Positioning calibration method for dynamic connection between low-altitude logistics aircraft take-off and landing point and vehicle

Abstract

The invention provides a low-altitude logistics aircraft take-off and landing point and vehicle dynamic connection positioning calibration method, and aims to solve the problems of insufficient positioning precision, poor collaboration and single path optimization in dynamic connection. The method comprises the following steps: firstly, constructing a multi-source fusion positioning module, fusing multi-sensor data through hardware terminal integration and data synchronization and an adaptive filtering algorithm, and realizing high-precision positioning by combining environment adaptive signal switching; secondly, a vehicle-aircraft cooperative calibration link is established, and real-time cooperation is guaranteed through low-delay data transmission, dynamic deviation prediction calibration and communication exception handling; and finally, establishing a multi-agent collaborative path optimization model, modeling based on multi-target parameters, modifying a path in real time by an improved algorithm, and improving a connection effect. The method is suitable for high shielding scenes such as an operation area, centimeter-level positioning calibration can be realized, and the accuracy and collaboration of dynamic connection are enhanced.

Description

Technical Field

[0001] This invention relates to the field of low-altitude logistics positioning calibration, and in particular to a positioning calibration method for the dynamic connection between the take-off and landing point of low-altitude logistics aircraft and vehicles. It is applicable to scenarios involving the collaboration between trucks and low-altitude aircraft, such as air-to-ground connection in work areas and collaborative transportation of emergency supplies. Background Technology

[0002] With the low-altitude economy being incorporated into the national strategy, the collaborative application of trucks and low-altitude aircraft (such as air-to-ground docking in work areas and collaborative transportation of emergency supplies) has become a core direction for the upgrading of smart logistics. The dynamic docking accuracy between the aircraft's take-off and landing point and the vehicle's cargo hold directly determines the operational efficiency. For example, in the work area, when vehicles are dynamically moving due to road conditions, the aircraft needs to match the cargo hold position in real time to complete loading and unloading. Its positioning calibration error needs to be controlled within the centimeter level to avoid cargo damage or operational delays.

[0003] However, existing positioning calibration technologies have shortcomings: traditional solutions rely on a single GNSS or basic RTK positioning. In environments with high obstruction and strong electromagnetic interference, such as the work area, satellite signals are easily affected by multipath effects, resulting in large positioning errors. Furthermore, in dynamic scenarios such as vibration and attitude changes during vehicle movement, positioning drift is further amplified, directly causing a high connection deviation rate and insufficient dynamic positioning accuracy. Truck positioning terminals (such as traditional T-BOX) and aircraft navigation modules operate independently. When the vehicle adjusts its driving trajectory, the aircraft cannot obtain the updated and accurate position in real time, which causes the alignment deviation between the take-off and landing point and the cargo hold to accumulate over time, reducing loading and unloading efficiency. Furthermore, dynamic route optimization lacks coordination and precise guidance. The optimization target only focuses on the truck as a single entity and does not incorporate the connection needs of low-altitude aircraft into a unified model. For example, vehicle route planning only considers road traffic efficiency and does not reserve the optimal time and space window for drone connection, which causes drones to have to wait or detour frequently, indirectly reducing operational efficiency. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a positioning calibration method for the dynamic docking of low-altitude logistics aircraft take-off and landing points with vehicles. The specific steps are as follows: S1: The hardware terminal collects multi-source data and preprocesses the collected multi-source data to obtain processed multi-source data. A multi-source data fusion algorithm is designed, and the processed multi-source data is fused using the multi-source data fusion algorithm to obtain the truck fusion positioning. Integrated positioning with aircraft ; The hardware terminals include a GNSS module, a 9-axis IMU, and a lidar integrated into the truck-mounted terminal T-BOX; and a GNSS module and a 9-axis IMU integrated into the aircraft DBOX terminal. The truck's onboard terminal T-BOX integrates the following hardware: a GNSS module that collects latitude and longitude (B, L) and elevation. 9-axis IMU acquires three-axis acceleration. Triaxial angular velocity Sampling frequency 100Hz; LiDAR collects point cloud data. The unit is m, where These are three-dimensional coordinates in the vehicle coordinate system. The hardware integrated into the aircraft's DBOX terminal includes: a GNSS module for collecting latitude and longitude coordinates (…). ), elevation 9-axis IMU acquires three-axis acceleration. =[ , , Triaxial angular velocity =[ , , Sampling frequency 100Hz; Next, data preprocessing is performed: mean filtering is used to remove impulse noise from the GNSS module data of the truck and aircraft, and static initialization calibration is used to eliminate acceleration zero bias from the IMU data of the truck and aircraft. With zero bias of angular velocity ; Furthermore, the data preprocessing also includes: processing the point cloud data output by the lidar. First remove those that satisfy the condition. The conditions are set for points to avoid occlusion by the vehicle body itself; and the distance from all points to the initial plane is calculated. All of the point cloud data The points are marked as non-ground points. ; =0.1m is the preset ground point determination threshold.

