radar point cloud measured by the secondary predetermined excitation motion is less than a preset translation threshold and the odometer-radar heading residual is less than a preset angle threshold; otherwise, it fails; (d) outputting a failure reason code when it fails. The failure reason code includes at least one or more of the following: insufficient excitation code, ICP mismatch code, coefficient divergence code, multi-radar consistency residual exceeding the limit code, solver non-convergence code, or covariance exceeding the limit code. Furthermore, combined with the above principles, this verification process can bring the unique and beneficial effect of performing final quality inspection of the calibration results, ensuring the reliability of the output parameters; while outputting specific failure reason codes provides clear guidance for the production line to quickly troubleshoot problems. As a specific implementation method, the translation residual threshold can be set to 0.01 meters, and the angle residual threshold can be set to 0.5 degrees; the coefficient divergence criterion can be: k_θ or k_d deviates from 1.0 by more than 0.05, or the absolute value of δ_heading exceeds 5 degrees.

[0037] Based on the above embodiment, S1 and S2 are completed sequentially by the same continuous motion sequence, and no downtime is required between the three phases; the robot automatically switches phases only according to the motion sequence order. Furthermore, combined with the above principle, this continuous automatic execution method can bring unique and beneficial effects such as significantly shortening the overall calibration time and improving the production line cycle time.

[0038] Based on the above embodiments, this method estimates all parameters solely through the robot's own motion without relying on any external calibration plates, reflective markers, or human intervention. Furthermore, the failure reason codes and suggested troubleshooting actions output by this method are reported to the production line MES system via a preset communication interface as a quality gate. In addition, combining the above principles, this fully self-excited and automated quality reporting method can bring unique and beneficial effects, achieving true "one-click calibration" and digital quality management, perfectly adapting to the needs of industrial production lines.

[0039] It should be noted that in some optional implementations, the specific implementation of the method of this application may include more operational details. For example, in the initialization phase, the robot's body geometry parameters and the coarse extrinsic parameters of each LiDAR (such as those from CAD drawings) are loaded as initial optimization values, and the data receiving threads of all sensors are started. When selecting the anchor LiDAR, it can be automatically determined based on its field of view coverage and historical point cloud stability. In the wheel rotation phase of S1, the fixed angle of the robot's stationary forward and reverse rotation is preferably in the range of ±π / 4 to ±π, for example, ±π / 2 can be selected. In S2, the motion sequence of the linear motion phase and the rotational motion phase can be designed as an alternating pattern of [forward → backward → forward → backward] and [left turn → right turn → left turn → right turn] to ensure sufficient excitation. When using the Ceres solver in S3, the initial values ​​of the extrinsic parameters can be set to coarse extrinsic parameters or zero values. In S4, failure reason codes are: (i) insufficient excitation, corresponding to any sub-phase in S2 not reaching the minimum sampling number min_size; (ii) ICP mismatch, corresponding to the anchor radar ICP in-point rate being lower than the threshold or the residual being higher than the threshold; (iii) coefficient divergence, corresponding to k_θ and k_d deviating from 1.0 and exceeding the threshold or the absolute value of δ_heading exceeding the threshold; (iv) multi-radar consistency residual exceeding the limit, corresponding to the point cloud alignment residual exceeding the limit in the secondary excitation; (v) solver not converging or covariance exceeding the limit, corresponding to Ceres solver iteration failure or the estimated covariance diagonal element being too large. These failure reason codes can be mapped to specific suggested troubleshooting actions, such as suggesting expanding the workstation space for insufficient excitation, and suggesting checking radar field of view obstruction for ICP mismatch, and reporting to MES via OPC-UA or MQTT protocol.

[0040] It should be noted that, in some alternative implementations, the specific parameters involved in the method of this application can be adjusted according to different robot platforms and workstation environments. The settings of the threshold intervals [L_min, L_max] and [θ_min, θ_max] need to balance sufficiency of excitation and workstation space constraints. The minimum number of samples, min_size, is crucial for ensuring observability and should not be too low. The specific implementations of the ICP matching algorithm and Ceres solver can use open-source or commercial libraries. This method has no limit on the number of lidars N, and multiple lidars can be calibrated simultaneously. This method is also applicable to wheeled odometer systems with IMU assistance, in which case S1 can fuse IMU data for more accurate straightness parameter estimation.

