A positioning deviation correction control method for precise butt joint of a parent-child shuttle vehicle

Abstract

The application relates to the technical field of shuttle vehicle control, in particular to a positioning and deviation correction control method for accurate docking of a parent-child shuttle vehicle, which comprises the following steps: laying a two-dimensional code reference at each docking station; connecting the incremental encoder pulse of the front axle driving motor of the child vehicle to the external trigger port of the vision sensor, triggering the vision sensor to capture the bottom view image to obtain coarse code reading and micro code reading; taking the drift increment of the parent vehicle state relative to the departure time as a state variable, combining the micro code to calculate the pose and the coarse code matrix to calculate the pose, calculating the drift difference observation, calculating the real-time drift and feeding it into the differential compensation loop for dynamic pose drift filtering and deviation correction; after entering the normal positioning two-dimensional code coverage area, determining the closing trigger signal and starting the exit process, and archiving the drift sequence as the prior compensation value for the next docking. The method effectively improves the continuity and accuracy of positioning and deviation correction during the seam passing.

Description

Technical Field

[0001] This application relates to the field of shuttle control technology, and more specifically, to a positioning and correction control method for precise docking of mother and child shuttles. Background Technology

[0002] With the development of intelligent warehousing and logistics technology, mother-daughter shuttle cars are widely used in high-density rack storage and retrieval operations. When performing docking operations, the mother-daughter shuttle cars need to cross the transition gap between the mother car track and the rack track. The positioning accuracy of this process directly determines the success rate of docking.

[0003] Currently, existing shuttle docking positioning methods mostly rely on static single-point calibration and timed periodic visual sampling. However, in actual working conditions where the shuttle crosses transition gaps, gap excitation can easily cause equipment vibration, leading to positioning blind spots. Timed sampling methods cannot adapt to changes in the speed of the shuttle's acceleration and deceleration, and are prone to mismatch phenomena such as low-speed redundancy and high-speed frame loss.

[0004] To address the existing technical problem that traditional positioning methods based on single-point calibration lack continuous spatial references and have rigid visual sampling methods when the mother-daughter shuttle crosses transition gaps, resulting in the inability to perform continuous high-frequency high signal-to-noise ratio correction control when the equipment undergoes body drift due to gap excitation, thus causing docking failure or forced biting of the rail gap, there is an urgent need to provide a continuous, adaptive positioning correction control method with high signal-to-noise ratio. Summary of the Invention

[0005] To address the existing technical problem that the lack of a continuous spatial reference and rigid sampling method of the mother-daughter shuttle car makes it impossible to perform continuous correction with high signal-to-noise ratio when the body drifts, this application provides a positioning correction control method for precise docking of mother-daughter shuttle cars.

[0006] In a first aspect, this application provides a positioning and correction control method for precise docking of a mother-daughter shuttle, comprising: simultaneously laying a QR code reference at each docking station, the QR code reference including a coarse reference set on the mother car rail and a fine reference set in the transition gap section at the entrance of the rack rail; connecting the incremental encoder pulse of the daughter car's front axle drive motor to the external trigger port of a vision sensor, triggering the vision sensor to capture a bottom view image according to the encoder pulse, obtaining the coarse code reading corresponding to the coarse reference as the absolute zero reference through the bottom view image at the moment when the daughter car has not yet left the mother car's cargo platform, and continuously obtaining the micro code reading corresponding to the fine reference through the bottom view image during the period when the daughter car crosses the transition gap section; The drift increment of the mother car relative to the departure time is used as a state variable. The drift differential observation is calculated using the microcode solution pose corresponding to the current microcode reading and the coarse code matrix solution pose corresponding to the coarse code reading at the departure time. The real-time drift is calculated by combining the improved extended Kalman fusion formula, and the real-time drift is fed into the differential compensation circuit of the left and right guide wheels of the daughter car for dynamic pose drift filtering and correction. When entering the conventional positioning QR code coverage area of ​​the rack aisle, the closing trigger signal is determined according to the first frame of rack aisle QR code captured by the vision sensor. The differential compensation amount is started to exit the process and stop the correction. The collected drift sequence is sent back to the host computer for archiving as the prior compensation value for the next docking.

[0007] By combining dual-scale spatial reference with hardware-level encoder-triggered sampling, time-driven operation is transformed into uniform position-driven operation. Differential observation is performed using relative drift increments, which effectively offsets the system common-mode bias and significantly improves the signal-to-noise ratio of drift filtering. Without disturbing the mother car braking closed loop, continuous high-frequency body posture accuracy correction is achieved by relying on the differential wheel of the daughter car.