[0005] Then, a multi-source data fusion algorithm is constructed: defining the truck state vector. , middle,( (This refers to the three-dimensional location, including the latitude and longitude output by the truck's GNSS module.) Eastward coordinates (X), northward coordinates (Y), and elevation (Z) after UTM projection transformation; These represent the vehicle's three-dimensional velocities in the east, north, and vertical directions, respectively, with initial values ​​obtained from GNSS Doppler velocity measurements; ) represents the vehicle's attitude angles, namely the roll angle, pitch angle, and yaw angle, which describe the vehicle's rotation angles around the X, Y, and Z axes, respectively. The initial values ​​are derived from the attitude angles after IMU static calibration. ; State transition matrix The dimension is 9×9, constructed based on the Euler angle kinematic model, reflecting the time evolution of each parameter in the state vector, and the control matrix. Compensation for gravity interference, initialization process noise covariance Covariance of observation noise ; Through the state transition equation Predicted state, combined with GNSS observation vectors Calculate the filter gain renew ;in g is the acceleration due to gravity; process noise It is the uncertainty error in the vehicle state evolution process, which comes from IMU measurement noise and vehicle dynamic disturbances, and is related to the state vector. One-to-one correspondence, each row corresponds to each noise component, following a mean of 0 and a covariance of . Gaussian distribution; The filter gain , It is the prior covariance matrix (a 9×9 symmetric matrix) at the previous time step k, and , describes the prior covariance at time k based on time k-1, the diagonal elements are the error variance of each state variable, and the off-diagonal elements are the error correlation between variables; For each time step k > 1, is the posterior covariance matrix of the filtering algorithm at time k-1, which is the output of the filtering update at the previous time step k-1, and serves as the input to the next time step k, describing the state vector at time k-1. The estimation error; finally, update the posterior covariance matrix. ; Initialize at the start of filtering, i.e., when k=1. Set as a diagonal matrix. Its elements are based on the initial error variance of the sensor.

[0006] When the truck's GNSS observation residual Then, the preprocessed non-terrestrial point cloud C' is matched with the offline constructed work area environment map using ICP to obtain the location correction amount. Update the state vector Where σ is the noise standard deviation, and σ = That is, take The square root of the diagonal element; the same process of inputting the data collected from the aircraft into S1 yields the aircraft fused positioning. The That is, the vehicle after the update The That is, the updated aircraft ; The updated As the fused positioning result, the vehicle fused positioning is obtained. The same process is performed on the S1 to obtain the aircraft's fused positioning by inputting the data collected from the aircraft. .

[0007] In addition, GNSS signal strength thresholds for trucks and aircraft are set. Mode switching is achieved through piecewise functions: ,in Signal strength collected by the GNSS module for trucks / aircraft; in Mode 1, RTK differential data is received in real time via the vehicle networking platform. Real-time differential signal with reference station (The base station is a fixed station around the work area.) For the pseudorange of the base station, Calculate the RTK differential correction amount using the base station timestamp. , where c (Speed ​​of light), through the formula Real-time correction of the original GNSS positioning to obtain corrected observations. Then re-execute the following filter update: ), to obtain the latest vehicle location results The vehicle positioning obtained from S1 is fused. Updated to the latest location results Soon The results are the latest ; The data collected by the aircraft is fed into the above process and processed in the same way to obtain the latest positioning result of the aircraft. The aircraft positioning obtained from S1 is fused. Updated to the latest location results Soon The results are the latest ; In Mode 2, the following correction is based on the "carrier phase smoothing pseudorange" model (utilizing carrier phase stability to suppress noise): ); Where k is the phase-pseudorange conversion coefficient, and , For GNSS L1 band wavelength, The initial carrier phase; Calculated using weighted PPK: ,in GNSS weights = ( The lower, The smaller the value, the lower the GNSS weight. Correct parameters for the broadcast ionospheric / tropospheric structures; That is, the optimal PPK position of the vehicle; the GNSS observation vector in the original observation vector. Replace with the optimal position of vehicle PPK ,Right now ; Then update the vehicle state vector. , For the final fusion positioning of vehicles under Mode 2; at last The vehicle positioning obtained from S1 is fused. Updated to the latest location results Soon The results are the latest ; The data collected by the aircraft is fed into the above process and processed in the same way to obtain the latest positioning result of the aircraft. The aircraft positioning obtained from S1 is fused. Updated to the latest location results Soon The results are the latest .

