An automated guided vehicle laser communication link maintenance method

Abstract

The application discloses an automatic guided vehicle laser communication link maintaining method, and belongs to the technical field of wireless communication. The method comprises the following steps: establishing a line-of-sight angle coordinate system; obtaining corresponding line-of-sight prediction angles after multiple time intervals in the line-of-sight angle coordinate system by using a pre-constructed neural motion field line-of-sight prediction network; and controlling a vehicle-mounted laser transceiver pan-tilt and a fixed laser transceiver pan-tilt according to the line-of-sight prediction angles. By establishing a unified line-of-sight angle coordinate system, introducing a coordinate alignment layer, a motion mode extraction layer and a time coupling layer on the prediction network, aligning the output line-of-sight angle with the line-of-sight angle coordinate system, and directly outputting the line-of-sight prediction angles after the corresponding preset time intervals, the line-of-sight prediction angles are provided to a line-of-sight target trajectory construction for controlling the motion of the pan-tilt, so that the continuous laser communication link between the automatic guided vehicle and the fixed laser transceiver pan-tilt is maintained.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically to a method for maintaining a laser communication link in automated guided vehicles. Background Technology

[0002] With the widespread application of automated guided vehicles (AGVs) in smart warehousing and manufacturing workshops, high-bandwidth, low-latency data exchange is required between the vehicle control system and the host computer and fixed infrastructure. Compared with traditional radio frequency (RF) communication, laser communication has advantages such as strong resistance to electromagnetic interference and good confidentiality, making it suitable for use in factories with strong electromagnetic equipment or strict radio control. However, the small divergence angle of the laser beam makes it highly sensitive to the alignment of the lines of sight between the transmitter and receiver. When the AGV moves, turns, or changes its attitude, the optical axes at both ends can easily deviate, leading to communication link interruption. Therefore, how to stably maintain the laser link under complex motion conditions has become a key issue in practical engineering.

[0003] In existing technologies, one approach involves deploying a wireless local area network (WLAN) or other radio frequency network within the factory. This wireless network transmits the position and attitude information of the automatically guided vehicle, along with control commands. The laser link primarily handles data payload transmission, while the two gimbals at both ends perform coarse pointing based on pose information from the wireless network, supplemented by simple power feedback adjustments. Another approach attempts to achieve tracking solely through the intensity feedback of the laser signal itself. This often employs regular sector scanning, step-by-step search, or single-target tracking algorithms based on Kalman filtering and inertial prediction. The pointing angle is independently controlled in the local coordinate systems of the vehicle-mounted gimbal and the fixed gimbal, respectively. The actions of both ends are coordinated through a fixed master-slave relationship or simple timing to find and maintain optical axis alignment near the approximate line-of-sight direction.

[0004] The existing solutions mentioned above still have shortcomings in practical applications: On the one hand, the collaborative approach relying on wireless LANs or external radio frequency networks is not suitable for factory environments where wireless LANs are prohibited or inconvenient to deploy. Once the wireless network is unavailable, the pointing control of the laser link is difficult to operate reliably on its own. On the other hand, the rule-based scanning or simple prediction based solely on power feedback lacks a unified line-of-sight coordinate system and feedforward modeling of the movement patterns of automated guided vehicles. When the vehicle is traveling at high speed, making frequent turns, or being briefly obstructed, the gimbal often needs a long time to re-search for the optical axis, and the link is prone to multiple interruptions, making it difficult to meet the requirements of continuous communication. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method for maintaining a laser communication link between an automated guided vehicle (AGV) and a fixed laser transceiver, in order to solve the problem of how to predict the line of sight direction using the AGV's own operating status information in a factory environment without relying on a wireless local area network, construct the line of sight target trajectory under a unified line of sight angle coordinate system, and drive the coordinated movement of the two gimbals, so as to maintain a continuous laser communication link between the AGV and the fixed laser transceiver even under continuous vehicle movement and short-term lockout conditions.

[0006] To achieve the above objectives, the method of the present invention includes the following steps:

[0007] S1. Establish a line-of-sight coordinate system under the factory building for operating automated guided vehicles, and align and fix the position parameters of the laser transceiver.

[0008] S2. Control the fixed laser transceiver and the vehicle-mounted laser transceiver of the automatic guidance vehicle to perform area scanning according to the line-of-sight angle coordinate system, obtain the laser signal intensity during the area scanning process, and determine the coarse line-of-sight direction under the line-of-sight angle coordinate system according to the laser signal intensity.

[0009] S3. Input the fixed laser transceiver position parameters, coarse line of sight direction, current operating status information of the automatic guidance vehicle, the identifier vector of the fixed laser transceiver, and the time features corresponding to multiple preset time intervals into the pre-constructed neural motion field line of sight prediction network, and output the line of sight prediction angle after multiple time intervals in the line of sight angle coordinate system.

[0010] The pre-constructed neural motion field gaze prediction network includes an input layer, an input splicing layer, a coordinate alignment layer, a dimensionality increase layer, a motion pattern extraction layer, a temporal coupling layer, an angle regression layer, a multi-time interval output layer, and an angle bias synthesis layer arranged sequentially. The fixed laser transceiver position parameters and coarse gaze direction are used as construction constraints to configure the parameterization of the coordinate alignment layer and the angle bias synthesis layer, so that the output of the neural motion field gaze prediction network is consistent with the gaze angle coordinate system. The current operating status information of the automated guided vehicle, the identifier vector of the fixed laser transceiver, and the temporal features corresponding to the preset multiple time intervals are used as inputs to the input layer.

[0011] S4. Based on the predicted angles of the line of sight after multiple time intervals in the line of sight angle coordinate system, obtain the line of sight target trajectory used to control the movement of the gimbal.

[0012] S5. Control the vehicle-mounted laser transceiver gimbal and the fixed laser transceiver gimbal according to the line-of-sight target trajectory for controlling the movement of the gimbal.

[0013] Its beneficial effects are as follows: This invention establishes a unified line-of-sight coordinate system and performs rigid registration and alignment benchmark calibration on the fixed laser transceiver position, enabling the vehicle-mounted gimbal and the fixed gimbal to coordinate control within the same angle domain. The neural motion field line-of-sight prediction network of this invention automatically guides the vehicle's current position, attitude, speed, and steering commands (eight-dimensional operating states), mapping them together with the fixed laser transceiver identifier and time interval encoding to predict line-of-sight angles on three scales: short time, medium time, and long time. Structurally, this network introduces a coordinate alignment layer, a motion pattern extraction layer, and a time coupling layer to align the output line-of-sight angle with the line-of-sight coordinate system. By generating an angle offset from the coarse line-of-sight direction, it achieves consistency between the prediction result and the actual coarse direction. This allows for explicit utilization of repeated paths and vehicle motion patterns within the factory, providing advance indication of line-of-sight change trends before the vehicle completes a turn or accelerates / decelerates. This enables the gimbal control layer to perform feedforward adjustment in the form of a line-of-sight target trajectory, thereby reducing the number of laser link lock-up failures and re-acquisition time in complex motion conditions and short-term occlusion scenarios, and maintaining a continuous laser communication link with the fixed laser transceiver.

[0014] Further, in step S1, establishing the line-of-sight coordinate system under the automated guided vehicle (AGV) workshop includes: using the structural reference plane of the workshop as a reference, determining the origin, horizontal reference axis, and vertical reference axis of the line-of-sight coordinate system, wherein the horizontal reference axis is consistent with the positive direction of the heading angle of the AGV, setting the zero position of the line-of-sight angle and the zero position of the pitch angle, and setting the upper and lower limits of the angle value range of the line-of-sight coordinate system according to the boundary of the area in the workshop where the AGV can pass.

[0015] Further, in step S1, aligning the position parameters of the fixed laser transceiver includes: setting at least three non-collinear positioning markers, determining the coordinates of the positioning markers in the installation coordinate system and the line-of-sight angle coordinate system of the fixed laser transceiver, obtaining the rotation matrix and displacement vector from the installation coordinate system to the line-of-sight angle coordinate system based on the coordinates of the same positioning marker in different coordinate systems, and transforming the coordinates of the fixed laser transceiver in the installation coordinate system to the line-of-sight angle coordinate system based on the rotation matrix and displacement vector.

[0016] This invention establishes a unified line-of-sight coordinate system, aligning the fixed laser transceiver position parameters with the angle expression of the automated guided vehicle under the same reference. It also provides a consistent angle reference for the input splicing layer, coordinate alignment layer, and angle offset synthesis layer of the neural motion field line-of-sight prediction network, thereby supporting the subsequent implementation of line-of-sight target trajectory construction and gimbal-coordinated rotation control.

[0017] Further, in step S2, controlling the fixed laser transceiver and the vehicle-mounted laser transceiver of the automated guided vehicle to perform area scanning according to the line-of-sight coordinate system includes:

[0018] Based on the line-of-sight coordinate system, the scanning sector, step angle, and dwell time are set. With the fixed laser transceiver's position in the line-of-sight coordinate system as the center, the scanning boundaries of the azimuth and elevation angles are defined. The fixed laser transceiver's gimbal is controlled to emit laser light towards the center of the sector. The vehicle-mounted laser transceiver's gimbal is controlled to perform sector scanning at step angles. The fixed laser transceiver's gimbal is controlled to perform elevation scanning around the azimuth angle corresponding to the intensity peak during scanning by the vehicle-mounted laser transceiver.