[0008] Preferably, the step of simultaneously laying QR code references at each docking station includes: affixing a coarse reference using a calibration QR code matrix label to the mother car rail directly below the mother car parking station; reading the coarse reference at the moment the daughter car is stationary before leaving the mother car loading platform to obtain the coarsely calibrated pose as the absolute zero reference; and affixing a fine reference using micro QR code strips at equal intervals along the entire transition gap section at the entrance of the rack rail, ensuring that the transverse center lines of adjacent QR codes are strictly aligned with the central axis of the rack rail; wherein the load field of each micro code only encodes the unique number of the coordinate along the rail direction and the rack rail it is located on.

[0009] Preferably, the fine reference using micro QR code strips is affixed at equal intervals along the entire transition gap section at the entrance of the shelf rail, including: the value of the micro code spacing is matched with the longitudinal dimension of the single effective field of view of the visual sensor under the vehicle, so that at least two adjacent micro codes are captured simultaneously in any transient field of view; when more than two micro codes appear simultaneously in the field of view, the one with the smallest lateral pixel deviation is taken as the effective observation of the current frame and the rest are used as redundancy checks; the coarse reference and the fine reference share the same set of laser rangefinder and total station external reference during the laying stage, and the affixing is completed by the same calibration according to the same tooling reference.

[0010] Preferably, the step of connecting the incremental encoder pulses of the front axle drive motor of the vehicle to the external trigger port of the vision sensor, and triggering the vision sensor to capture the bottom view image according to the encoder pulses, includes: triggering the vision sensor to capture a frame of bottom view image for each multiple pulses output by the encoder, the number of pulses being matched according to the wheel circumference and the microcode spacing, so that two adjacent frames of images exactly cover two adjacent micro-QR codes respectively; reconstructing the one-dimensional time series image stream into a spatially sampled sequence that is strictly uniformly distributed along the track, the row index being determined by the microcode's track coordinates, and the column index being determined by the lateral pixel deviation of the microcode within the field of view relative to the center of the field of view; in event-driven mode, the camera shutter is triggered by the wheel position to lock the shutter phase, and the phase naturally locks as the wheel rolls.

[0011] By reconstructing a one-dimensional time-series image stream into a spatially sampled sequence that is strictly uniformly distributed along the rail direction, the sampling frame rate is adaptively adjusted with the vehicle speed. This avoids redundant accumulation in the low-speed range and avoids missing key frames in the high-speed range. At the same time, the shutter is naturally locked by the rolling phase of the wheels, effectively avoiding the worst transient vibration interference at the rail joint.

[0012] Preferably, the step of triggering the vision sensor to capture the bottom view image based on the encoder pulse further includes: when the vehicle briefly stops in the transition gap section due to an obstacle or control command, the encoder pulse output returns to zero and the vision sensor automatically stops triggering frame output; when the vehicle restarts, the encoder pulse output resumes seamlessly and the sampling link continues; if the encoder signal line experiences a brief circuit break due to vibration, the vehicle controller detects the trigger pulse timeout and automatically switches to a frequency reduction and fixed-period sampling backup mode, and switches back to the event-driven mode after the encoder signal recovers.

[0013] Preferably, the calculation of real-time drift using the improved Extended Kalman fusion formula includes: defining each component of the drift state vector as representing the lateral drift, track drift, and heading drift of the mother car relative to the coarse calibration attitude at the current moment; the initial state covariance matrix is ​​directly estimated from the static observation variance obtained by continuously collecting coarse reference readings in the stationary state of the subcar; the observation matrix in the Extended Kalman fusion formula degenerates into an identity matrix in the drift domain so that all common-mode components related to the absolute coordinates are natively canceled out in the difference operation.

[0014] By reducing the absolute pose to the relative drift increment and introducing drift differential observation, the state space of the Kalman filter is freed from the masking effect of the large cardinality of the absolute coordinates. The observation matrix is ​​degenerated into an identity matrix, which effectively cancels the posting translation error and sensor common mode bias, further improving the data stability of the correction process during the cross-slit transition period.

