Transfer AGV parking control method based on commercial vehicle tire contour memory inference
By installing 2D LiDAR and GPS systems on the transfer AGV and combining them with support vector regression algorithms, the problem of inaccurate longitudinal parking control caused by point cloud occlusion in traditional methods has been solved, achieving high-precision and stable longitudinal parking control.
Patent Information
- Application Number
- CN202510836060.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-20
- Publication Date
- 2025-11-18
AI Technical Summary
When traditional algorithms rely on visual or 3D LiDAR point cloud recognition to identify vehicle information, the point cloud is obscured after the vehicle enters the AGV, resulting in insufficient accuracy and robustness of longitudinal parking control, which affects vehicle retrieval operations.
A parking control method for transport AGVs based on the memory inference of the tire contour of commercial vehicles is adopted. Using 2D LiDAR and GPS positioning system, the wheel pose is corrected in real time through support vector regression algorithm and complementary calculation between wheel positions to ensure accurate parking.
It improves the accuracy and robustness of longitudinal parking control of transfer AGVs, ensuring stability and efficiency in different application scenarios and meeting the operational needs of commercial vehicle transfer AGVs.
Smart Images
Figure CN120972784A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of commercial vehicle taking, and relates to a transfer AGV parking control method based on commercial vehicle tire contour memory inference and complementation. BACKGROUND
[0002] Commercial vehicle transfer AGVs play a key role in the fields of automobile manufacturing and logistics, and they are responsible for transporting commercial vehicles in the commonly used areas of port, such as ship loading and unloading areas, yards and stereo garages. In this process, automatic docking as the core link of the transfer process directly affects the safety of the operation. The docking process can be divided into horizontal error correction and longitudinal parking control, and the longitudinal parking control is related to whether the commercial vehicle can be accurately and safely taken out. Traditional algorithms usually rely on vision or 3D laser radar point cloud recognition to obtain a bounding box and extract information about the front and rear of the vehicle, but after the point cloud is blocked and the features are lost after the commercial vehicle enters the inside of the AGV, the vehicle cannot be accurately recognized, resulting in deviation of the longitudinal parking control and affecting the taking operation. Therefore, a new method is needed to improve the accuracy and robustness of the longitudinal parking control of the transfer AGV. SUMMARY
[0003] In order to overcome the problems in the prior art, the purpose of the present application is to provide a transfer AGV parking control method based on commercial vehicle tire contour memory inference, which can improve the accuracy and robustness of the longitudinal parking control of the transfer AGV.
[0004] The technical solution of the present application to solve the above technical problems is as follows:
[0005] The transfer AGV parking control method based on commercial vehicle tire contour memory inference disclosed by the present application comprises the following steps:
[0006] Step 1: Collect and process commercial vehicle wheel data to obtain its relative pose sequence relative to the AGV. In addition, the AGV is equipped with a GPS positioning system to provide AGV global pose information.
[0007] Step 2: After the AGV detects the rear wheel of the commercial vehicle, it starts to decelerate, and after the deceleration is completed, it continues to move forward and records the relative pose sequence of the rear wheel of the commercial vehicle and the global pose information-time sequence of the AGV in real time; if the sequence length exceeds the preset value, the data with the longest time interval is deleted to maintain real-time performance; it is judged whether the relative pose sequence jumps or not, if there is no jump, it is directly output, if there is a jump, it is further corrected according to the jump duration, and the corrected relative pose sequence is output.
[0008] Step three: further slow down when the AGV detects the front wheels of the commodity vehicle; continue to move forward after the deceleration is completed and detect the positions of the four wheels of the commodity vehicle in real time, and record the relative pose sequence of the front wheels; for the rear wheels of the commodity vehicle, supplement the record of the relative pose sequence based on the output sequence in step two; if the sequence length exceeds the preset value, delete the data with the longest time interval to maintain real-time performance; judge whether the relative pose sequence starts to jump and correct and output the latest relative pose after correction.
[0009] Step four: when the front wheel clamping device of the AGV is aligned with the front wheels of the commodity vehicle, it means that the AGV has reached the target parking position, and the AGV stops moving forward.