[0008] S2: Design a low-latency transmission link between the vehicle and the aircraft and based on... and Low-latency transmission of truck positioning data was obtained. and low-latency transmission of aircraft positioning data ,according to Calculate the predicted location of the truck Construct a dynamic calibration feedback and communication anomaly handling mechanism, based on... The calibrated aircraft positioning is obtained based on a dynamic calibration feedback mechanism. Alternatively, the local location of the aircraft during a communication interruption can be obtained based on the communication anomaly handling mechanism. ; Calculate the set of preset edge node locations of trucks and work areas. How to determine the signal strength of each node? ,filter The node and through Select the nearest node As a forwarding node; among which This refers to the transmit power of the 5G industrial module it is equipped with. The horizontal distance between the truck and the edge node. Environmental damage in the work area was measured based on actual measurements of mountain and vegetation obstruction; and Data marked as high priority is allocated 70% of peak bandwidth for transmission, while the remaining 30% of bandwidth is used to transmit low priority data such as historical trajectories; edge nodes are directly connected to the cloud via fiber optic cables; via the aforementioned forwarding nodes Truck location data is forwarded and transmitted with high priority bandwidth to obtain low-latency transmission. After processing using the same transmission mechanism as in this step, low-latency transmission of aircraft positioning data is obtained. .

[0009] Positioning timestamps output by the aircraft's GNSS module Based on this, the timestamps of all sensor data inside the aircraft are unified to its own GNSS. A baseline is established to ensure internal data time consistency; then cross-terminal time synchronization is performed, extracting the GNSS timestamps from the aircraft's internal memory. Calculate the deviation from the truck's GNSS timestamp. = | - |; like >1ms, calculate uniform correction amount = - The timestamps of all sensors on the aircraft that have been uniformly calibrated will be uniformly corrected to... = + As the latest timestamp; Next, the truck's position is predicted 1 second later using the S1-based filtering model, and the predicted truck position is calculated using the state transition equation. = ; Then based on Calculate the alignment deviation vector between the predicted positions of the aircraft and the truck. and deviation amplitude And obtain the calibrated aircraft positioning ; The basis Calculate the magnitude of the deviation between the predicted positions of the aircraft and the truck. And the calibrated aircraft positioning The specific process is as follows: calculate the magnitude of the deviation between the predicted positions of the aircraft and the truck. ; Finally, calibration coefficients are introduced. ,when When ≤2cm, =0.1+0.2 ; when When >2cm, Calculate the aircraft positioning after calibration ; Specifically, when the 5G signal strength RSSI < -85dBm and the interruption duration is... If the time exceeds 100ms, a 5G communication interruption is determined, indicating a communication anomaly; at this time, the truck's predicted position is used... 2. Aircraft positioning after calibration The aircraft's local positioning during communication interruption is obtained from the visual sensor image I output by the visual sensor. As a temporary alternative in case of communication failure The downgraded location data; the specific process is as follows: First, extract the truck's location history sequence for the past 5 seconds. The standard deviation of the angular velocity calculated by combining the data collected by the truck's IMU within 5 seconds Construct a prediction model, and the prediction formula is as follows: Where i is the number of samples corresponding to the interrupt duration, based on Calculated with a 10ms sampling interval; t represents the start time of the communication interruption, a discrete-time index in units of IMU sampling periods. and The two numbers represent the truck fusion positioning data from the one and two sampling times preceding the start time of the interruption, respectively, and are derived from historical sequences. In the model, the coefficients are dynamically adjusted; specifically... , The When communication is interrupted, the real-time three-dimensional position vector of the truck is calculated based on the truck's historical location and IMU vibration intensity prediction. The following three formulas are used in the calculation: The predicted information in the X, Y, and Z directions is obtained separately, and then combined to obtain the predicted center position coordinates, i.e. ; Next, the truck cargo compartment image I captured by the aircraft is preprocessed, and the rectangular edge corners of the cargo compartment are extracted using the Canny edge detection algorithm. Using the pixel-to-physical conversion coefficient calibrated by the sensor =0.001m, according to Convert the corner points of the rectangle to physical coordinates ; Then according to Calculate the physical coordinates of the cargo hold center ,pass Obtain visual correction amount ; Finally passed Obtain the aircraft's local location when communication is interrupted .