[0015] Preferably, the step of feeding the real-time drift into the differential compensation circuit of the left and right guide wheels of the subcar for dynamic pose drift filtering and correction includes: the Kalman gain in the improved formula is derived from the standard formula through a self-closed loop; the observation noise covariance is taken from the sample variance of the baseline noise sample obtained by collecting microcode readings for several consecutive seconds when the subcar is stationary directly above the coarse code matrix; the process noise covariance is taken from the upper bound of the variance calculated by the maximum driftable displacement of the mother car within a unit sampling interval allowed by the physical limit of the suspension compression stroke; the lateral component of the real-time drift corresponds to the instantaneous speed difference of the differential wheel, and the yaw component corresponds to the continuous speed difference of the differential wheel.

[0016] Preferably, the step of determining the closing trigger signal based on the first frame of the shelf aisle QR code captured by the vision sensor includes: the closing trigger signal can be determined to be valid only when both conditions are met simultaneously, namely, the vision sensor successfully decodes the shelf aisle QR code in two consecutive frames, and the horizontal deviation between the decoded aisle coordinates and the target storage location coordinates sent by the host computer does not exceed a preset threshold, thus eliminating interference from single-frame false triggering and blurry areas at the end of the transition gap; if the dual conditions are not met in three consecutive frames, the sub-cart controller determines that the closing is abnormal, actively decelerates to the crawling speed, and reports an alarm to the host computer, waiting for manual confirmation before continuing to advance.

[0017] Preferably, the differential compensation amount exit process to stop correction includes: the exit process adopts a first-order smooth transition, linearly decaying the differential compensation amount to zero within several frames; after the smooth exit is completed, the left and right wheel speeds of the sub-cart are resynchronized, and the control is switched to the conventional three-dimensional coordinate positioning stage of the rack aisle.

[0018] Preferably, the step of transmitting the collected drift sequence back to the host computer for archiving as a priori compensation value for the next docking includes: archiving the collected drift sequence in buckets according to the mother car number, row number, and load weight level; the host computer takes the average of the historical drift sequences in the same bucket as the coarse calibration priori compensation value for the next docking at that station; after the coarse calibration reading is completed before the next departure of the subcar, the controller superimposes the priori compensation value onto the coarse calibration pose, so that the initial drift estimation of the filter starts from the historical average.

[0019] By archiving the dynamic drift sequences of each operation in buckets and extracting the historical average as a priori compensation term for subsequent docking, the initial estimation of the filter directly inherits the historical pattern, effectively shortening the convergence time in the initial stage and realizing the gradual convergence and optimization of the correction accuracy of the same workstation as the number of operations accumulates.

[0020] This application effectively integrates the dual-scale code position reference and the encoder collaborative triggering sampling logic, getting rid of the blind spot and system jitter failure problems that are easily generated at the docking gap in traditional fixed-period visual recognition, and ensuring that the vehicle always has high redundancy and high spatial resolution pose extraction capability in the complex cross-gap transition section.

[0021] Furthermore, the drift differential observation and filtering process based on dimensionality reduction and reconstruction directly outputs the dynamic correction compensation amount to the inter-wheel differential circuit, which not only achieves adaptive suppression of pose drift error, but also avoids secondary disturbance to the braking state of the mother car. Combined with the prior data compensation method based on historical averages, the long-term operating accuracy and reliability stability of the equipment are significantly improved. Attached Figure Description

[0022] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart illustrating a positioning and correction control method for precise docking of mother and child shuttle vehicles according to the present invention.

[0023] Figure 2 A schematic diagram illustrating the deployment principle of a dual-scale spatial reference system provided in this application. Detailed Implementation

[0024] 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, not all, of the embodiments of the present invention. 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.

[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] This invention discloses a positioning and correction control method for precise docking of mother-daughter shuttle vehicles, referring to... Figure 1 This includes steps S1-S4: S1, Dual-scale code point reference laying.

[0027] In an optional embodiment, such as Figure 2 The diagram shown illustrates the deployment principle of the dual-scale spatial reference system, providing a clear view of its structure.

[0028] Two types of QR code references with different scales and functions are simultaneously laid at each docking station, forming a two-stage spatial reference system for coarse calibration before parking and fine tracking during gap crossing. The first type is the coarse reference, which uses existing 4×4 positioning QR code matrix labels and is affixed to the mother car rail directly below the mother car parking station. It is responsible for the static position of the mother car body relative to the rack rail entrance before the daughter car departs. The coarse calibration is the responsibility of the vehicle. The coarse reference is read at the moment when the vehicle is stationary and the trolley has not yet left the cargo platform of the mother vehicle. At this time, the mother vehicle has completed braking and locking, the suspension is in a static equilibrium state, the reading noise is minimal, and the obtained pose is used as the absolute zero reference for subsequent drift increment calculation.