[0010] Further, in step one, the specific implementation method is that the transfer AGV is installed with four wheel clamping devices carrying 2D laser radars on the left front, right front, left rear and right rear inside the vehicle, and commodity vehicle wheel point clouds are collected; based on the geometric features of the wheel contour, the collected commodity vehicle wheel point clouds are fitted using a search-based best bounding box fitting algorithm to construct the wheel contour to obtain the longitudinal pose-time sequence of each wheel relative to the center of the AGV, which is referred to as the relative pose sequence.
[0011] Further, in step two, the method for correction according to the jump duration is that when the jump duration of the relative pose sequence does not exceed the threshold, the pose of the rear wheel of the commodity vehicle is fitted using a support vector regression algorithm according to the relative pose sequence and the current time to correct the error, and then the corrected relative pose sequence of the rear wheel of the commodity vehicle is output.
[0012] When the jump duration of the relative pose sequence exceeds the threshold, the prediction of the support vector regression algorithm may be inaccurate at this time, so the last frame of information of the relative pose sequence is converted using the relative pose sequence, the distance of the AGV movement is calculated according to the AGV global pose at the corresponding time, and the relative pose of the commodity vehicle wheel is reversely calculated to correct the error, and then the corrected relative pose sequence of the rear wheel of the commodity vehicle is output.
[0013] Further, in step three, the method for judging and correcting the jump is that if there is no jump, it means that the relative pose sequence is relatively stable, and the last frame of information of each relative pose sequence is the latest relative pose of the commodity vehicle wheel and is output.
[0014] When only one of the four wheels of the commodity vehicle appears a jump in the relative pose sequence, the positions of the other three wheels are used to calculate the position of this wheel according to the rectangular relationship, and the latest relative pose of each wheel of the commodity vehicle is output.
[0015] If the relative pose sequence of two or more commodity car wheels jumps, the support vector regression algorithm is used to fit the wheel position to correct the error, and the latest relative pose of each commodity car wheel is output.
[0016] If the relative pose sequence of at least one commodity car wheel jumps for more than a threshold, the last frame information of the relative pose sequence is converted using the relative pose sequence, the distance moved by the AGV is calculated based on the corresponding time of the jump, and the relative pose of the commodity car wheel at that time is inversely calculated to correct the error, and the latest relative pose of each commodity car wheel is output.
[0017] Further, when the commodity car transfer AGV moves to a distance of a preset value from the rear edge of the front commodity car, the laser radar starts to collect point clouds. The preset value is 2 meters.
[0018] Further, in step two, the commodity car transfer AGV is slowed down to 0.4m / s, and after waiting for 2s, it is determined that the deceleration is completed. The wheel recognition loss time threshold is 1s.
[0019] Further, in step three, the commodity car transfer AGV is slowed down to 0.1m / s, and after waiting for 2s, it is determined that the deceleration is completed.
[0020] Further, in step four, when the 2D laser radar carried by the wheel clamping device is aligned with the longitudinal center position in the latest relative pose of the recognized front wheel of the commodity car, the distance is less than 2cm, the commodity car transfer AGV is limited to 0m / s, and after waiting for 1s, it is determined that the commodity car stops running.
[0021] Advantages:
[0022] 1. The commodity car tire contour memory-based transfer AGV parking control method disclosed in the application uses the internal 2D laser radar of the transfer AGV to perceive the target vehicle wheel point cloud, and uses the support vector regression algorithm, wheel position complementation and GPS inverse calculation to calculate the target vehicle wheel position, thereby performing longitudinal parking control of the transfer AGV, avoiding relying on incomplete 3D point cloud target vehicle position of the commodity car transfer AGV for parking control, and greatly improving the longitudinal parking control accuracy of the commodity car transfer AGV. The method is easy to implement, stable in effect, high in precision, and fully guarantees the stability and efficiency of the longitudinal parking of the transfer AGV.
[0023] 2. The commodity car tire contour memory-based transfer AGV parking control method disclosed in the application designs a complete wheel perception abnormality processing procedure, which can autonomously select a reasonable processing method for different jump conditions of a single or multiple wheels, and can fully guarantee the robustness in different application scenarios.