[0010] S3: Establish a path optimization model and algorithm, based on the vehicle... With aircraft or The actual path of the vehicle is obtained based on the optimization model and algorithm. With the optimal trajectory of the aircraft ; according to , or Construct a multi-objective optimization function J and introduce road slope constraints and minimum turning radius constraints; The specific formula for constructing the multi-objective optimization function J is as follows: ; in + + =1, the three weight values ​​are derived from optimization experiments based on actual scenarios in the work area to ensure a balance between different objectives; The distance (in meters) between adjacent nodes on the preset path is specifically the distance from the previous node k-1 to the current node k. The UTM back-projection algorithm is used to convert the previous node k-1 and the current node k into geographic latitude and longitude coordinates. , )and( ),according to The calculation yields the result, where R is the Earth's radius; when k=1, ( , )for plane coordinates ( Converted to geographic latitude and longitude coordinates, ( , Predicted location for trucks plane coordinates ( Converted to geographic latitude and longitude coordinates; Simultaneously preset truck planning routes and aircraft path planning ; (k) represents the waiting time (in seconds), specifically the time it takes for the aircraft / truck to reach the corresponding node on the pre-planned path after arriving at the current node of the pre-planned path, measured by the truck's real-time speed in the east and north directions. Through formula The synthesis rate was calculated. (k) (unit: m / s) and the cruising speed of the aircraft ( (Based on the performance parameters of the aircraft), according to the formula Seek; For path feasibility parameters, specifically, = , where m is the number of path nodes from the starting point of the preset path to the current node k; m_feasible is the number of nodes in the path from the starting point to the current node k that satisfy the constraints.

[0011] This step yields a set of multi-objective optimization functions and constraints, providing a quantitative optimization benchmark for subsequent path correction.

[0012] For trucks, based on preset routes With the multi-objective optimization function J, vibration weights are introduced into the heuristic function, which is: , This represents the standard deviation (in rad / s) of the angular velocity calculated by the truck's IMU over the last 5 seconds. The traditional Euclidean distance heuristic value, , ( , ) represents the three-dimensional coordinates of the current node n, , , () represents the three-dimensional coordinates of the target connection point, derived from the preset path. The end point; The vibration impact coefficient, calibrated through experiments in the work area, has a value range of [0.1, 0.5]. The path search iterates every 500ms, combining slope constraints and minimum turning radius constraints to select feasible path nodes, based on the actual cost from the starting point to the node. Computation node cost The node with the lowest cost is selected for expansion until the target point is reached; finally, the actual path of the truck is obtained. ; For the aircraft, based on the actual truck route The initial population is the candidate solution for the aircraft trajectory. Each solution in the population is a sequence containing 10 waypoints. Next, based on the alignment deviation magnitude of S2 Design dynamic crossover probability The mutation probability is fixed at 1. =0.1; Finally, based on the multi-objective optimization function J, non-dominated sorting and crowding degree calculation are performed to select the Pareto optimal solution. After 50 iterations, the solution that satisfies the condition is selected from the optimal solution set. The optimal trajectory for the final aircraft .

[0013] The beneficial effects of the positioning calibration method for the dynamic connection between the take-off and landing point of the low-altitude logistics aircraft and the vehicle in this invention are as follows: This invention effectively solves the problems of insufficient positioning accuracy and environmental adaptability in existing technologies. It can maintain centimeter-level positioning accuracy even in complex environments such as high obstruction and strong electromagnetic interference, and can maintain positioning continuity when the signal is interrupted. It solves the problem of disconnected cooperation between trucks and aircraft by achieving real-time and accurate cooperation between the two through low-latency transmission and dynamic calibration, avoiding the accumulation of alignment deviations. It also solves the problem of lack of cooperation in route optimization by optimizing the path through multi-entity collaborative path optimization, making the path planning more in line with the connection requirements, and improving the overall efficiency and stability of dynamic connection. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall workflow of the positioning calibration method for the dynamic connection between the take-off and landing point of the low-altitude logistics aircraft and the vehicle according to the present invention. Detailed Implementation

[0015] To provide a further understanding of the purpose, structure, features, and functions of the present invention, detailed descriptions are provided below with reference to specific embodiments.