[0029] The second type is a fine reference, defined as a micro QR code strip along the edge of the rail seam, with a side length of... Miniature QR codes are spaced evenly along the entire transition gap at the entrance of the shelf rail. The QR codes are affixed tightly, with the horizontal center lines of adjacent QR codes strictly aligned with the central axis of the shelf rail. The load field of each QR code only encodes its coordinates along the rail. The unique identifier is associated with the shelf track it is on. The microcode field design is extremely simple, deliberately omitting the complete information of the absolute coordinates on both the horizontal and vertical axes, retaining only the one-dimensional coordinates along the track. The purpose is to compress the decoding time of a single frame to the shortest possible time, in order to match the subsequent high-density rolling sampling rhythm and avoid the bloated code information from slowing down the recognition response.

[0030] Specifically, microcode spacing The value of must match the longitudinal dimension of the single effective field of view of the vehicle's under-vehicle vision sensor, ensuring that at least two adjacent microcodes are captured simultaneously within any transient field of view, thus guaranteeing code redundancy. For example, suppose... The value is one-third to one-half of the longitudinal dimension of the field of view. At this time, even if a single microcode fails to be decoded due to dirt or occlusion, the adjacent microcodes can still provide valid orbital coordinate readings without interrupting the continuous tracking link. When two or more microcodes appear in the field of view at the same time, the one with the smallest lateral pixel deviation is taken as the valid observation of the current frame, and the rest are used as redundancy checks.

[0031] Furthermore, the application process for both coarse and fine reference standards must meet specific constraints: both types of labels must share the same set of laser rangefinders and total station external reference standards during the installation phase, and be applied by the same calibration, the same work team, and the same tooling reference standards. After this treatment, the translational deviations of both types of labels relative to the common construction reference standard are strictly homogeneous, belonging to structural common-mode components; the steady-state biases within the visual sensor's dark current and other channels also appear homogeneously in both frames of readings. This constraint is the physical prerequisite for the native cancellation of the common tooling reference error and the sensor channel common-mode deviation in subsequent differential observations. If the two types of labels belong to different construction batches and are applied using different external reference standards, the common-mode cancellation prerequisite for differential analysis will not hold, and the subsequent filtering signal-to-noise ratio will significantly decrease.

[0032] In this way, the dual-scale code reference uses the division of labor of coarse code to establish zero position and fine code to measure drift, so that after the mother car stops, it no longer relies on a single static reading. Every millimeter of travel in the transition gap section has a corresponding spatial code element that can be queried, effectively filling the observation blind spot of traditional single-point calibration in the spatial dimension.

[0033] S2, encoder-triggered scrolling visual sampling.

[0034] In an optional embodiment, the incremental encoder pulses of the front axle drive motor of the vehicle are directly connected to the external trigger port of the vision sensor to construct an event-driven sampling link, replacing the traditional independent sampling mode of the vision sensor with a fixed period and fixed frame rate. Each encoder output... Each pulse triggers the vision sensor to capture a frame of bottom view image; the number of pulses... Based on wheel circumference and microcode spacing Matching ensures that two adjacent image frames exactly cover two adjacent micro QR codes, guaranteeing the uniformity of spatial sampling from a hardware perspective. For example, suppose the circumference of a wheel is... The number of pulses per encoder revolution is ,but Values The actual spatial step size after rounding is equal to The deviation does not exceed the displacement order of one pulse.

[0035] Specifically, this event-driven sampling reconstructs a traditional one-dimensional time-series image stream into a spatially sampled sequence that is strictly uniformly distributed along the track direction, with the row index determined by the microcode along the track coordinates. The column index is determined by the lateral pixel deviation of the microcode within the field of view relative to the center of the field of view. As the vehicle speed increases, the frame rate naturally increases synchronously, and when the speed approaches zero, the frame rate automatically returns to zero. In principle, this avoids the mismatch blind spot of traditional fixed-period sampling, which results in redundant frames at low speeds and missed key frames at high speeds. It also avoids the non-linear correlation deviation between camera exposure time and vehicle travel distance at the software level.