[0024] 3. The application discloses a transfer AGV parking control method based on commodity vehicle tire contour memory inference, which meets the demand of longitudinal parking control of the transfer AGV by means of the existing structural characteristics and sensor units of the straddle double-body commodity vehicle transfer AGV; the method has a simple process, high operation efficiency and high operation accuracy, and can be well applied to the operation of the commodity vehicle transfer AGV, and meets the actual demand of the longitudinal parking process of the commodity vehicle transfer AGV. BRIEF DESCRIPTION OF DRAWINGS
[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art and the advantages thereof, a brief introduction will be given to the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.
[0026] Figure 1 It is a process schematic diagram of the transfer AGV parking control method based on commodity vehicle tire contour memory inference of the present application. DETAILED DESCRIPTION
[0027] In order to further explain the technical means and effects taken by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the technical solutions proposed according to the present application will be described in detail below in combination with the drawings and preferred embodiments. The specific features, structures or characteristics in one or more embodiments can be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used in the present application have the same meaning as understood by those skilled in the art of the technology to which the present application belongs.
[0028] The purpose of the present application is to provide a transfer AGV parking control method based on commodity vehicle tire contour memory inference, which can safely and accurately complete the longitudinal parking process control of the transfer AGV in the case that the target vehicle is blocked by the transfer AGV body or the point cloud features are lost due to other reasons.
[0029] The method is based on a straddle double-body commodity vehicle transfer AGV, which is provided with a 3D laser radar as a target vehicle sensing system at each corner of the vehicle, and is provided with a GPS positioning system to provide body pose information. The straddle double-body commodity vehicle transfer AGV has an arch shape in appearance. When transferring the commodity vehicle, the transfer AGV will "straddle" on the commodity vehicle, wrap the commodity vehicle inside the "arch", and fix the four tires of the commodity vehicle by means of four clamping devices installed inside, and then lift the commodity vehicle. A 2D laser radar is installed in the center of each clamping device to sense the target vehicle wheel information. After reaching the designated location, the commodity vehicle is lowered in a similar process, and the transfer process is completed.
[0030] The 3D laser radar installed on the four corners of the commodity car transfer AGV can construct the laser point cloud around the body and inside the body, but due to the interference of the metal body to the laser radar signal and the shielding of the body itself, the internal point cloud data is incomplete, which affects the accurate perception of the space inside the car and the adjacent obstacles. If only relying on 3D laser radar for vehicle longitudinal parking control cannot guarantee the operation accuracy. After using the supplementary 2D laser radar carried on the clamping device, the target vehicle information can be more accurately perceived, but there may still be a situation of identification failure due to obstacles such as weeds, affecting the longitudinal parking accuracy.
[0031] The present application solves the execution method of the longitudinal parking control link of the straddle two-body commodity car transfer AGV vehicle taking task. The longitudinal parking control link can be abstracted as: according to the position and orientation of the target vehicle that has been parked, the transfer AGV is required to align the clamping device with the wheel position of the vehicle in the longitudinal direction and keep the transfer AGV running speed at 0.
[0032] The task in this example is to ensure the longitudinal accurate parking control of the commodity car transfer AGV during the vehicle taking process, as shown in Figure 1 The commodity car wheel tire contour memory-based complementary transfer AGV parking control method disclosed in the embodiment is implemented as follows:
[0033] Step one: four 2D laser radar-equipped wheel clamping devices are installed on the left front, right front, left rear and right rear of the inside of the vehicle, and commodity car wheel point cloud is collected; the collected commodity car wheel point cloud is fitted based on the geometric features of the wheel contour using a search-based best bounding box fitting algorithm.
[0034] Specifically, when the transfer AGV moves to the preset position, i.e. 2m behind the target commodity car, the transfer AGV body is basically opposite to the target position, and the commodity car transfer AGV vehicle taking longitudinal parking task process is formally entered.