[0041] This application also provides an automatic calibration device for a wheeled odometer with multiple external lidar inputs. For example... Figure 3As shown, the device includes a motion control module 10, a data acquisition module 20, a straightness parameter estimation module 30, an extrinsic parameter estimation module 40, a verification module 50, and a configuration writing module 60. The motion control module 10 drives the robot to perform straightness calibration and extrinsic parameter calibration movements according to a preset sequence; the data acquisition module 20 receives data from the wheel odometer and each lidar; the straightness parameter estimation module 30 executes the logic of S1 (wheel rotation, wheel straight movement, wheel compensation); the extrinsic parameter estimation module 40 executes the logic of S2 (linear motion, rotational motion) and solves the ICP-Ceres equation in S3; the verification module 50 executes the secondary excitation and residual check in S4, and generates failure reason codes; the configuration writing module 60 persists the calibration results to the robot parameter file. These modules work together to execute the steps of any of the aforementioned method embodiments, realizing an automated and integrated calibration process.

[0042] This application also provides a mobile robot. The mobile robot includes: a mobile chassis, N two-dimensional LiDAR sensors (N≥2) mounted on the mobile chassis, a wheeled odometer, and a processing unit connected to these sensors. The processing unit is configured to execute the steps of any of the foregoing method embodiments and write the obtained calibration parameters into the mobile robot's parameter file for subsequent use by the SLAM and navigation modules.

[0043] This application also provides a calibration system. The system includes the mobile robot described in the aforementioned mobile robot embodiment and a calibration host computer. The calibration host computer receives the status field and failure reason code output by the mobile robot in S4 via a communication interface, and reports the corresponding result (pass or fail and specific reason) to the production line MES system based on the status field.

[0044] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the foregoing method embodiments.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art should be able to make equivalent embodiments by making some changes or modifications to the above-disclosed technical content without departing from the scope of the present invention. Any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An automatic calibration method for a wheeled odometer with multiple lidar external inputs, characterized in that, Includes the following steps: S1. Perform straightness parameter calibration, which includes three sub-phases: wheel rotation phase, wheel straight-line phase, and wheel compensation phase; trigger the wheel rotation phase through robot in-situ rotation motion excitation, and output the heading ratio correction coefficient; The straight-line phase of the wheel is triggered by the robot's forward straight-line motion excitation, and the straight-line heading deviation is output; the wheel compensation phase is triggered by the robot's parameter-corrected straight-line motion excitation, and the displacement scale correction coefficient is output. S2. Perform multi-LiDAR extrinsic parameter calibration, which includes two sub-phases: linear motion phase and rotational motion phase. The linear motion phase is triggered by the robot's forward and backward linear motion excitation, and the rotational motion phase is triggered by the robot's stationary forward and reverse rotational motion excitation. The sampling event is triggered when the robot's cumulative displacement or cumulative angle falls into a preset threshold range. For each LiDAR, the first and last two frames of scanned point cloud and the corresponding odometry pose of the sampling event are recorded, and a minimum sampling threshold is set for both sub-phases. S3. Perform joint optimization of extrinsic parameters based on ICP-Ceres. For each lidar, select the effective samples collected in step S2, construct the residuals with SE(2) constraints, use Huber loss as the robust kernel, and solve the extrinsic parameters of each lidar relative to the robot body through a nonlinear least squares solver. S4. Perform verification. Apply the straightness parameters calibrated in step S1 and the LiDAR extrinsic parameters optimized in step S3 to the robot odometry and point cloud fusion, respectively. Through the robot's secondary predetermined excitation motion, collect the consistency residuals of the multi-radar point cloud and the odometry-radar heading residuals. Determine whether the calibration is successful based on the residual values. If the calibration fails, output the corresponding failure reason code.

2. The automatic calibration method for a wheeled odometer with multiple lidar external inputs according to claim 1, characterized in that, The parameter calibration method for each sub-phase in step S1 is as follows: Wheel rotation phase: By comparing the heading change measured by the anchor lidar using the ICP algorithm with the heading change output by the wheel odometer, and after multiple sampling calculations, the heading ratio correction coefficient k_θ is obtained; Straight-line phase: The average value of the displacement direction angle of the anchor lidar during the straight-line motion segment is used to calculate the straight-line heading deviation δ_heading; Wheel compensation phase: Based on the application of the heading ratio correction coefficient and the straight heading deviation, the displacement amplitude of the anchor lidar in the straight movement segment is compared with the displacement amplitude output by the wheel odometer to calculate the displacement scale correction coefficient k_d.

3. The automatic calibration method for a wheeled odometer with multiple external lidar inputs according to claim 1, characterized in that, The specific triggering condition for the sampling event in step S2 is as follows: If the robot performs linear motion, sampling is triggered when the cumulative displacement falls within the preset displacement threshold range [L_min, L_max]; if the robot performs rotational motion, sampling is triggered when the cumulative rotation angle falls within the preset angle threshold range [θ_min, θ_max]. When the cumulative motion amount is less than the lower limit of the corresponding threshold, sampling is not triggered. When the cumulative motion amount exceeds the upper limit of the corresponding threshold, sampling is forcibly triggered and the cumulative count is cleared.