[0036] Furthermore, the longitudinal excitation of the ground rail joint at the transition gap to the vehicle causes slight jitter in the vision sensor itself. If the periodic sampling happens to fall on the peak phase of the jitter, the image will inevitably be blurred, and the code reading failure rate will increase significantly. In event-driven mode, the camera shutter is triggered by the wheel position rather than the system clock, and the shutter phase naturally locks as the wheel rolls. The worst transient of the rail joint excitation always corresponds to the instant when the wheel passes through the rail weld, but at that instant the camera has not yet been triggered to generate a frame. Therefore, the worst transient transmitted from the rail excitation to the vision channel is filtered out in the sampling link itself, without the need to add an additional anti-jitter filtering algorithm at the image processing level.

[0037] For fault-tolerant handling under extreme conditions, when the subcarriage briefly stops in the transition gap due to obstacles or control commands, the encoder pulse output returns to zero, and the vision sensor automatically stops triggering frame output, preventing redundant static frame accumulation. When the subcarriage restarts, the encoder pulse output resumes, and the sampling link is seamlessly connected without any software-level state reset operation. If the encoder signal line experiences a brief break due to vibration, the subcarriage controller detects the trigger pulse timeout and automatically switches to a reduced-frequency, fixed-period sampling backup mode to maintain basic drift observation capability with lower spatial resolution. It then switches back to event-driven mode after the encoder signal recovers.

[0038] In this way, the encoder hardware trigger transforms visual sampling from time-driven to position-driven, and the sampling density adapts to the vehicle speed, ensuring no frame loss at high speeds and no redundancy at low speeds. At the same time, it avoids the worst phase of rail joint excitation in the sampling link, providing a uniform and stable spatial sample sequence for subsequent drift observation.

[0039] S3, dynamic pose drift filtering and correction during the transition period.

[0040] In an optional embodiment, instead of using absolute pose as the state variable, the state space is directly reconstructed as the drift increment of the mother vehicle's body relative to the departure time. The drift state vector is defined as... Each component represents mother car relative to time The lateral drift, along-track drift, and heading drift of the attitude are coarsely calibrated at all times. This refers to the static instant when the tractor has not yet left the mother vehicle's cargo platform and the coarse baseline matrix is ​​read. The drift is used as the state variable, with its prior mean always being a zero vector. The initial state covariance matrix... The static observation variance is directly estimated from the coarse reference readings continuously collected while the vehicle is stationary, providing a basis for the filter's operation. It provides an objective statistical lower bound for the initial uncertainty at all times.

[0041] Specifically, this scheme makes targeted improvements to the standard Extended Kalman fusion formula. The original form of the existing Extended Kalman update formula is:

[0042] in For prior estimation, For the current frame observation, For the observation matrix, Kalman gain, which means the gain of the observed quantity. Weighted adjustments are made to the forecasts. The original formula assumes the observations are... For absolute pose readings, in this scenario, it means that the entire absolute pose read from the fine reference is fed into the filter in each frame. The absolute zero reference information carried by the coarse reference is diluted in consecutive frames. The filter must resolve sub-millimeter drifts against a background where the absolute coordinate values ​​are much larger than the drift variable, severely compromising the signal-to-noise ratio. To address the characteristic that the drift variable during the transition period across the slit is much smaller than the absolute pose, but the signal-to-noise ratio requirement is more stringent, this scheme reconstructs the observations into drift differential observations. That is, the current microcode solution pose. Pose calculation using coarse matrix at departure time The difference, the improved update relation is:

[0043] The relational form is consistent with the existing formula, but the semantics of the state variables are reduced from absolute pose to relative drift. Corresponding to the The difference between the frame microcode reading and the starting frame coarse code reading, rather than the absolute reading. The observation matrix no longer contains constants whose absolute coordinate values ​​are much larger than the drift variable. It degenerates into an identity matrix in the drift domain because of differential observations. With drift state The three-dimensional pose increments, which are relative to the starting time, correspond one-to-one and do not require linear mapping expansion. All common-mode components related to the absolute coordinates, including the translational deviation of the common tooling reference and the common-mode bias within the vision sensor channel, are natively canceled out in the difference operation. The filter only needs to resolve millimeter-level drift increments, resulting in a significant improvement in the signal-to-noise ratio.