[0035] During this period, the commodity car transfer AGV will acquire the point cloud data collected by the 2D laser radar installed on the four corners of the commodity car transfer AGV body. In order to ensure that all point cloud data is in the same reference frame, facilitate subsequent processing and analysis, the point cloud data collected by the four laser radars is uniformly summarized into the body Cartesian coordinate system, and point cloud data preprocessing is performed, including merging, filtering, etc. Merging is to merge the point cloud data from different laser radars into a complete two-dimensional point cloud graph, filtering is to remove noise points and smooth the point cloud edge to improve the accuracy and reliability of the data, and common filtering methods include Gaussian filtering, median filtering, bilateral filtering, etc. The preprocessed point cloud data is used to construct a two-dimensional model of the target commodity car wheel, which usually involves steps such as point cloud segmentation and clustering.
[0036] The processed point cloud data is used to obtain the effective part of the point cloud using fast Euclidean clustering. Fast Euclidean clustering is a clustering algorithm based on Euclidean distance, mainly used for efficient clustering of data. In this scenario, fast Euclidean clustering is used to extract the part of the point cloud that belongs to the target vehicle wheel, so as to perform subsequent analysis and processing. The position of the wheel is fitted using the obtained clustered point cloud, which usually involves steps such as segmentation, feature extraction, and recognition of the point cloud, and through the fitting of the wheel rectangular geometric features, the transfer AGV can determine the position and attitude of the target vehicle wheel.
[0037] The wheel contour is constructed by the above-mentioned technology to obtain the longitudinal pose-time sequence of each wheel relative to the center of the AGV, referred to as the relative pose sequence; in addition, the AGV is equipped with a GPS positioning system to provide AGV global pose information;
[0038] Step two: start to decelerate when the AGV detects the rear wheel of the commodity vehicle; after completing the deceleration, continue to advance longitudinally; while advancing, the AGV detects the position of the rear wheel of the commodity vehicle in real time and records the relative pose sequence thereof; at the same time, record the AGV global pose information-time sequence; when the sequence length exceeds the preset value after data recording, the data with the longest time interval from the current time is popped out from the sequence to maintain the real-time nature of the sequence;
[0039] Specifically, the commodity vehicle transfer AGV is limited to 0.4 m / s, and after waiting for 2 seconds, it is determined that the deceleration is completed.
[0040] During this period, point cloud data may be missing or the wheel position may jump discontinuously due to obstacles such as weeds. It is judged whether the relative pose sequence starts to jump. If there is no jump, it indicates that the relative pose sequence is relatively stable, and the relative pose sequence of the rear wheel of the commodity vehicle is directly output. If the jump starts, further judgment is made on the duration of the jump of the relative pose sequence.
[0041] When the duration of the jump of the relative pose sequence does not exceed the threshold, the pose of the rear wheel of the commodity vehicle is fitted using a support vector regression algorithm according to the relative pose sequence and the current time to correct the error, and then the corrected relative pose sequence of the rear wheel of the commodity vehicle is output; the support vector regression algorithm will be modeled based on the time sequence data, and the known position points are used to fit the wheel position. The support vector regression algorithm predicts the continuous position of the wheel in the missing data part by calculating the optimal regression function, ensuring high-precision wheel positioning of the AGV during longitudinal parking. Through prediction, the transfer AGV can overcome the problem of point cloud missing caused by obstacle blocking, and accurately identify the wheel position.
[0042] When the relative pose sequence jump duration exceeds the threshold, the prediction of the support vector regression algorithm may be inaccurate at this time, the relative pose sequence is used to convert the last frame information of the relative pose sequence, the AGV global pose corresponding to the time is calculated according to the AGV moving distance, the relative pose of the commodity car wheel is reversely calculated to correct the error, and then the corrected relative pose sequence of the rear wheel of the commodity car is output;
[0043] Step three: further decelerate when the AGV detects the front wheel of the commodity car; after completing the deceleration, continue to advance longitudinally; when advancing, the AGV detects the positions of the four wheels of the commodity car in real time, and records the relative pose sequence of the front wheel. For the rear wheel of the commodity car, the relative pose sequence is supplemented on the basis of the output sequence in step two. When the sequence length exceeds the preset value after data recording, the data with the longest time interval from the current time is popped out from the sequence to maintain the real-time performance of the sequence;
[0044] Specifically, the commodity car transfer AGV is limited to 0.1 m / s, and waits for 2 seconds, and then determines that the deceleration is completed.