[0019] Further, in step S2, determining the coarse line-of-sight direction in the line-of-sight angle coordinate system based on the laser signal intensity includes:

[0020] An intensity threshold is set, and samples below the threshold are eliminated when the onboard laser transceiver gimbal performs a fan-shaped scan at a step angle. The laser signal intensities of the retained samples are sorted in the line-of-sight angle coordinate system, and the azimuth angle corresponding to the intensity peak is selected as a candidate angle. The fixed laser transceiver gimbal is controlled to perform pitch scan with the candidate angle, and the laser signal intensity is collected. The coarse line-of-sight direction is determined in the line-of-sight angle coordinate system based on the angle corresponding to the intensity peak in the pitch scan.

[0021] In this invention, the fixed laser transceiver gimbal is first controlled to remain stationary, and the vehicle-mounted laser transceiver gimbal is automatically guided to scan to select the azimuth angle position corresponding to the peak value. Then, the fixed laser transceiver gimbal is controlled to perform a pitch angle scan around the azimuth angle. Based on the angle corresponding to the intensity peak value in the pitch micro-scan, the coarse line of sight is determined in the line of sight angle coordinate system. The coarse line of sight is then provided to the coordinate alignment layer and angle bias synthesis layer of the neural motion field line of sight prediction network for parameterization.

[0022] Further, in step S3, the input layer is used to acquire the current operating status information of the automated guided vehicle, the identifier vector of the fixed laser transceiver, and multiple preset time intervals; the input stitching layer is used to stitch together the information acquired by the input layer; the coordinate alignment layer is used to obtain an alignment vector that expresses the spatial relationship of the fixed laser transceiver in the line-of-sight coordinate system by using the position parameters of the fixed laser transceiver and the coarse line-of-sight direction as construction constraints, and the alignment vector and the stitching result obtained by the input stitching layer are used together to form the alignment input; the dimensionality enhancement layer is used to obtain multidimensional motion state features through multiple fully connected neurons; the motion pattern extraction layer is used to obtain multidimensional motion state features through multiple fully connected neurons. A two-level fully connected structure extracts multidimensional motion pattern features; a temporal coupling layer couples the temporal features with the multidimensional motion pattern features element by element through multiple fully connected neurons to obtain multidimensional time-related features; an angle regression layer compresses the multidimensional time-related features into angle features through multiple fully connected neurons; a multi-time interval output layer outputs the predicted gaze angles after multiple time intervals; and an angle bias synthesis layer generates the predicted gaze angles after multiple time intervals in the gaze angle coordinate system based on the coarse gaze direction.

[0023] Further, in step S3, generating the predicted viewing angles after multiple time intervals in the viewing angle coordinate system by using the multi-time interval output layer to output corresponding viewing angles according to the coarse viewing direction includes: generating an angle offset using the coarse viewing direction as input, adding the angle offset to the predicted viewing angles after multiple time intervals in the coarse viewing direction output layer to obtain a sum result; if the sum result is between the lower and upper azimuth boundaries of the viewing angle coordinate system, then the sum result is taken as the predicted viewing angle after multiple time intervals in the viewing angle coordinate system; if the sum result is less than the lower azimuth boundary of the viewing angle coordinate system, then the lower azimuth boundary of the viewing angle coordinate system is taken as the predicted viewing angle after multiple time intervals in the viewing angle coordinate system; if the sum result is greater than the upper azimuth boundary of the viewing angle coordinate system, then the upper azimuth boundary of the viewing angle coordinate system is taken as the predicted viewing angle after multiple time intervals in the viewing angle coordinate system.

[0024] During the operation phase of the neural motion field gaze prediction network, the current operating status information of the automated guided vehicle, the fixed laser transceiver identifier, and multiple preset time intervals are used as inputs. The output is the gaze prediction angles after multiple time intervals in the gaze angle coordinate system. Each gaze prediction angle after multiple time intervals in the gaze angle coordinate system is expressed in a unified gaze angle coordinate system and corresponds to its preset time interval, serving as the angle target point for the subsequent gaze target trajectory, supporting the time-series control of the gimbal movement. To ensure consistency with the training phase, the parameterization method of the coordinate alignment layer is configured during the construction phase, forming a fixed alignment constraint with the fixed laser transceiver position parameters and the coarse gaze direction; the angle bias synthesis layer uses the coarse gaze direction as the only input for bias generation, ensuring that the source of the angle bias is unique and stable. This enables the neural motion field gaze prediction network to form a continuous mapping from the current operating status information of the automated guided vehicle to the future gaze prediction angles within a unified gaze angle coordinate system, and the aligned gaze prediction angles directly serve the construction of the gaze target trajectory for controlling the gimbal movement.

[0025] Further, in step S4, obtaining the line-of-sight target trajectory for controlling the gimbal movement includes: determining the current line-of-sight angle based on the coarse line-of-sight direction, and combining the current line-of-sight angle with the predicted line-of-sight angles after multiple time intervals in the line-of-sight angle coordinate system in chronological order to form the line-of-sight target trajectory for controlling the gimbal movement.

[0026] Furthermore, in step S5, controlling the onboard laser transceiver gimbal and the fixed laser transceiver gimbal of the automated guided vehicle includes:

[0027] The laser follow control command is generated based on the line-of-sight target trajectory and the laser signal intensity. When the laser signal intensity is lower than the preset threshold, the local search direction is determined based on the line-of-sight prediction angle in the line-of-sight target trajectory, and the laser follow control command is updated.

[0028] According to the laser follow control command, the vehicle-mounted laser transceiver gimbal and the fixed laser transceiver gimbal of the automated guided vehicle are driven to rotate, forming a continuous laser communication link between the automated guided vehicle and the fixed laser transceiver.

[0029] Furthermore, in step S5, when the laser signal intensity is not lower than a preset threshold, the laser following control command is corrected according to the error between the current line of sight angle and the target angle, and the angular velocity is limited to not exceeding the upper limit of the angular velocity, so that the gimbal follows the trajectory of the line of sight target.

[0030] This invention integrates the aforementioned unified line-of-sight coordinate system and neural motion field line-of-sight prediction network with the construction of the line-of-sight target trajectory, the generation of laser following control commands, and the local search strategy, enabling the vehicle-mounted gimbal and the fixed gimbal to rotate collaboratively within the same line-of-sight coordinate system. When the link is normal, both gimbals smoothly track the line-of-sight target trajectory according to the time-calibrated parameters using limited angular velocity and angular acceleration. When the laser signal intensity is below a threshold, a unilateral local search angle sequence is automatically constructed around the line-of-sight prediction angle corresponding to the current time, and seamlessly returns to the original trajectory control after signal recovery is detected. This overall technical approach allows the system to maintain a continuous laser communication link with the fixed laser transceiver even under continuous vehicle movement and short-term lockout conditions, without relying on a wireless LAN or external radio frequency network, solely based on the vehicle's own operating status information and laser signal intensity feedback. This is particularly suitable for factory applications with strict limitations on the radio environment.

[0031] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0033] Figure 2 This is a flowchart illustrating the establishment of a unified line-of-sight coordinate system used in the method of this invention.

[0034] Figure 3 This is a flowchart of the area scanning and coarse line-of-sight direction determination method used in this invention;

[0035] Figure 4 This is a flowchart illustrating the construction and operation of the neural motion field gaze prediction network used in the method of this invention.

[0036] Figure 5 This is a flowchart of the line-of-sight target trajectory construction method used in the present invention;

[0037] Figure 6 This is a schematic diagram illustrating the process of maintaining the laser communication link of an automated guided vehicle in a factory environment under one scenario.

[0038] Figure 7 This is a schematic diagram illustrating the process of maintaining the laser communication link of the automated guided vehicle in another scenario within a factory environment, as described in this embodiment.

[0039] Figure 8 This is a schematic diagram illustrating the working process of the method in this embodiment using a three-scenario approach;

[0040] Figure 9 This is a diagram of the network model architecture used in the method of this embodiment. Detailed Implementation

[0041] The technical solution of the present invention will be clearly and completely described below with reference to specific embodiments. However, those skilled in the art should understand that the embodiments described below are only for illustrating the present invention and should not be regarded as limiting the scope 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.

[0042] Example of laser communication link maintenance method for automated guided vehicles

[0043] The innovation of this application lies in proposing an improved laser communication link maintenance method. By establishing a unified line-of-sight coordinate system in the factory environment and combining it with a coarse line-of-sight direction acquisition mechanism, coordinate alignment processing is performed on the operating status of the automated guided vehicle and the position of the fixed laser transceiver. Compared with existing schemes that independently control the gimbal in their respective local coordinate systems and rely solely on power feedback for sector scanning, this technique limits the range of azimuth and elevation angles under a unified coordinate system and uses the alignment reference vector obtained by rigid registration to compensate for the installation error of the fixed end. This allows for accurate local searching within the angular domain even after the vehicle turns or changes its attitude, reducing unnecessary global scanning and repeated pointing adjustments. This makes the pointing control of the laser transmitter and receiver more consistent with the actual constraints of the factory's traffic area and the mechanical travel of the gimbal, which is beneficial for maintaining a relatively stable line-of-sight geometry in the absence of wireless LAN assistance.