[0044] Furthermore, the Kalman gain in the improved relation is... From the standard formula Self-closed-loop derivation; observation noise covariance The sample variance is taken from the baseline noise samples collected over several consecutive seconds when the vehicle is stationary directly above the coarse code matrix; the process noise covariance is also included. A separate, independent structure is constructed, taking the maximum drift displacement of the mother vehicle within a unit sampling interval allowed by the physical limits of the suspension compression stroke. The upper bound of the variance is calculated using this limit displacement, reflecting the physical reachability of the suspension itself. The data sources are strictly distinguished. When the suspension is sharply excited, the prediction uncertainty increases sharply, causing the prediction covariance to... Much larger than the observation noise covariance hour, Automatically approaches 1, the filter naturally weights the current observation; when the vehicle is at the rail weld excitation peak, observation variance When rising, Automatic fallback, automatic regression of filter to confidence prediction, and adaptive adjustment of gain throughout the process according to physical conditions.

[0045] Real-time drift of the filter output on the correction execution side The differential compensation circuit directly feeds into the left and right guide wheels of the subcar, with the lateral component... Corresponding instantaneous speed difference of differential wheel, yaw component The corresponding differential wheel speed difference forms a high-density rolling control that corrects the vehicle's posture once for every microcode passed. The entire gap crossing process does not require any positional correction from the mother car. The responsibility for correction is entirely borne by the wheel differential of the daughter car itself, avoiding secondary drift caused by disturbing the mother car's braking closed loop at the moment of gap crossing.

[0046] In this way, the state variables are reduced from absolute pose to drift increments, and the existing extended Kalman update formula is specifically improved in terms of observation semantics, so that all common mode deviations are natively canceled in the difference. The filter can maintain an effective signal-to-noise ratio at the millimeter-level drift variable level and form continuous real-time suppression of the micro-drift of the mother vehicle.

[0047] S4, Differential Exit and Prior Archive Closure.

[0048] In an optional embodiment, when the front axle of the trolley passes the last micro QR code in the docking gap and enters the area covered by the regular positioning QR code in the rack aisle, the first frame of the rack aisle QR code captured by the vision sensor serves as the closing trigger signal. Upon receiving this signal, the drift state... After the final update and archiving are completed, the differential compensation loop immediately initiates the exit process. The exit process employs a first-order smooth transition, linearly decaying the differential compensation amount to zero over several frames, rather than instantaneously clearing it to zero, thus avoiding new deflection disturbances caused by speed jumps. After the smooth exit is completed, the left and right wheel speeds of the sub-carriage are resynchronized, and control is switched to the conventional three-dimensional coordinate positioning stage of the rack aisle. Subsequent driving no longer relies on the drift domain filter.

[0049] Specifically, the closing trigger signal is considered valid only when both conditions are met simultaneously: the visual sensor successfully decodes the QR code of the aisle in two consecutive frames, and the lateral deviation between the decoded aisle coordinates and the target storage location coordinates sent by the host computer does not exceed a preset threshold. This eliminates interference from single-frame false triggers and ambiguous areas at the end of the transition gap. For example, assuming the lateral deviation threshold is one-quarter of the width of the aisle guide rail, the single-sided allowance of the trolley guide wheel relative to the inner side of the aisle rail is not less than the upper limit of the design tolerance of the gap between the guide wheel and the rail. Geometrically, this ensures that the guide wheel has completely crossed the edge of the rail gap and entered the effective engagement zone, without any geometric conditions for re-engaging the rail gap. If the above two conditions are not met simultaneously for three consecutive frames, the trolley controller determines that the closing is abnormal, actively decelerates to the crawling speed, and reports an alarm to the host computer. It waits for manual confirmation before continuing to advance, preventing the trolley from forcibly entering the aisle at normal speed when the closing status is unclear.

[0050] Furthermore, the subcar will collect the drift sequence during this seam crossing process. The data is transmitted back to the host computer and archived in bins according to the mother car number, row number, and load weight level. The host computer averages the historical drift sequences within the same bin and uses this average as the coarse calibration prior compensation value for the next docking at that station. After the coarse calibration readings are completed before the next departure of the daughter car, the controller superimposes this prior compensation value onto the coarse calibration pose, so that the initial drift estimation of the filter no longer starts from zero but from the historical average, shortening the convergence time of the filter in the initial stage of the gap crossing. As the number of dockings at the same station accumulates, the prior compensation value gradually approaches the systematic drift law of that station, and the correction accuracy shows a qualitative trend of gradual convergence.