[0045] Whether the relative pose sequence starts to jump is judged. If no jump occurs, it indicates that the relative pose sequence is relatively stable, and the last frame information of each relative pose sequence is the latest relative pose of the wheel of the commodity car and is output.
[0046] When only one wheel of the commodity car appears relative pose sequence jump among the four wheels, the positions of the other three wheels are used to calculate the position of the wheel according to the rectangular relationship, and the latest relative pose of each wheel of the commodity car is output. Since the positions of the wheels of the target vehicle have a certain geometric relationship, if the position of one wheel is missing or jumping, the AGV can calculate the possible position of the missing wheel according to the rectangular geometric constraint through the positions of the three known wheels. The known three wheel coordinates A(x1, y1), B(x2, y2), C(x3, y3), and the missing wheel position (x4, y4) can be calculated using the following formula:
[0047] x4=x1+x2-x3
[0048] y4=y1+y2-y3
[0049] If two or more wheels of the commodity car appear relative pose sequence jump, the rectangular relationship cannot effectively calculate the wheel position, then the support vector regression algorithm is used to fit the wheel position to correct the error, and the latest relative pose of each wheel of the commodity car is output.
[0050] If the relative position sequence of at least one commodity vehicle wheel exceeds the threshold, the last frame information of the relative position sequence is converted, the distance of AGV movement is calculated according to the AGV global position at the corresponding time of the jump, the relative position of the commodity vehicle wheel at this time is reversely calculated to correct the error, and the latest relative position of each commodity vehicle wheel is output. In this way, when the wheel recognition fails, the AGV can ensure that the positions of the four wheels of the target vehicle are always accurately tracked through the multiple safeguards of geometric calculation, support vector regression algorithm and inverse calculation, thereby providing reliable data support for subsequent parking and taking tasks.
[0051] Step four: when the front wheel clamping device of the AGV is aligned with the front wheel of the commodity vehicle, it means that the AGV has reached the target parking position, and the AGV stops moving forward;
[0052] Specifically, according to the target vehicle wheel position detected or fitted by the above steps, when the 2D laser radar carried by the clamping device is less than 2cm away from the longitudinal center position of the recognized target vehicle front wheel, it is considered to be aligned, the commodity vehicle transfer AGV is limited to 0m / s, and after waiting for 1s, it is determined that the commodity vehicle stops running. At this time, the rear clamping device wheelbase is adjusted to align the position of the rear wheel of the target vehicle, and after the adjustment is completed, the four clamping devices start to fix the four tires of the commodity vehicle respectively, and the commodity vehicle is clamped to carry out the next step of transfer operation.
[0053] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.
Claims
1. A parking control method for a transfer AGV based on the memory inference of the tire contour of a commercial vehicle, characterized in that: Includes the following steps: Step 1: Collect and process the wheel data of the commercial vehicle to obtain its relative pose sequence with respect to the AGV. Step 2: After the AGV detects the rear wheels of the vehicle, it begins to decelerate. After deceleration, it continues to move forward and records the relative pose sequence of the rear wheels of the vehicle and the global pose information of the AGV in real time. If the sequence length exceeds the preset value, the data with the longest time interval is deleted to maintain real-time performance. It is determined whether there is a jump in the relative pose sequence. If there is no jump, it is output directly. If there is a jump, it is further corrected according to the duration of the jump, and the corrected relative pose sequence is output. Step 3: When the AGV detects the front wheels of the vehicle, it further decelerates; after deceleration, it continues to move forward and detects the positions of the four wheels of the vehicle in real time, and records the relative pose sequence of the front wheels; for the rear wheels of the vehicle, it supplements the record with the relative pose sequence based on the output sequence in Step 2; if the sequence length exceeds the preset value, it deletes the data with the longest time interval to maintain real-time performance; it judges and corrects whether the relative pose sequence has started to change and outputs the latest corrected relative pose. Step 4: When the front wheel clamping device of the AGV is aligned with the front wheel of the commercial vehicle, it means that the AGV has reached the target parking position and the AGV stops moving forward.