[0044] This application also employs a line-of-sight prediction method based on neural motion fields. The neural motion field line-of-sight prediction network maps the eight-dimensional operational state of the automatically guided vehicle (including its current position, attitude, speed, and steering commands) to fixed laser transceiver identifiers and time interval encoding, resulting in line-of-sight prediction angles at short, medium, and long time scales. Structurally, this network introduces a coordinate alignment layer, a motion pattern extraction layer, and a temporal coupling layer to align the output line-of-sight angle with a unified line-of-sight angle coordinate system. Furthermore, it achieves consistency between the predicted result and the actual coarse direction by using an angle offset generated from the coarse line-of-sight direction. Compared to existing prediction methods based on simple linear models or Kalman filters, this technique explicitly utilizes repetitive paths within the factory and vehicle motion patterns to provide the line-of-sight change trend before the vehicle completes a turn or acceleration / deceleration. This allows the gimbal control layer to perform feedforward adjustment in the form of the line-of-sight target trajectory, thereby reducing the number of laser link lock-offs and re-acquisition time in complex motion conditions and short-term occlusion scenarios.

[0045] The proposed method for maintaining a laser communication link for automated guided vehicles (AGVs) in factory environments integrates a unified line-of-sight coordinate system and a neural motion field line-of-sight prediction network with line-of-sight target trajectory construction, laser follow control command generation, and local search strategies. This allows the vehicle-mounted gimbal and the fixed gimbal to rotate collaboratively within the same line-of-sight coordinate system. When the link is normal, both gimbals smoothly track the time-calibrated line-of-sight target trajectory using limited angular velocity and angular acceleration. When the laser signal intensity falls below a threshold, a unilateral local search angle sequence is automatically constructed around the line-of-sight prediction angle corresponding to the current time, and seamlessly returns to the original trajectory control upon signal recovery. This overall technical approach enables the system to maintain a continuous laser communication link with the fixed laser transceiver, relying solely on the AVR's own operating status information and laser signal intensity feedback, without relying on a wireless LAN or external radio frequency network. This makes it particularly suitable for factory applications with strict limitations on radio environment.

[0046] like Figure 1 The method of this application includes:

[0047] S1. In a factory building where wireless LAN communication is not used, obtain the position parameters of a fixed laser transceiver and establish a unified line-of-sight coordinate system to obtain the line-of-sight coordinate system.

[0048] The process of step S1 is as follows: Figure 2 As shown, it includes:

[0049] In a factory building where wireless LAN communication is not used, the origin, horizontal reference axis, and vertical reference axis of a unified line-of-sight angle coordinate system are determined with reference to the factory building structure reference plane. The horizontal reference axis is aligned with the positive direction of the heading angle of the automated guided vehicle, and the zero position of the line-of-sight angle and the zero position of the pitch are set.

[0050] Obtain the position parameters of the fixed laser transceiver, represent the spatial position of the fixed laser transceiver in three dimensions, and define the direction consistent with the horizontal and vertical reference axes;

[0051] Align the fixed laser transceiver position parameters to a unified line-of-sight coordinate system, establish consistency between the spatial position of the fixed laser transceiver and the zero point of the line-of-sight angle, and form an alignment reference for the coordinate alignment layer.

[0052] Based on the boundary of the area accessible to automated guided vehicles within the factory, the upper and lower limits of the angle values ​​in the line-of-sight coordinate system are set to complete the establishment of a unified line-of-sight coordinate system.

[0053] S2. Based on the line-of-sight angle coordinate system, control the vehicle-mounted laser transceiver and the fixed laser transceiver to perform area scanning, receive the intensity of the laser signal propagating through a straight line, determine the coarse line-of-sight direction based on the laser signal intensity, and obtain the coarse line-of-sight direction.

[0054] The process of step S2 is as follows: Figure 3 As shown, it includes:

[0055] The scanning sector, step angle, and dwell time are set according to the line-of-sight coordinate system. The scanning boundaries of the azimuth and elevation angles are defined with the fixed position of the laser transceiver in the line-of-sight coordinate system as the center, and the intensity threshold is set.

[0056] The fixed laser transceiver gimbal is controlled to emit laser light towards the center of the sector. The vehicle-mounted laser transceiver gimbal is controlled to perform sector scanning at step angles. At each step angle, the intensity of the laser signal arriving through straight-line propagation is collected, and samples below the intensity threshold are discarded.

[0057] The intensity of the retained sampled laser signal is sorted in the line-of-sight coordinate system, and the azimuth angle corresponding to the intensity peak is selected as the candidate angle. The fixed laser transceiver gimbal is controlled to perform pitch micro-scan around the candidate angle to collect the laser signal intensity in the pitch direction and update the candidate angle.

[0058] Based on the angle corresponding to the intensity peak in the pitch microscan, the coarse line of sight is determined in the line of sight angle coordinate system. The coarse line of sight direction is then provided to the coordinate alignment layer and angle bias synthesis layer of the neural motion field line of sight prediction network for parameterization.

[0059] S3. Based on the automatic guidance vehicle's operating status information, fixed laser transceiver position parameters, and coarse line of sight direction, construct a neural motion field line of sight prediction network. The neural motion field line of sight prediction network takes the automatic guidance vehicle's current operating status information, fixed laser transceiver identifier, and preset time interval as input, and outputs the line of sight prediction angle after the preset time interval, thus obtaining the neural motion field line of sight prediction network.

[0060] The process of step S3 is as follows: Figure 4 As shown, it includes:

[0061] Using fixed laser transceiver position parameters and coarse line of sight as construction constraints, the parameterization method of coordinate alignment layer and angle offset synthesis layer is configured to make the output of neural motion field line of sight prediction network consistent with a unified line of sight angle coordinate system, and three preset time intervals are set.

[0062] The current operating status information of the automated guided vehicle is organized into eight dimensions: lateral position, longitudinal position, heading angle, pitch angle, roll angle, linear velocity, angular velocity, and steering command. The fixed laser transceiver identifier is mapped to a four-dimensional identifier vector through an identifier vector mapping layer, and the preset time interval is encoded to obtain a two-dimensional time feature.

[0063] The current operating status information of the eight-dimensional automated guided vehicle, the four-dimensional identification vector, and the two-dimensional time features are input into the input stitching layer to form a fourteen-dimensional input vector. Based on the fixed laser transceiver position parameters, a six-dimensional alignment vector is generated by the coordinate alignment layer and merged with the fourteen-dimensional input vector to form a twenty-dimensional aligned input.

[0064] The 20-dimensional aligned input is fed into the dimensionality-upgrading layer, and 128 fully connected neurons are used to obtain 128-dimensional motion state features. Then, through the two-level fully connected structure of the motion pattern extraction layer, the first layer has 256 fully connected neurons and the second layer has 128 fully connected neurons, and 128-dimensional motion pattern features are extracted.

[0065] Two-dimensional time features and 128-dimensional motion pattern features are coupled element-by-element in the time coupling layer. One hundred and twenty-eight fully connected neurons are used to obtain 128-dimensional time-related features. In the angle regression layer, sixty-four fully connected neurons are used to compress the 128-dimensional time-related features into angle features.

[0066] The angle features are input into the multi-time interval output layer, which outputs the three-dimensional line-of-sight prediction angles corresponding to short, medium and long time intervals. An angle offset synthesis layer is used to generate a one-dimensional angle offset based on the coarse line-of-sight direction, and the offset is synthesized with the three-dimensional line-of-sight prediction angle to obtain the three-dimensional line-of-sight prediction angle aligned with a unified line-of-sight angle coordinate system.

[0067] During the operation phase of the neural motion field line-of-sight prediction network, the current operating status information of the automatic guidance vehicle, the fixed laser transceiver identifier, and the preset time interval are used as inputs. The line-of-sight prediction angle after the preset time interval is directly output, and the line-of-sight prediction angle is provided to the line-of-sight target trajectory construction used to control the movement of the gimbal.

[0068] S4. Calculate at least one line-of-sight prediction angle based on the neural motion field line-of-sight prediction network and the current operating status information of the automated guided vehicle. Determine the current line-of-sight angle based on the coarse line-of-sight direction. Combine the current line-of-sight angle with at least one line-of-sight prediction angle in chronological order to form a line-of-sight target trajectory for controlling the movement of the gimbal.

[0069] The process of step S4 is as follows: Figure 5 As shown, it includes:

[0070] Using the lateral position, longitudinal position, heading angle, pitch angle, roll angle, linear velocity, angular velocity, steering command, and fixed laser transceiver identifier from the current operating status information of the automated guided vehicle as input, the angle regression layer and multi-time interval output layer of the neural motion field line-of-sight prediction network are called to obtain the line-of-sight prediction angles corresponding to short time intervals, medium time intervals, and long time intervals. The angle bias synthesis layer outputs the three-dimensional line-of-sight prediction angles aligned with the line-of-sight angle coordinate system. At the same time, the output is checked to be within the angle value range of the line-of-sight angle coordinate system. If the value exceeds the limit, it is truncated according to the boundary.