[0016] like Figure 1 As shown, the positioning calibration method for the dynamic docking of the take-off and landing point of a low-altitude logistics aircraft with a vehicle includes the following steps: S1: The hardware terminal collects multi-source data and preprocesses the collected multi-source data to obtain processed multi-source data. A multi-source data fusion algorithm is designed, and the processed multi-source data is fused using the multi-source data fusion algorithm to obtain the truck fusion positioning. Integrated positioning with aircraft ; S11: Data is collected via hardware terminal; The hardware terminals include a GNSS module, a 9-axis IMU, and a lidar integrated into the truck-mounted terminal T-BOX; and a GNSS module, a 9-axis IMU, and a monocular vision sensor integrated into the aircraft-mounted DBOX terminal.

[0017] Furthermore, the GNSS module and 9-axis IMU integrated in the truck-mounted terminal T-BOX are the same model as those integrated in the aircraft DBOX terminal.

[0018] The GNSS module integrated into the truck's onboard terminal T-BOX collects latitude, longitude (B, L) and elevation data. Location timestamp Raw observation data and signal strength Pg, where For carrier phase, It is a pseudo-range; The truck's on-board terminal T-BOX integrates a 9-axis IMU to collect three-axis acceleration data. (Unit: m / s²) Triaxial angular velocity (Unit: rad / s), sampling frequency: 100Hz; The truck's onboard terminal T-BOX integrates a lidar system to collect point cloud data. The unit is m, where These are three-dimensional coordinates in the vehicle coordinate system. The GNSS module integrated into the aircraft's DBOX terminal collects latitude and longitude coordinates. ), elevation Location timestamp ; 9-axis IMU acquires three-axis acceleration =[ , , (Unit: m / s²) Triaxial angular velocity =[ , , (Unit: rad / s), sampling frequency: 100Hz; The monocular vision sensor integrated into the aircraft's DBOX terminal acquires images of the cargo hold area I. Furthermore, image I has a resolution of 1920×1080; Furthermore, the GNSS module uses the GNSS-M100, which supports the fusion of multiple systems including BDS / GPS / StarNet; the 9-axis IMU uses the IMU-9250, which includes a 3-axis accelerometer, a 3-axis gyroscope, and a 3-axis magnetometer; the lidar uses the LiDAR-L16 model 16-line lidar with a scanning frequency of 10Hz and a ranging accuracy of ±2cm; both the truck T-BOX and the aircraft DBOX are equipped with the 5G industrial module AG580, and all hardware outputs multi-source data through the CAN bus.

[0019] S12: Perform data preprocessing on the data collected in S11, including the following steps: S121: Perform mean filtering on latitude and longitude data; For latitude and longitude (B, L), respectively The eastward coordinates (X) of the truck, the northward coordinates (Y) of the truck, and the eastward coordinates of the aircraft were obtained through UTM projection transformation. North coordinates of the aircraft Impulse noise was removed using a sliding window mean filter (window size 5 frames) to obtain the preprocessed 3D position of the truck. and the three-dimensional position of the aircraft The parameter t is the index of the current sampling time (or the sampling time identifier of the current data frame). In sliding window mean filtering, the window size is 5 frames, and t represents the sampling time / frame number corresponding to the current set of data to be processed. (i=0,1,2,3,4) represents the historical data frame of the i-th sampling time counting backwards from the current sampling time t; the formula achieves impulse noise suppression by averaging the GNSS three-dimensional position data of the current sampling time and the previous 4 historical sampling times. S122: Eliminate the zero bias of acceleration and angular velocity from the data collected by the IMUs of the truck and the aircraft respectively, and obtain the calibrated acceleration and angular velocity respectively; Truck IMU acceleration zero bias Angular velocity zero bias This refers to the measurement deviation of the sensor in a static state, which leads to cumulative errors during dynamic positioning and requires static initialization calibration. The specific steps include: Before starting the vehicle, the vehicle was placed in a horizontal and stationary state, and the IMU sampled and output 1000 sets of raw acceleration measurements within 10 seconds. and 1000 sets of raw angular velocity measurements (Unit: m / s², t=1,2,...,1000, representing sampling times).