[0051] When the daughter car returns to the mother car after completing this operation, it reuses the differential correction loop of this docking process. The return direction only reverses the reading order of the microcode sequence, without the need to develop independent control logic. The return drift sequence is also transmitted back for archiving and merged into the same bin as the de-program sequence to further enrich the prior statistical sample. The entire method consists of four stages: dual-scale code position, encoder-triggered sampling, drift domain Kalman filtering, and differential wheel correction, all integrated on the original electromechanical architecture of the mother-daughter shuttle. No additional actuators are added, and the mother car's main control loop is not modified. It can be implemented simply by adding a vision and encoder hard trigger link and a filter running code in the daughter car controller.

[0052] In this way, the differential wheel assumes all the responsibility for cross-gap correction, and the mother car braking closed loop does not need to be disturbed throughout the entire process, effectively avoiding the risk of secondary drift introduced by the correction action itself; the station-level archiving of the drift sequence further precipitates the observation results of a single cross-gap as a priori compensation term for the next docking at the same station, so that the correction accuracy shows a qualitative trend of gradual convergence with the cumulative number of dockings.

Claims

1. A positioning and correction control method for precise docking of mother-daughter shuttle vehicles, characterized in that, include: A QR code reference is laid at each docking station. The QR code reference includes a coarse reference set on the mother car rail and a fine reference set in the transition gap section at the entrance of the rack rail. The incremental encoder pulse of the front axle drive motor of the sub-carriage is connected to the external trigger port of the vision sensor. The vision sensor is triggered to capture the bottom view image according to the encoder pulse. At the moment when the sub-carriage is stationary before leaving the cargo platform of the mother car, the coarse code reading corresponding to the coarse reference is obtained through the bottom view image as the absolute zero reference. During the sub-carriage crossing the transition gap section, the micro code reading corresponding to the fine reference is continuously obtained through the bottom view image. The drift increment of the mother car body relative to the departure time is used as the state variable. The drift differential observation is calculated by using the microcode solution pose corresponding to the current microcode reading and the coarse code matrix solution pose corresponding to the coarse code reading at the departure time. The real-time drift is calculated by combining the improved extended Kalman fusion formula and feeding the real-time drift into the differential compensation circuit of the left and right guide wheels of the sub-car for dynamic pose drift filtering and correction. Upon entering the area covered by the regular positioning QR code in the rack aisle, the closing trigger signal is determined based on the first frame of the rack aisle QR code captured by the vision sensor. The differential speed compensation exit process is initiated to stop the correction, and the collected drift sequence is sent back to the host computer for archiving as the prior compensation value for the next docking.

2. The positioning and correction control method for precise docking of mother-daughter shuttles according to claim 1, wherein a QR code reference is simultaneously laid at each docking station, characterized in that, include: The coarse reference using the calibration QR code matrix label is affixed to the mother car rail directly below the mother car parking position. The coarse reference is read at the moment when the daughter car is stationary before leaving the mother car loading platform to obtain the coarsely calibrated pose as the absolute zero reference. The fine reference strips of micro QR codes will be closely affixed at equal intervals along the entire transition gap section at the entrance of the shelf rail, and the transverse center lines of adjacent QR codes will be strictly aligned with the central axis of the shelf rail. The load field of each microcode encodes only the unique number of the coordinate along the rail and the rail on the shelf it is located on.

3. The positioning and correction control method for precise docking of mother-daughter shuttles according to claim 2, wherein the fine reference strips of micro QR codes are applied at equal intervals along the entire transition gap section at the entrance of the rack rail, characterized in that... include: The value of the microcode spacing is matched with the longitudinal dimension of the single effective field of view of the vehicle's under-vehicle vision sensor, so that at least two adjacent microcodes are captured simultaneously in any transient field of view. When two or more microcodes appear in the field of view at the same time, the one with the smallest lateral pixel deviation is taken as the valid observation of the current frame and the rest are used as redundancy checks. During the laying phase, the coarse and fine references share the same set of laser rangefinders and total station external references, and are affixed using the same tooling references after the same calibration.