4. The automatic calibration method for a wheeled odometer with multiple external lidar inputs according to claim 1, characterized in that, In step S2: Both the linear motion phase and the rotational motion phase are configured with a minimum sampling threshold and a maximum sampling limit; the external parameter solution process in step S3 can only proceed when both the linear motion phase and the rotational motion phase meet the minimum sampling requirement. If the motion excitation times out and the minimum number of samples is not reached, the sampling excitation is determined to be abnormal, and the insufficient excitation code in the failure reason code is recorded; if the number of samples reaches the maximum sampling limit, the sampling operation of the corresponding phase is terminated immediately.

5. The automatic calibration method for a wheeled odometer with multiple external lidar inputs according to claim 1, characterized in that, The specific method for constructing the cost function in step S3 is as follows: For each sampling event, construct the residual formula: In the formula, The relative pose of the i-th lidar sensor measured by the ICP algorithm at the k-th sampling point. This represents the cumulative pose of the wheel odometer within the same sampling period. Let be the extrinsic parameters of the i-th lidar to be solved; Using Huber robust kernel function The residuals are weighted and accumulated to construct the total cost function and then minimized. At the same time, all sampled samples are pre-screened to remove invalid samples with in-point rate of ICP or residual exceeding the threshold.

6. The automatic calibration method for a wheeled odometer with multiple external lidar inputs according to claim 1, characterized in that, The verification and determination logic in step S4 specifically includes: (1) Write the heading ratio correction coefficient, straight heading deviation, and displacement scale correction coefficient obtained in step S1 into and correct the wheel odometer parameters; (2) Apply the external parameters of each lidar obtained in step S3 to the robot's multi-radar point cloud fusion algorithm; (3) Control the robot to perform a second predetermined excitation motion, and collect the consistency residual of multiple radar point clouds and the odometry-radar heading residual; when the consistency residual of multiple radar point clouds is less than the preset translation threshold and the odometry-radar heading residual is less than the preset angle threshold, the calibration is deemed qualified; otherwise, the calibration is deemed unqualified. (4) When the calibration is not qualified, output the corresponding failure reason code; the failure reason code includes at least one of the following: insufficient excitation code, ICP mismatch code, coefficient divergence code, multi-radar consistency residual exceeding limit code, solver non-convergence code, and covariance exceeding limit code. The straightness parameter calibration in step S1 and the multi-LiDAR extrinsic parameter calibration in step S2 are completed through the same continuous robot motion sequence. There is no need to stop between the three straightness parameter calibration sub-phases. The robot automatically completes the phase switching according to the preset motion sequence to achieve continuous calibration. The calibration method does not rely on external calibration boards or reflective markers throughout the entire process, and requires no manual intervention. It uses only the robot's own movement as a stimulus to complete all parameter estimations. The failure reason codes and troubleshooting suggestions output by the method are reported to the production line MES system through a preset communication interface, serving as the robot's factory quality verification threshold.

7. An automatic calibration device for a wheeled odometer with multiple lidar external inputs, characterized in that, It includes a motion control module, a data acquisition module, a straightness parameter estimation module, an extrinsic parameter estimation module, a verification module, and a configuration writing module; The motion control module is used to control the robot to perform preset excitation motion; the data acquisition module is used to acquire lidar point cloud data and odometry pose data; the straightness parameter estimation module is used to perform the straightness parameter calibration step according to any one of claims 1-6; the extrinsic parameter estimation module is used to perform the lidar extrinsic parameter joint optimization step according to any one of claims 1-6; the verification module is used to perform the calibration verification step according to any one of claims 1-6; and the configuration writing module is used to write the valid parameters after calibration into the robot configuration file.

8. A mobile robot, characterized in that, The system includes a mobile chassis, a wheeled odometer, at least two two-dimensional lidar sensors, and a processing unit; the two-dimensional lidar sensors and the wheeled odometer are both mounted on the mobile chassis and are communicatively connected to the processing unit. The processing unit is configured to execute all the steps of the automatic calibration method according to any one of claims 1-6, and write the final calibration parameters into the parameter configuration file of the mobile robot.

9. A calibration system, characterized in that, It includes a calibration host computer and the mobile robot as described in claim 8; the calibration host computer establishes communication with the mobile robot through a preset communication interface, and is used to receive the status field and failure reason code output by the mobile robot during the verification stage, and report the calibration result to the production line MES system according to the calibration status to complete the quality verification.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the automatic calibration method for a multi-laser radar externally-assisted wheeled odometer as described in any one of claims 1-6.

Images