[0020] In the static state, the theoretical acceleration should be: (where g is the acceleration due to gravity), the theoretical angular velocity should be: Therefore, zero bias is the average deviation between the original measured value and the theoretical value within 10 seconds: , ; in, The raw acceleration (including zero bias error) is collected in real time by the IMU. For zero acceleration bias, This is the original angular velocity (including zero bias error). The angular velocity is zero bias (unit: rad / s). Theoretical acceleration; During dynamic driving, the real-time output of the IMU and After zero bias subtraction, the calibrated acceleration of the truck is obtained. and angular velocity This eliminates the influence of static deviation on dynamic measurement.

[0021] The IMU integrated into the aircraft's DBOX performs the same zero-bias calibration procedure as the truck's. The aircraft, stationary at the takeoff and landing points, collects 10 seconds of static data to calculate the acceleration zero bias. With zero bias of angular velocity To ensure consistent data quality standards between the aircraft and the truck, the raw acceleration and raw angular velocity output from the aircraft's IMU in real time were subtracted by zero bias to obtain the calibrated acceleration of the aircraft. and angular velocity .

[0022] S123: Filter non-ground points from the point cloud data of LiDAR; The lidar point cloud contains ground points and non-ground points. Non-ground points also need to be extracted for environment matching. The specific process is as follows: Point cloud data output by lidar First remove those that satisfy the condition. The conditions should be met to avoid the vehicle itself obstructing the view; Calculate the distance from all points to the initial plane. All of the point cloud data The points are marked as non-ground points. This is used for subsequent ICP matching with the work area map to correct positioning drift caused by GNSS signal obstruction. The threshold for determining ground points is preset based on the ground features of the work area.

[0023] By standardizing hardware models, GNSS data consistency and CAN bus synchronization are ensured, which solves the error accumulation caused by hardware asynchrony and provides high-precision, synchronized raw data for subsequent multi-source data fusion.

[0024] S13: Design a multi-source data fusion algorithm to obtain fused localization results; Construct an adaptive filtering model; first, define the truck's state vector. State vector It describes the truck's motion state at time k, used for real-time estimation of vehicle position, speed and attitude, and integrates multi-source data.

[0025] middle,( The three-dimensional position of the truck is given by its eastward coordinate (X), northward coordinate (Y), and elevation (Z), which is subsequently updated iteratively through filtering. These represent the vehicle's three-dimensional velocities (in m / s) in the east, north, and vertical directions, respectively. Initial values ​​were obtained from GNSS Doppler velocity measurements, and updated dynamically via IMU acceleration integration. ) represents the vehicle's attitude angles, namely the roll angle, pitch angle, and yaw angle (in rad) to describe the vehicle's rotation angles around the X-axis, Y-axis, and Z-axis, respectively, which are updated dynamically through IMU angular velocity integration.

[0026] Furthermore, the ( The specific process of updating via IMU acceleration integration during dynamic operation is as follows: First, the calibrated acceleration after IMU zero bias subtraction is called. ( for The component represents the acceleration about the X-axis, ensuring that the acceleration of the input integral has eliminated static errors. The motion velocity update formula is based on the state vector. The update of motion speed depends on the integral of the motion speed at the previous moment and the current motion speed, as shown in the formula: ;in: The eastward velocity at time k-1 (from the state vector of the filtered output at the previous time step); The IMU sampling interval (since the IMU sampling frequency f=100Hz, therefore...) = =0.01s); Similarly, , ; The ( The specific process of updating the attitude angle through IMU angular velocity integral during dynamic operation is as follows: The attitude angle update formula is based on the state vector. The formula is ;in: The roll angle at time k-1 (from the state vector of the filtered output at the previous time step); For the IMU sampling interval; similarly, , .

[0027] State transition equation This is used to describe the dynamic evolution of the vehicle's state from time k-1 to time k, where... Let k be the predicted state of the vehicle. Let K be the vehicle state vector at time k-1, and let the state transition matrix be K. With dimensions of 9×9, it is constructed based on the Euler angle kinematics model, reflecting the time evolution of each parameter in the state vector, and its structure is as follows: In the matrix, the three-dimensional position parameters ( Updates and speed parameters () Related, specific The velocity and attitude angle remain approximately constant over a short period of time; among which The IMU sampling time interval; Control Matrix With dimensions of 9×3, it describes the influence of control inputs on the state. Its core function is to project the gravitational acceleration g onto the vehicle coordinate system to compensate for gravitational interference in IMU acceleration measurements. Its structure is as follows:

[0028] in , , , Only for velocity parameters ( This will have an impact.