[0071] Within the line-of-sight angle coordinate system, determine the current line-of-sight angle based on the coarse line-of-sight direction, and set the current line-of-sight angle as the starting point and reference of the time series for subsequent line-of-sight target trajectory construction;

[0072] The predicted angle of the line of sight corresponding to the short time interval is placed after the current line of sight angle. The predicted angles of the line of sight corresponding to the medium time interval and the long time interval are added in ascending order of time to form an initial angle sequence. The angle difference and angular velocity reference of adjacent angle points are calculated, and the preset time interval to which each angle point belongs is determined.

[0073] Based on the upper limits of gimbal angular velocity and angular acceleration, the time step, the upper limit of angle difference, and the upper limit of angular velocity change are set to constrain the initial angle sequence. When the angle difference between adjacent angle points exceeds the upper limit of angle difference, the sequence is truncated according to the upper limit of angle difference, and a transition angle point is inserted between the two angle points to obtain a constrained angle sequence.

[0074] A short time interval priority strategy is adopted to divide the restricted angle sequence into segments. The starting segment consists of angles corresponding to short time intervals and transition angle points, and the angular velocity is kept continuous. The turning segment is mainly composed of angles corresponding to medium time intervals, and transition angle points are inserted at the beginning and end of the segment to achieve smooth angular velocity. The far segment is mainly composed of angles corresponding to long time intervals, and the difference between adjacent angles does not exceed the upper limit of the angle difference. The segments are rearranged into a time-ordered angle point sequence according to the segment order.

[0075] The current line of sight angle is combined with the time-ordered line of angle points to form the line of sight target trajectory. A unique time label is assigned to each angle point so that the line of sight target trajectory maintains a monotonic time order on the time axis and uses the line of sight angle coordinate system to express all angle values, covering the angle sequence from the starting point to the corresponding time range of the long time interval.

[0076] For each time calibration in the line-of-sight target trajectory, a corresponding line-of-sight prediction angle is determined. The local search direction is limited by this line-of-sight prediction angle. During execution, the corresponding angle values ​​are taken sequentially according to the time calibration, and the line-of-sight target trajectory is used as the angle reference input for controlling the gimbal movement to the next control process.

[0077] S5. Generate laser following control commands based on the target trajectory and laser signal intensity. When the laser signal intensity is lower than a preset threshold, determine the local search direction based on the predicted angle of the target trajectory and update the laser following control commands.

[0078] The process of step S5 includes:

[0079] The target trajectory is input into the control calculation according to the time calibration, the target angle at the current moment and the target angle at the next moment are extracted, the expected angular velocity and angular acceleration are calculated, and the initial laser following control command is generated, which includes the gimbal target angle and angular velocity, and is kept consistent with the line of sight angle coordinate system;

[0080] The laser signal intensity is collected and compared with a preset threshold. When the laser signal intensity is not lower than the preset threshold, the laser following control command is corrected according to the error between the current line of sight angle and the target angle, and the angular velocity is limited to not exceeding the upper limit of the angular velocity, so that the gimbal follows the trajectory of the line of sight target.

[0081] When the laser signal intensity is lower than the preset threshold, the local search direction is determined based on the predicted angle of the line of sight in the target trajectory. In the line of sight angle coordinate system, the unilateral angle offset and step angle are set around the predicted angle of the line of sight to construct a local search angle sequence.

[0082] The local search angle sequence is embedded into the laser follower control command. The corresponding angle and angular velocity are output sequentially according to the time calibration. The laser signal intensity is monitored in real time. When the laser signal intensity is not lower than the preset threshold, the local search is stopped and the angle and angular velocity corresponding to the line-of-sight target trajectory are restored.

[0083] S6. Drive the vehicle-mounted laser transceiver gimbal and the fixed laser transceiver gimbal to rotate according to the laser follow control command, so as to form a continuous laser communication link between the automatic guide vehicle and the fixed laser transceiver.

[0084] The process of step S6 includes:

[0085] The laser follow control command is parsed into the target azimuth angle, target elevation angle and corresponding angular velocity and angular acceleration in the line-of-sight angle coordinate system, and the control quantities of the vehicle-mounted laser transceiver gimbal and the fixed laser transceiver gimbal are generated according to the equipment allocation.

[0086] The two gimbals are driven to rotate based on the gimbal control input. Angle error closed-loop correction is used to convert the target azimuth and target pitch angles into motor actuation inputs, and the angular velocity and angular acceleration are limited to the set upper limit to keep them consistent with the target trajectory in line of sight.

[0087] During the rotation of the two gimbals, the intensity of the laser signal is collected and compared with a preset threshold. When the intensity of the laser signal is not lower than the preset threshold, the angle tracking is maintained according to the laser follow control command. When the intensity of the laser signal is lower than the preset threshold, the local search angle sequence in the laser follow control command is executed.

[0088] By synchronizing the target azimuth and elevation angles of the two gimbals according to the line-of-sight coordinate system, the transmitting optical axis and the receiving optical axis are aligned, and control quantities are continuously output to form a continuous laser communication link between the automated guided vehicle and the fixed laser transceiver.

[0089] The specific steps of the automatic guidance vehicle laser communication link maintenance method in this embodiment include:

[0090] S1. In this embodiment, a unified line-of-sight coordinate system is established in the factory building without using wireless local area network communication. The factory building structural reference plane F0 is used as the reference plane (F0 represents a structural plane with stable geometric properties). The origin O (O represents the geometric origin of this coordinate system) and the horizontal reference axis X are defined. h With vertical reference axis Z v With zero line of sight and pitch zero position As a reference for angle expression, the automated guided vehicle is stationary along the centerline of the channel, and the heading angle sensor output is read, making X... h Consistent with the positive direction of the heading angle, Z is determined by the F0 normal. v Move upwards to complete the construction of the angle reference plane.

[0091] Obtain the position parameters of the fixed laser transceiver end using P f =(x f ,y f ,z f ) describes spatial location, where x f Along X h Positive, y f With X h In the same plane and with X h Orthogonal, z f Along Z v Forward. To achieve coordinate alignment, a rotation matrix R and a displacement vector t are introduced, where R represents the rotation from the installation coordinate system to the unified line-of-sight angle coordinate system, and t represents the displacement of the origin of the two coordinate systems. A rigid registration process is used to calculate R and t: three non-collinear positioning markers M1, M2, and M3 are set on the fixed laser transceiver mounting surface, and their coordinates in the installation coordinate system are recorded. Corresponding points N1, N2, and N3 are measured in the unified line-of-sight angle coordinate system. R and t are obtained through rotation and translation registration with minimum mean square error. A threshold is determined for the registration residuals; if the residual exceeds the threshold, the measurement is repeated until the threshold requirement is met.

[0092] According to O and Pf The directional relationship is defined by the direction vector d. f To point from O to P f The vector representation, in d f In X h With Z v The projection relationship under the base determines the azimuth reference value of the fixed laser transceiver. Compared with the pitch angle reference value and encode it as an alignment reference vector V a As a parameter input to the coordinate alignment layer, it is used to align the current operating status information of the automated guided vehicle with the fixed laser transceiver position parameters in a unified line-of-sight coordinate system. The P parameter, after R and t transformation... f With V a Perform consistency checks to ensure that the angular representation and spatial location are valid under the same reference datum.

[0093] Based on the boundary set C of the accessible area for automated guided vehicles within the factory building lane The range of angle values ​​for the gimbal's mechanical stroke constraint settings. Where C... lane The set of polygon vertices representing the boundary of the passable area is derived from the fusion of structural lines on the factory construction drawings and boundary points acquired by laser ranging, and is expressed using a unified coordinate system. The azimuth range is set to... Set the pitch angle range to ,in and The turning radius of the passage, wall obstruction, and the upper limit of the pan-tilt unit's azimuth travel A are all factors. max limited, and The installation height, the upper beam and the upper limit of the gimbal's pitch angle travel B max limited. A max B represents the upper limit angle value of the mechanical travel of the gimbal in the azimuth direction. max This indicates the upper limit of the gimbal's mechanical travel angle in the pitch direction. A boundary consistency check is performed on the angle range to ensure that any angle value within a unified line-of-sight coordinate system falls within the above range.

[0094] Will The coordinate system definition and the uniform position P after R and t transformations. f Alignment reference vector V a and the range of angle values and The configuration output, solidified into a line-of-sight angle coordinate system, is provided to the coordinate alignment layer and serves as a unified reference for the angle offset synthesis layer to superimpose coarse line-of-sight directions. Through the above implementation method, a unified line-of-sight angle coordinate system is established, aligning the fixed laser transceiver position parameters with the angle expression of the automated guided vehicle under the same reference. This provides a consistent angle reference for the input stitching layer, coordinate alignment layer, and angle offset synthesis layer of the neural motion field line-of-sight prediction network, thereby supporting the subsequent implementation of line-of-sight target trajectory construction and gimbal-coordinated rotation control.