4. The positioning and correction control method for precise docking of mother-daughter shuttle vehicles according to claim 1, wherein the incremental encoder pulse of the front axle drive motor of the daughter vehicle is connected to the external trigger port of the vision sensor, and the vision sensor is triggered to capture the bottom view image according to the encoder pulse, characterized in that, include: The encoder outputs multiple pulses to trigger the vision sensor to capture a frame of bottom view image. The number of pulses is matched according to the wheel circumference and the micro-code spacing, so that two adjacent frames of images exactly cover two adjacent micro-QR codes respectively. The one-dimensional time series image stream is reconstructed into a spatially sampled sequence that is strictly uniformly distributed along the track. The row index is determined by the track coordinates of the microcode, and the column index is determined by the lateral pixel deviation of the microcode within the field of view relative to the center of the field of view. In event-driven mode, the camera shutter is locked by the position of the wheel, and the shutter phase locks naturally as the wheel rolls.

5. The positioning and correction control method for precise docking of mother-daughter shuttles according to claim 4, wherein the visual sensor captures the bottom view image based on encoder pulse triggering, is characterized in that... Also includes: When the vehicle stops briefly in the transition gap due to an obstacle or control command, the encoder pulse output returns to zero and the vision sensor automatically stops triggering frame output. When the vehicle restarts, the encoder pulse output resumes seamlessly, continuing the sampling link. If the encoder signal line experiences a brief break due to vibration, the sub-carriage controller will automatically switch to a frequency reduction and fixed-period sampling backup mode after detecting a trigger pulse timeout. It will then switch back to event-driven mode once the encoder signal recovers.

6. The positioning and correction control method for precise docking of mother-daughter shuttles according to claim 1, wherein the real-time drift is calculated using the improved extended Kalman fusion formula, characterized in that... include: The components of the drift state vector are defined to represent the lateral drift, track drift, and heading drift of the mother car relative to the coarsely calibrated attitude at the current moment, relative to the static instant. The initial state covariance matrix is ​​directly estimated from the static observation variance obtained by continuously collecting coarse reference readings under the stationary state of the vehicle; The observation matrix in the extended Kalman fusion formula degenerates into an identity matrix in the drift domain, causing all common-mode components related to the absolute coordinates to cancel each other out during the difference operation.

7. The positioning and correction control method for precise docking of mother and child shuttle vehicles according to claim 6, wherein the real-time drift is fed into the differential compensation circuit of the left and right guide wheels of the child vehicle for dynamic pose drift filtering and correction, characterized in that, include: The Kalman gain in the improved formula is derived from the standard formula through a closed loop, and the observation noise covariance is taken from the sample variance of the baseline noise sample obtained by collecting microcode readings for several consecutive seconds while the vehicle is stationary directly above the coarse code matrix. The process noise covariance is taken from the upper bound of the variance calculated by the maximum drift displacement of the mother vehicle within a unit sampling interval allowed by the physical limit of the suspension compression stroke; The lateral component of the real-time drift corresponds to the instantaneous speed difference of the differential wheel, while the yaw component corresponds to the continuous speed difference of the differential wheel.

8. The positioning and correction control method for precise docking of mother-daughter shuttle cars according to claim 1, wherein the termination trigger signal is determined based on the first frame of the shelf aisle QR code captured by the vision sensor, characterized in that, include: The closing trigger signal can be deemed valid only when both of these conditions are met simultaneously: the visual sensor successfully decodes the QR code of the shelf aisle in two consecutive frames, and the horizontal deviation between the decoded aisle coordinates and the target storage location coordinates sent by the host computer does not exceed a preset threshold. If the above dual conditions are not met for three consecutive frames, the sub-vehicle controller determines that the closing is abnormal, actively decelerates to the crawling speed, and reports an alarm to the host computer, waiting for manual confirmation before continuing to advance.

9. The positioning and correction control method for precise docking of mother-daughter shuttles according to claim 1, wherein the step of stopping the correction by exiting the differential compensation process is characterized in that, include: The exit process uses a first-order smooth transition, linearly decaying the differential compensation amount to zero within a few frames; After the smooth exit is completed, the speeds of the left and right wheels of the trolley are resynchronized, and control is switched to the conventional three-dimensional coordinate positioning process of the rack aisle.

10. The positioning and correction control method for precise docking of mother-daughter shuttles according to claim 1, wherein the acquired drift sequence is transmitted back to the host computer for archiving as a priori compensation value for the next docking, characterized in that, include: The collected drift sequences are archived in buckets according to the mother car number, row number, and load weight level; The host computer takes the average of the historical drift sequences within the same bin and uses it as the coarse calibration prior compensation value for the next docking at that workstation. After the coarse calibration readings are completed before the next vehicle departure, the controller will add the prior compensation value to the coarse calibration pose, so that the initial drift estimate of the filter starts from the historical mean.

Images