2. The parking control method for transfer AGVs based on the memory inference of vehicle tire contours as described in claim 1, characterized in that: The specific implementation method of step one is as follows: The transfer AGV has four wheel clamping devices equipped with 2D LiDAR installed on the inner left front, right front, left rear, and right rear sides of the vehicle, and collects wheel point clouds of the commercial vehicle. The collected wheel point clouds are fitted based on the geometric features of the wheel contour using a search-based optimal bounding box fitting algorithm to construct the wheel contour and obtain the longitudinal pose-time series of each wheel relative to the center of the AGV, referred to as the relative pose sequence. In addition, the AGV is equipped with a GPS positioning system to provide global pose information of the AGV.
3. The parking control method for transfer AGVs based on the memory inference of vehicle tire contours as described in claim 1, characterized in that: The method for correction based on the transition duration described in step two is as follows: When the duration of the relative pose sequence jump does not exceed the threshold, the support vector regression algorithm is used to fit the pose of the rear wheel of the commercial vehicle based on the relative pose sequence and the current time to correct the error, and then the corrected relative pose sequence of the rear wheel of the commercial vehicle is output. When the duration of the relative pose sequence jump exceeds the threshold, the prediction of the support vector regression algorithm may be inaccurate. At this time, the information of the last frame of the relative pose sequence is transformed by using the relative pose sequence. The distance moved by the AGV is estimated based on the global pose of the AGV at the corresponding time. The relative pose of the vehicle wheels is calculated in reverse to correct the error. Then, the corrected relative pose sequence of the rear wheels of the vehicle is output.
4. The parking control method for transfer AGVs based on the memory inference of vehicle tire contours as described in claim 1, characterized in that: The method for judging and correcting based on the jump situation described in step three is as follows: If no jump occurs, it indicates that the relative pose sequence is relatively stable. The last frame of each relative pose sequence is the latest relative pose of the vehicle wheel and is output. When only one of the four wheels of the vehicle experiences a relative pose sequence jump, the position of this wheel is calculated based on the rectangular relationship using the positions of the other three wheels, and the latest relative pose of each vehicle wheel is output. If two or more vehicle wheels exhibit relative pose sequence jumps, support vector regression algorithms are used to fit the wheel positions to correct the errors, and the latest relative pose of each vehicle wheel is output. If the duration of the relative pose sequence jump of at least one vehicle wheel exceeds the threshold, the last frame information of the relative pose sequence is converted using the relative pose sequence. The distance moved by the AGV is estimated based on the global pose of the AGV at the time corresponding to the jump. The relative pose of the vehicle wheel and the AGV at this time is calculated in reverse to correct the error, and the latest relative pose of each vehicle wheel is output.
5. The parking control method for a transfer AGV based on the memory inference of the tire contour of a commercial vehicle as described in claim 1, characterized in that: When the AGV transporting the goods vehicle moves to a preset distance from the rear edge of the preceding goods vehicle, the lidar begins to collect point clouds; the preset distance is 2 meters.
6. The parking control method for a transfer AGV based on the memory inference of the tire contour of a commercial vehicle as described in claim 1, characterized in that: In step two, the AGV for transporting the goods vehicle is decelerated to 0.4 m / s, and after waiting for 2 seconds, the deceleration is determined to be complete; the wheel recognition loss time threshold is 1 second.
7. The parking control method for transfer AGVs based on the memory inference of vehicle tire contours as described in claim 1, characterized in that: In step three, the AGV for transporting the goods vehicle is decelerated to 0.1 m / s, and after waiting for 2 seconds, the deceleration is determined to be complete.
8. The parking control method for a transfer AGV based on the memory inference of the tire contour of a commercial vehicle as described in claim 1, characterized in that: In step four, when the distance between the 2D LiDAR mounted on the wheel clamping device and the longitudinal center position of the latest relative pose of the front wheel of the identified vehicle is less than 2cm, it is considered to be aligned. The speed of the vehicle transfer AGV is limited to 0m / s, and the vehicle is stopped after 1s.