[0095] S2. The process of using a line-of-sight angle coordinate system to complete area scanning, acquire laser signal intensity, and determine the coarse line-of-sight direction is explained. The area scanning sector is set according to the line-of-sight angle coordinate system, and the azimuth angle of the sector center is denoted as... The pitch angle at the center of the sector is denoted as The azimuth reference of the fixed laser transceiver in the line-of-sight coordinate system is denoted as . The pitch angle reference is denoted as This is obtained from the alignment reference established when a unified line-of-sight coordinate system is used. The azimuth angle of the sector center is taken as... Take the pitch angle of the sector center The azimuth scan boundary is defined as follows: and The elevation angle scan boundary is limited to and The aforementioned boundaries all lie within the angular range of the line-of-sight coordinate system. The step angle is set to... Set the stay time as ,in It must be no less than the minimum resolvable angle of the pan-tilt unit and meet the requirements for discrete sector coverage. Not less than the receiver's integration sampling period. The intensity threshold is set to I. th The background intensity collected in the absence of emission is denoted as I. bg The minimum detectable intensity, combined with the receiver sensitivity, is denoted as I. min , will I th byI bg with I min This is jointly determined and used to eliminate unusable samples.

[0096] Control the fixed laser transceiver pan / tilt unit to point towards the sector center, using the sector center azimuth angle. pitch angle with sector center Continuous transmission is performed. The gimbal of the vehicle-mounted laser transceiver is controlled to perform a sector scan in step angles along the azimuth direction, while the elevation angle is maintained at its initial value near the center of the sector. Angle and intensity samples are recorded at each step angle position, and the azimuth angle is denoted as... The pitch angle is denoted as The laser signal intensity is denoted as I. i Where i is the sampling sequence number. Dwell time at each location. The intensity is acquired internally, forming a sampling set S composed of triples. Composition. Samples below the intensity threshold are discarded, retaining only those satisfying I. i Not lower than I th The samples are used for subsequent processing; the scanning order is from azimuth to... to The system advances monotonously, stopping scanning upon reaching the boundary and maintaining a safe gimbal position.

[0097] The intensity of the retained laser signals is sorted within the line-of-sight coordinate system, and the sample with the highest intensity is selected as a candidate. Let the index of the sample with the highest intensity be denoted as . The corresponding candidate azimuth angle is denoted as The corresponding pitch angle is near the current initial pitch angle. Control the fixed laser transceiver gimbal to orbit around the candidate azimuth angle. Perform pitch micro-scan, and record the micro-scan step angle as . Its value is less than And not less than the minimum resolvable angle of the gimbal's pitch angle; the pitch micro-scan range is limited to and Between these points, local sequences of pitch angles and corresponding laser signal intensities are obtained in order from the lower bound to the upper bound. The local sequences are sorted by intensity, and the candidate angles in the pitch direction are updated so that the laser signal intensity corresponding to the candidate angle reaches its peak value in the pitch direction. If a boundary is triggered during the pitch microscan, the candidate pitch angle is fixed at the trigger boundary to avoid exceeding the angle value range.

[0098] The coarse line-of-sight direction is determined in the line-of-sight angular coordinate system based on the angle corresponding to the peak intensity obtained from the pitch micro-scan. The azimuth angle corresponding to the peak intensity obtained from the pitch micro-scan is denoted as... The peak value corresponds to the pitch angle denoted as The coarse line of sight is taken as the one-dimensional angular offset reference and denoted as... ,Pick As an expression of this one-dimensional perspective, and The pitch direction verification reference is stored in the control layer. The coordinate alignment layer and angle bias synthesis layer of the neural motion field gaze prediction network are parameterized to ensure that the gaze prediction angle output by the network is consistent with the gaze angle coordinate system, and to support the construction of gaze target trajectory and the execution of gimbal motion control.

[0099] S3 describes the construction and operation of the neural motion field gaze prediction network, enabling the network to directly map the current operating status information of the automatically guided vehicle to the future gaze prediction angle, and maintain consistency with a unified gaze angle coordinate system. The neural motion field gaze prediction network structure used in this embodiment is as follows: Figure 9 As shown. The fixed laser transceiver position parameters are denoted as P. f , representing the three-dimensional position in a unified line-of-sight coordinate system; the coarse line-of-sight direction is denoted as . , representing a one-dimensional angular offset reference in a unified line-of-sight coordinate system; the alignment reference vector is denoted as V. a , representing a binary vector composed of a fixed laser transceiver azimuth reference and an elevation reference; three preset time intervals are set, denoted as . , , Based on the gimbal's dynamic response and the controller's clock resolution, the system is designed to cover short, medium, and long time interval prediction requirements. The current operating status information of the automated guided vehicle is organized into an eight-dimensional vector U, specifically representing the lateral position x. v Vertical position y v Heading angle Pitch angle Roll angle Linear velocity v v angular velocity Turning command u s The fixed laser transceiver identifier is denoted as e. A lookup table mapping layer is used to map e to obtain a four-dimensional identifier vector E. m The three preset time intervals are input into the time interval encoding layer respectively to obtain two-dimensional time features, which are denoted as follows: , , Each two-dimensional time feature corresponds to a preset time interval, which is used to participate in coupling operations in the time coupling layer.

[0100] The eight-dimensional vector U and the four-dimensional identifier vector E m With two-dimensional time features T c The vectors are sequentially input to the input concatenation layer to form a fourteen-dimensional input vector, denoted as H. 14 Within the coordinate alignment layer, based on the fixed laser transceiver position parameter P... f Alignment reference vector V a Generate a six-dimensional aligned vector, denoted as V. a6 This six-dimensional alignment vector, composed of a combination of position and orientation components, is used to express the spatial relationship of a fixed laser transceiver within a unified line-of-sight coordinate system. Specifically, it fixes the position parameter P of the fixed laser transceiver. fIt refers to the three-dimensional position components in a unified line-of-sight coordinate system, aligned with the reference vector V. a Given as a binary vector, it represents the azimuth and elevation references of a fixed laser transceiver, essentially corresponding to a unique spatial pointing line in this coordinate system. The coordinate alignment layer will... The "angle expression" can be deterministically converted into three-dimensional direction components, for example, to obtain the unit direction vector. Then, the three-dimensional position component and the three-dimensional orientation component are concatenated to form a six-dimensional aligned vector. Therefore, V a6 The six dimensions are formed by P f Position components and V a The mapped directional components are used to simultaneously constrain the spatial positional relationship and pointing reference of the laser transceiver within a unified line-of-sight coordinate system, thereby ensuring the consistency of angular semantics between subsequent network inputs and outputs.

[0101] H 14 With V a6 Merge to form a 20-dimensional aligned input, denoted as H. 20 The aligned representation is then fed into the subsequent network. The 20-dimensional aligned input is then fed into H. 20 The input layer is an up-dimensional layer, which uses 128 fully connected neurons to obtain 128-dimensional motion state features, denoted as F. 128 F 128 The input motion pattern extraction layer employs a two-level fully connected structure for non-linear extraction. The first layer has 256 fully connected neurons, and the second layer has 128 fully connected neurons. The output is a 128-dimensional motion pattern feature, denoted as M. 128 This is used to carry information about the operating mode and repeating paths of automated guided vehicles within the channel. The two-dimensional time features are coupled with M within the time coupling layer. 128 Perform element-by-element coupling. For three preset time intervals, respectively... , , With M 128 Coupling was used to obtain three sets of 128-dimensional time-related features using 128 fully connected neurons, denoted as follows: , , The aforementioned time-related features are input into the angle regression layer, which employs 64 fully connected neurons. Each group of 128-dimensional time-related features is compressed into angle features for output, denoted as follows: , , .

[0102] The angular features are input into the multi-time interval output layer, which outputs the predicted gaze angles for short, medium, and long time intervals, denoted as . In the angle-biased synthesis layer, with a coarse line-of-sight orientation... As input, a one-dimensional angular bias is generated, denoted as... The one-dimensional angle offset is combined with the three sets of sight prediction angles to obtain a three-dimensional sight prediction angle aligned with a unified sight angle coordinate system, denoted as . To ensure that the offset synthesis aligns with the boundary of a unified view angle coordinate system, a saturation constraint is applied to the boundary of the synthesis result in the angle offset synthesis layer. Specifically, the alignment angle for each time interval is defined by the following formula:

[0103] ;

[0104] in: This represents the line-of-sight prediction angle of the multi-time interval output layer over time interval p; Indicates angular offset, taken from the coarse line of sight. ; This represents the lower bound of the azimuth angle in a unified line-of-sight coordinate system. Represents the upper bound of the azimuth angle in a unified line-of-sight coordinate system; This represents the aligned line-of-sight prediction angle after boundary constraints; clip represents the saturation function, whose output is... Inside, when the input is less than Time to take When the input is greater than Time to take Otherwise, the input value is used. The above boundary values ​​are obtained by configuring the angle range of a unified viewpoint coordinate system. They are fixed during the construction phase to ensure that all alignment angles satisfy the coordinate system constraints during the runtime phase.

[0105] During the operation phase of the neural motion field gaze prediction network, the system will automatically guide the vehicle's current operating status information U, the fixed laser transceiver identifier e, and three preset time intervals. As input, U and e are processed through an identifier vector mapping layer and an input concatenation layer to form H. 14 , will P f With V a Generate V in the coordinate alignment layer a6 and H 14 Merge into H 20 M is obtained through the dimensionality enhancement layer and motion pattern extraction layer. 128 ,Will , , respectively with M 128 Formed in the time coupling layer , , Generated through the angle regression layer , , Y is obtained in the multi-time interval output layer and then synthesized in the angle bias layer. generate Complete the bias synthesis and output Y according to the above saturation constraint. align Y align Each predicted gaze angle is expressed in a unified gaze angle coordinate system and corresponds to a preset time interval, serving as the angle target point for the subsequent gaze target trajectory and supporting the time-series control of the gimbal motion. To ensure consistency with the training phase, the parameterization method of the coordinate alignment layer is configured during the construction phase, making it P f With V a Forming fixed alignment constraints; for angle-biased synthesis layers... As the sole input for bias generation, the source of the one-dimensional angle bias is guaranteed to be unique and stable; a static lookup table method is used for the identifier vector mapping layer to ensure that the fixed laser transceiver identifier e maps to the same four-dimensional identifier vector E in different operating cycles. m Through the above implementation method, the neural motion field gaze prediction network forms a continuous mapping from the current operating status information of the automated guided vehicle to the future gaze prediction angle within a unified gaze angle coordinate system, and directly serves the construction of the gaze target trajectory for controlling the gimbal movement with three sets of aligned gaze prediction angles.

[0106] S4 describes the construction of the line-of-sight target trajectory used to control gimbal movement. It organizes a time-series of angle points around the output of the neural motion field line-of-sight prediction network and completes constraints, segmentation, and time calibration within the line-of-sight angle coordinate system. The current operating status information of the eight-dimensional automated guided vehicle is denoted as U, including lateral position, longitudinal position, heading angle, pitch angle, roll angle, linear velocity, angular velocity, and steering command; a fixed laser transceiver identifier is denoted as e; and a coarse line-of-sight direction is denoted as... When the neural motion field gaze prediction network is invoked, the angle regression layer and the multi-time interval output layer are triggered to obtain the gaze prediction angles corresponding to short, medium, and long time intervals, denoted as . The 3D line-of-sight prediction angle has been determined by the angle offset synthesis layer. Alignment complete. Perform angle value range verification on the 3D line-of-sight prediction angle to ensure the output falls within the lower bound of the azimuth angle in the line-of-sight angle coordinate system. and the upper limit of azimuth Between, among and The coordinate system configuration is given by the line-of-sight angle.

[0107] Determine the current line-of-sight angle within the line-of-sight angle coordinate system based on the coarse line-of-sight direction, denoted as . Let the current time be t0. This serves as the starting point and baseline for the time series, used for subsequent line-of-sight target trajectory construction. To ensure consistency with the temporal semantics of the network prediction, three preset time intervals are used, denoted as follows: , , These three factors originate from the controller clock and the dynamic response configuration of the PTZ.

[0108] When constructing the initial angle sequence, the angles corresponding to short time intervals are... Placed The angle corresponding to the intermediate time interval Angle corresponding to long time intervals Placed separately and This forms an initial angle sequence that increases over time, denoted as... Calculate the angle difference between adjacent angle points in the initial angle sequence, denoted as . The unit is angle; the time difference between adjacent times is... The ratio is used to obtain the angular velocity reference, denoted as . The unit is angles per second, and the preset time interval for each angle point is recorded for subsequent segmentation.

[0109] The initial angle sequence is constrained based on the upper limits of the gimbal's angular velocity and angular acceleration. Let the upper limit of angular velocity be denoted as... The unit is angles per second; let the upper limit of angular acceleration be denoted as . The unit is angles per second squared; let the upper limit of the angle difference be denoted as . The unit is angle; let the upper limit of the change in angular velocity be denoted as . The unit is angles per second; the selected time step is denoted as... The unit is seconds, which is not less than the controller clock resolution and is consistent with the gimbal driver update cycle. To determine the required number of transition angle points between two angle points, the minimum insertion number is calculated using the following formula, and insertion and time allocation are performed accordingly:

[0110] ;

[0111] Where: N i This represents the minimum number of transition angle points to be inserted between the i-th and (i+1)-th angle points in the sequence, and is a dimensionless integer. This represents the angle difference between adjacent angle points, expressed in degrees. This indicates the upper limit of angular velocity, measured in degrees per second. This indicates the upper limit of angular acceleration, expressed in degrees squared per second. Indicates the time step, in seconds; This represents the floor function; This indicates that the larger of the two input values ​​is selected. Represents absolute value; This represents the square root operation. When... N is evenly inserted between the two angle points. i There are several transition angle points, such that the time interval between adjacent transition pairs is... Furthermore, the transition angles are allocated in a monotonic direction to ensure that the angle difference and angular velocity change do not exceed the corresponding upper limit.

[0112] After the insertion is completed, a restricted angle sequence is obtained, denoted as . For constrained angle sequences, a short time interval priority strategy is used to divide them into segments. A starting segment, denoted as S, is constructed. seg ,Depend on and near It consists of the restricted angle points and the transition angle points between them, requiring the angular velocity to be continuous within the segment and not exceeding [a certain value]. Construct a transition segment, denoted as M. seg to get closer The focus is on the limited angle points, inserting a minimum number of transition angle points at the beginning and end of each segment to smooth the angular velocity and satisfy the requirements. Constraints. Construct the remote segment, denoted as L. seg to get closer The focus is on the limited angle points, keeping the difference between adjacent angles no more than [missing information]. S seg M seg L seg Rearrange and connect the points in chronological order to obtain a time-ordered sequence of angle points, denoted as . .

[0113] The current line-of-sight angle is combined with the time-ordered sequence of angle points to form the line-of-sight target trajectory, denoted as . .right Each angle point in the equation is assigned a unique time calibration, denoted as t. j The time calibration is required to be monotonically increasing and cover the interval. .right All angle values ​​in the coordinate system are expressed using the line-of-sight coordinate system, ensuring that the angle falls within the coordinate system at any given time. Within the range. Once completed, As a time-series angular reference for the gimbal's movement, it can be read by the control layer in a time-calibrated order.

[0114] For each time point in the target trajectory, determine the corresponding predicted line-of-sight angle, denoted as . When t j equal , or At that time, take respectively , or When t jWhen the time falls within the above-mentioned time range, the method used is... Interpolating adjacent angle points within a linear time interval yields the results. And maintain consistency with the restricted constraints. Limiting the local search direction, denoted as During execution, data is retrieved sequentially according to the time marker. The angle value in the middle, and will As an angle reference input for controlling the movement of the gimbal, the subsequent control process enables the gimbal to respond in advance within a short time interval, complete the transition smoothly at a medium time interval, and follow stably within a long time interval, while maintaining consistency with the output of the neural motion field line-of-sight prediction network.

[0115] S5 explains the use and local search embedding of the line-of-sight target trajectory for controlling gimbal movement at the runtime layer. Let the line-of-sight target trajectory be denoted as... It consists of a time calibration and an angle value; the time calibration is denoted as t. j Let represent the j-th time point in the trajectory; let be the trajectory angle value. , indicating time calibration t j The corresponding line-of-sight angle; recorded as a control cycle. , representing the time length of a single controller update. In control calculations, the current time is extracted and denoted as t. k Let t be the time interval between the next two moments. k+1 The target angles are denoted as follows: and Let the desired angular velocity be denoted as It is calculated from the angles of the two targets and the corresponding time difference; the desired angular acceleration is denoted as... The expected angular velocity of adjacent angular velocities and Calculation. Generate the initial laser follower control command, denoted as C0, including the gimbal target angle, denoted as... The angular velocity of the target gimbal is denoted as And maintain the line-of-sight coordinate system as the reference for angle expression.

[0116] The intensity of the acquired laser signal is denoted as I(t) k ), and the preset threshold is denoted as I. th Compare. When I(t) k Not lower than I th At that time, calculate the error between the current line-of-sight angle and the target angle, and record the error angle as . The current line of sight angle is denoted as The angle is read by the gimbal angle encoder. The error angle is used to correct the laser following control command, and the gimbal target angular velocity is weighted and adjusted, with the gain denoted as . The unit is per second per angle, obtained by controller tuning. The angular velocity is limited to a maximum angular velocity, denoted as . The unit is angles per second; if the corrected angular velocity exceeds the limit, it will be truncated to the upper limit. Keep the gimbal pressed. The time sequence and angle value follow.

[0117] When I(t) k (Below I) th At that time, the local search direction is determined based on the predicted angle of the line of sight in the target trajectory. This will be done at time t. k The predicted line of sight angle is denoted as Recorded during the target trajectory construction phase; the local search direction is denoted as D. loc (t k ), set as and Unilateral directions with consistent signs; unilateral angular offset is denoted as The unit is angle; the step angle is denoted as... The unit is angle; the number of local search steps is denoted as N. loc , which is a dimensionless integer. Around In direction D loc (t k ) and step angle Construct a local search angle sequence, denoted as Its angle value increases monotonically and does not exceed And assign continuous time calibration to each local angle point, with the time interval being... .

[0118] The local search angle sequence is embedded into the laser follower control command. At each local angle point, the gimbal target angle is set and denoted as . Set the gimbal target angular velocity as . Its size does not exceed Real-time monitoring of laser signal intensity, when I(t) k Not lower than I th Immediately stop the local search and resume using the angle and angular velocity corresponding to the target trajectory along the line of sight. During the recovery process, record the current time as t. r ,from Select with t r The most recent time setting, denoted as t j Take the corresponding angle As a new and the angle difference calibrated by adjacent time points Calculate new Continue to output control commands in a time sequence.

[0119] To ensure consistent quantitative constraints throughout the control process, angle values ​​are expressed using a line-of-sight coordinate system at all stages, and numerical parameters with consistent units are used for angular velocity and angular acceleration; , , , N loc The time calibration is given during the construction phase and remains unchanged during the runtime phase; a monotonically increasing sequence is used for the time calibration, maintaining consistency with... Consistency ensures that the gimbal remains in place even when unlocked. Fast convergence, seamless return to normal during recovery Time series control.

[0120] S6. This section explains how to execute laser follow control commands at the operating layer and drive the two gimbals to rotate.

[0121] Record the trajectory of the line of sight as It consists of a time calibration and an angle value; the time calibration is denoted as t. j Let represent the j-th time point in the trajectory; let the trajectory angle value be denoted as . , indicating time calibration t j The corresponding line-of-sight angle; the peak pitch in the coarse line-of-sight direction is denoted as... The pitch reference angle is denoted as C; the laser follow control command is denoted as C, representing the angle and velocity target issued from the control layer; the control period is denoted as... This indicates the duration of a single update by the controller.

[0122] The laser tracking control command is resolved into the target azimuth and target elevation angles within the line-of-sight coordinate system. The target azimuth is denoted as... , set as The target pitch angle is denoted as... , set as Let the target angular velocity be denoted as and The target angle determined by adjacent time intervals and Calculate; denote the target angular acceleration as and The angular velocities of adjacent targets and Calculate and generate control variables according to equipment allocation.

[0123] The control quantity of the gimbal for the onboard laser transceiver of the automated guided vehicle is denoted as... ,set up , .

[0124] The control value of the fixed laser transceiver gimbal is denoted as... ,set up , .

[0125] The two gimbals are driven to rotate based on the gimbal control input. The current azimuth angle of the vehicle-mounted gimbal for the automated guided vehicle is denoted as... The current pitch angle is denoted as The angle is read by the gimbal angle encoder; the current azimuth angle of the fixed laser transceiver gimbal is recorded as... The current pitch angle is denoted as Angle error closed-loop correction is adopted. The azimuth error is denoted as... , The pitch angle error is denoted as... , Let the upper limit of angular velocity be denoted as , representing the maximum permissible angular velocity per unit time; the upper limit of angular acceleration is denoted as , representing the maximum permissible rate of change of angular velocity per unit time. , , , Apply no more than Restrictions on , , , Apply no more than Restrictions.

[0126] The motor's input is synthesized based on error and limited speed and acceleration, and then output to the gimbal driver to make the gimbal rotate and... Maintain consistency.

[0127] The intensity of the laser signal was collected during the rotation of the two gimbals.

[0128] Let the laser signal intensity be denoted as I(t). j The data is collected by the current receiving side photoelectric receiver; the preset threshold is recorded as I. th , indicating the lower bound of the intensity. If I(t) j Not lower than I th Then, according to the laser follow control command, the angle tracking is maintained, and U is continuously output. v with U f If I(t) j (Below I) th Then, the local search angle sequence in the laser follower control command is executed. Let the local search angle sequence be denoted as... In the azimuth channel, local angles and angular velocities are sequentially taken according to time calibration, and the pitch angle is maintained. Execute simultaneously on both gimbals until I(t) j Not lower than I th Stop local search and resume using The corresponding angle and angular velocity.

[0129] The target azimuth and target elevation angles of the two gimbals are synchronized according to the line-of-sight coordinate system to align the transmitting optical axis with the receiving optical axis.

[0130] Set the synchronization strategy to synchronize with the same angle value, that is... , In each t j Continuous output U v with U f This ensures that the two gimbals are aligned in the same direction within the line-of-sight coordinate system. Through the above control process, a continuous laser communication link is established between the automated guided vehicle and the fixed laser transceiver, ensuring that the optical axis alignment remains stable and consistent over time.

[0131] like Figure 6 as well as Figure 7 The diagram illustrates the process of maintaining a laser communication link for an automated guided vehicle (AGV) in a factory environment using the method described in this embodiment. The gray area represents the factory and shelving aisles, and the right wall shows the fixed laser transceiver. Figure 6 as well as Figure 7 The multiple poses represent the historical and current positions of the automated guided vehicle (AGV) within the channel. The blue beams indicate the laser communication link between the fixed end and the AGV; the historical positions are displayed semi-transparently, indicating that the link remains continuous and uninterrupted during vehicle movement. The purple dashed lines and discrete points on them represent the line-of-sight target trajectory generated based on a neural motion field line-of-sight prediction network, enabling advance adjustment of the gimbal's pointing before vehicle turns and acceleration / deceleration. This invention combines geometric constraints, vehicle motion states, and gimbal dynamics through a unified line-of-sight angle coordinate system and line-of-sight prediction trajectory, achieving forward-looking, smooth, and robust maintenance of the laser link in complex factory environments without relying on wireless LANs.

[0132] like Figure 8 The diagram illustrates the working process of the method in this embodiment using a three-scenario approach. The left side shows the initial coarse line-of-sight acquisition scenario: the automated guided vehicle (AGV) establishes an initial link with the fixed laser transceiver. The vehicle-mounted gimbal performs a region scan within a certain sector, and the fixed end emits a probe beam. After multiple weak echo comparisons, the beam gradually converges to the brightest beam, forming the coarse line-of-sight direction. The middle section shows the prediction-driven smooth tracking scenario: driven by a neural motion field line-of-sight prediction network, the system provides the target trajectory of the line-of-sight in advance based on the vehicle's operating status. The gimbal rotates continuously along the smoothly curved predicted trajectory, and the laser link remains stable during vehicle movement and turning. The right side shows the local search and recovery scenario: when channel obstruction or sudden attitude changes cause signal attenuation, the system no longer blindly rescans the entire system but instead quickly searches within a narrow local sector around the predicted direction. Once the signal is recaptured, stable tracking is restored. Through the synergy of "coarse acquisition + prediction tracking + local search," this invention significantly improves the continuity, response speed, and robustness of the laser link in complex factory environments and without wireless LAN assistance, while reducing invalid scans and improving overall control efficiency.

[0133] Compared to the closest existing technology (which relies solely on laser power feedback, performs sector scanning in their respective local coordinate systems, and uses a simple linear / Kalman prediction-based link-keeping method), this application establishes a unified line-of-sight coordinate system and performs rigid registration and alignment benchmark calibration on the fixed laser transceiver positions, enabling coordinated control of the vehicle-mounted gimbal and the fixed gimbal within the same angular domain. This design pre-defines the effective range of azimuth and pitch angles within the factory traffic area and under the constraints of gimbal mechanical travel. The coarse line-of-sight direction is locked and fixed as an offset benchmark through a single area scan. Simulation and prototype testing show that, under the same scanning step angle conditions, the effective search sector area of ​​this application's scheme can be reduced by approximately 50%–65% compared to traditional global sector scanning. Under typical operating conditions, the re-acquisition time after a single lock-off is reduced from approximately 1.0 s to less than approximately 0.3 s. Simultaneously, since frequent large-scale reciprocating angle scans are unnecessary, the peak angular velocity and cumulative mechanical travel of the gimbal's azimuth drive are significantly reduced, which is beneficial for improving the service life and pointing stability of the gimbal actuator.

[0134] On the other hand, this application introduces a neural motion field line-of-sight prediction network, which jointly models the pose, linear velocity, angular velocity, and steering commands of the automatically guided vehicle with fixed-end identifiers and time interval encoding. It outputs line-of-sight prediction angles at short, medium, and long time scales, and aligns them with the coarse line-of-sight direction in the angle bias synthesis layer, further constructing a line-of-sight target trajectory constrained by angular velocity and angular acceleration. During link maintenance, the gimbal uses this trajectory as a feedforward reference, smoothly tracking the laser signal in a time sequence when it is normal, and automatically generating a one-sided local search sequence around the predicted angle at the corresponding moment when the signal attenuates or is obstructed. Compared with existing technologies that rely solely on passive search based on power feedback, in channel scenarios with small vehicle turning radii and rapid attitude changes, the link availability rate of this application's solution is increased from approximately 92% to over 99%. The average number of link interruptions per task is reduced by approximately 70% during multiple round trips. Without relying on wireless LANs or external radio frequency networks, it achieves continuous laser communication link maintenance capabilities more suited to factory scenarios.

[0135] With the development of intelligent manufacturing and smart logistics, the number of automated guided vehicles (AGVs) in warehousing, workshop delivery, and flexible production lines continues to increase, placing higher demands on the real-time performance and reliability of data interaction between equipment within the factory. In some factories with strong electromagnetic interference or strict radio control, the deployment of traditional wireless local area networks is limited, making short-range laser communication, with its high security and anti-interference capabilities, a promising application area. This invention provides a technical approach for maintaining link stability during continuous vehicle movement in such scenarios, demonstrating considerable market potential.

[0136] This invention exhibits good compatibility with existing automated guided vehicle (AGV) control systems, scheduling systems, and onboard sensor platforms, with a relatively clear technology implementation path. On one hand, this solution does not rely on factory-wide wireless LANs, serving as a communication supplement for specific process areas where wired / wireless networks are difficult to cover or where radio frequency (RF) technology is unsuitable. This facilitates deployment in high-intensity motors, areas with high electromagnetic interference, or confidential workshops. On the other hand, through a unified line-of-sight coordinate system and a neural motion field line-of-sight prediction network, the continuity and maintainability of the laser link can be improved without significantly altering the existing vehicle control architecture. This makes it suitable for differentiated functional configurations in new projects and for modular upgrades of existing AGV systems.

[0137] From an industry perspective, the link maintenance mechanism proposed in this invention is also attractive to other AGV manufacturers, system integrators, and suppliers of dedicated laser communication equipment. For users in industries such as steel, petrochemicals, and power that have management requirements for radio environments, third-party companies can incorporate this technology as part of their high-reliability in-plant logistics solutions. For integrated projects that have already deployed multi-brand AGVs, the unified line-of-sight coordinate system and predictive control approach provided by this invention can also be easily integrated or licensed by other manufacturers through standardized interfaces.

[0138] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for maintaining a laser communication link in an automated guided vehicle, characterized in that, Includes the following steps: S1. Establish a line-of-sight coordinate system under the factory building for operating automated guided vehicles, and align and fix the position parameters of the laser transceiver. S2. Control the fixed laser transceiver and the vehicle-mounted laser transceiver of the automatic guidance vehicle to perform area scanning according to the line-of-sight angle coordinate system, obtain the laser signal intensity during the area scanning process, and determine the coarse line-of-sight direction under the line-of-sight angle coordinate system according to the laser signal intensity. The process of controlling the fixed laser transceiver and the vehicle-mounted laser transceiver of the automated guided vehicle to perform area scanning according to the line-of-sight coordinate system includes: setting the area scanning sector, step angle, and dwell time according to the line-of-sight coordinate system; defining the scanning boundaries of azimuth and elevation angles with the azimuth of the fixed laser transceiver in the line-of-sight coordinate system as the center; controlling the fixed laser transceiver gimbal to emit laser light towards the center of the sector; controlling the automated guided vehicle's vehicle-mounted laser transceiver gimbal to perform sector scanning at step angles; and controlling the fixed laser transceiver gimbal to perform elevation scanning around the azimuth angle corresponding to the intensity peak value when the automated guided vehicle's vehicle-mounted laser transceiver is scanning. Determining the coarse line-of-sight direction in the line-of-sight angle coordinate system based on the laser signal intensity includes: setting an intensity threshold, eliminating samples below the intensity threshold when the onboard laser transceiver gimbal performs a fan-shaped scan at a step angle in the automatic guidance vehicle, sorting the laser signal intensity of the retained samples in the line-of-sight angle coordinate system, selecting the azimuth angle corresponding to the intensity peak as a candidate angle, controlling the fixed laser transceiver gimbal to perform pitch scan with the candidate angle, and collecting the laser signal intensity, and determining the coarse line-of-sight direction in the line-of-sight angle coordinate system based on the angle corresponding to the intensity peak in the pitch scan; S3. Input the fixed laser transceiver position parameters, coarse line of sight direction, current operating status information of the automated guided vehicle, the identifier vector of the fixed laser transceiver, and the time features corresponding to multiple preset time intervals into the pre-constructed neural motion field line of sight prediction network, and output the line of sight prediction angle after multiple time intervals in the line of sight angle coordinate system. The pre-constructed neural motion field gaze prediction network includes an input layer, an input splicing layer, a coordinate alignment layer, a dimensionality increase layer, a motion pattern extraction layer, a temporal coupling layer, an angle regression layer, a multi-time interval output layer, and an angle bias synthesis layer arranged sequentially. The fixed laser transceiver position parameters and coarse gaze direction are used as construction constraints to configure the parameterization method of the coordinate alignment layer and the angle bias synthesis layer, so that the output of the neural motion field gaze prediction network is consistent with the gaze angle coordinate system. The current operating status information of the automated guided vehicle, the identifier vector of the fixed laser transceiver, and the temporal features corresponding to the preset multiple time intervals are used as inputs to the input layer. S4. Based on the predicted angles of the line of sight after multiple time intervals in the line of sight angle coordinate system, obtain the line of sight target trajectory used to control the movement of the gimbal. S5. Control the vehicle-mounted laser transceiver gimbal and the fixed laser transceiver gimbal according to the line-of-sight target trajectory for controlling the movement of the gimbal.

2. The method for maintaining a laser communication link for automated guided vehicles according to claim 1, characterized in that, In step S1, establishing the line-of-sight coordinate system under the automated guided vehicle (AGV) workshop includes: using the structural reference plane of the workshop as a reference, determining the origin, horizontal reference axis, and vertical reference axis of the line-of-sight coordinate system, wherein the horizontal reference axis is consistent with the positive direction of the AGV's heading angle, setting the zero position of the line-of-sight angle and the zero position of the pitch angle, and setting the upper and lower limits of the angle value range of the line-of-sight coordinate system according to the boundary of the AGV's passable area within the workshop.

3. The method for maintaining a laser communication link for automated guided vehicles according to claim 2, characterized in that, In step S1, aligning the position parameters of the fixed laser transceiver includes: setting at least three non-collinear positioning markers, determining the coordinates of the positioning markers in the installation coordinate system and the line-of-sight angle coordinate system of the fixed laser transceiver, obtaining the rotation matrix and displacement vector from the installation coordinate system to the line-of-sight angle coordinate system based on the coordinates of the same positioning marker in different coordinate systems, and transforming the coordinates of the fixed laser transceiver in the installation coordinate system to the line-of-sight angle coordinate system based on the rotation matrix and displacement vector.

4. The method for maintaining a laser communication link for automated guided vehicles according to claim 1, characterized in that, In step S3, the input layer is used to obtain the current operating status information of the automated guided vehicle, the identification vector of the fixed laser transceiver, and the time features corresponding to multiple preset time intervals; The input splicing layer is used to splice the information obtained from the input layer; The coordinate alignment layer is used to obtain an alignment vector that expresses the spatial relationship of the fixed laser transceiver in the line-of-sight coordinate system by using the fixed laser transceiver position parameters and coarse line-of-sight direction as construction constraints. The alignment vector and the splicing result obtained by the input splicing layer are used to form the alignment input. The dimensionality-upgrading layer is used to obtain multidimensional motion state features through multiple fully connected neurons. The motion pattern extraction layer is used to extract multidimensional motion pattern features through a two-level fully connected structure; the temporal coupling layer is used to couple the temporal features with the multidimensional motion pattern features element by element through multiple fully connected neurons to obtain multidimensional time-related features; the angle regression layer is used to compress the multidimensional time-related features into angle features through multiple fully connected neurons; the multi-time interval output layer is used to output the gaze prediction angles after multiple time intervals; the angle bias synthesis layer is used to generate the gaze prediction angles after multiple time intervals output by the multi-time interval output layer according to the coarse gaze direction, based on the gaze direction.

5. The method for maintaining a laser communication link for automated guided vehicles according to claim 4, characterized in that, In step S3, generating the predicted viewing angles after multiple time intervals output by the multi-time interval output layer according to the coarse viewing direction includes: generating an angle offset using the coarse viewing direction as input, adding the angle offset to the predicted viewing angles after multiple time intervals output by the multi-time interval output layer to obtain a sum. If the sum is between the lower and upper azimuth boundaries of the viewing angle coordinate system, the sum is taken as the predicted viewing angle after multiple time intervals in the viewing angle coordinate system. If the sum is less than the lower azimuth boundary of the viewing angle coordinate system, the lower azimuth boundary of the viewing angle coordinate system is taken as the predicted viewing angle after multiple time intervals in the viewing angle coordinate system. If the sum is greater than the upper azimuth boundary of the viewing angle coordinate system, the upper azimuth boundary of the viewing angle coordinate system is taken as the predicted viewing angle after multiple time intervals in the viewing angle coordinate system.

6. The method for maintaining a laser communication link for automated guided vehicles according to claim 1, characterized in that, In step S4, obtaining the line-of-sight target trajectory for controlling the gimbal movement includes: determining the current line-of-sight angle based on the coarse line-of-sight direction, and combining the current line-of-sight angle with the predicted line-of-sight angles after multiple time intervals in the line-of-sight angle coordinate system in chronological order to form the line-of-sight target trajectory for controlling the gimbal movement.

7. The method for maintaining a laser communication link for automated guided vehicles according to claim 1, characterized in that, Step S5 involves controlling the onboard laser transceiver gimbal and the fixed laser transceiver gimbal of the automated guided vehicle, including: The laser follow control command is generated based on the line-of-sight target trajectory and the laser signal intensity. When the laser signal intensity is lower than the preset threshold, the local search direction is determined based on the line-of-sight prediction angle in the line-of-sight target trajectory, and the laser follow control command is updated. According to the laser follow control command, the vehicle-mounted laser transceiver gimbal and the fixed laser transceiver gimbal of the automated guided vehicle are driven to rotate, forming a continuous laser communication link between the automated guided vehicle and the fixed laser transceiver.

8. The method for maintaining a laser communication link for automated guided vehicles according to claim 7, characterized in that, In step S5, when the laser signal intensity is not lower than the preset threshold, the laser following control command is corrected according to the error between the current line of sight angle and the target angle, and the angular velocity is limited to not exceeding the upper limit of the angular velocity, so that the gimbal follows the trajectory of the line of sight target.

Images