[0029] The It is the vehicle's roll angle, one of the core parameters describing the truck's attitude. Specifically, it refers to the angle of rotation of the vehicle around its own longitudinal axis (usually defined as the X-axis of the vehicle's forward direction), indicating the degree of tilt of the vehicle to the left or right, and the unit is rad; when the right side of the vehicle is higher than the left side, roll is a positive value; when the left side is higher than the right side, roll is a negative value; when the vehicle is level, roll = 0; The initial roll angle is derived from the IMU's static calibration: when the vehicle is stationary before startup, the IMU measures the gravity field distribution using a triaxial accelerometer and calculates the initial roll angle. During dynamic driving, roll is updated in real time via the IMU's three-axis gyroscope.

[0030] Process noise covariance It is a 9×9 diagonal matrix, and the diagonal elements are the evolution error variances of each parameter of the state vector. The initial values ​​are based on the factory parameters of the IMU. To calibrate the angular velocity The standard deviation is calculated by setting a sliding window IMU to acquire angular velocity sequences and calibrating it, reflecting the vehicle vibration intensity (the more intense the vibration, the lower the standard deviation). The larger (the larger).

[0031] Standard deviation of angular velocity based on IMU Dynamic adjustment ,Right now ,in The initial covariance matrix is ​​based on the IMU static noise parameter settings, and the noise density parameter from the IMU manufacturer's manual is converted into the error variance in discrete time. The weighting coefficients are calibrated through scenario-based experiments (testing the impact of different values ​​on filtering accuracy in a work area scenario); when vehicle vibration intensifies ( When (increase), Increase the size to improve the filter's response speed to dynamic changes; Observation noise covariance It is a 6×6 diagonal matrix, with the diagonal elements being... Error variance, initial value Based on GNSS module factory parameters; process noise It represents the uncertainty error in the vehicle state evolution process, originating from IMU measurement noise and vehicle dynamic disturbances, with a dimension of 9×1, and is related to the state vector. One-to-one correspondence, each row corresponds to each noise component, following a mean of 0 and a covariance of . The Gaussian distribution.

[0032] The observation equation is The observation equation describes the mapping relationship between the system state vector and GNSS observations, where the observation vector... The dimension is 6×1, containing the position and attitude observations output by GNSS; Observation matrix The dimension is 6×9, which is a 9-dimensional state vector. Mapping to 6-dimensional observation vector The sparse matrix has the following structure:

[0033] In the matrix, "1" corresponds to the parameter in the state vector that is directly observed by the GNSS. , "0" corresponds to an unobserved parameter. The source is the physical correspondence between the observed values ​​and the state vector; Observation noise The dimension is 6×1, and the observation vector A one-to-one correspondence describes the deviation between GNSS observations and actual values, reflecting the random uncertainty of GNSS measurements, and follows a mean of 0 and a covariance of 0. The Gaussian distribution of the signal originates from environmental interference and hardware noise in the GNSS signal.

[0034] Observation residuals This represents the deviation between the current and predicted observations, used to determine whether lidar-assisted data is needed for correction. When the GNSS module signal from the truck is blocked, it causes the observation residual to... hour( The noise standard deviation, and = That is, take (Square root of diagonal element), fused with LiDAR-assisted data: The preprocessed non-terrestrial point cloud C' is matched with the offline constructed work area environment map using ICP (Iterative Nearest Point) matching to obtain the position correction amount. ; Input ΔX as a supplementary observation into the filtering model to update the state vector. ,in For the filter gain, and , It is the prior covariance matrix (a 9×9 symmetric matrix) at the previous time step k, and , describes the prior covariance at time k based on time k-1. The diagonal elements are the error variance of each state variable, and the off-diagonal elements are the error correlation between variables.

[0035] For each time step k>1 It is the posterior covariance matrix of the filtering algorithm at time k-1 (the final error after filtering update at time k-1), which is the output of the filtering update at the previous time step k-1 and serves as the input to the next time step k, describing the state vector at time k-1. The estimation error; finally, update the posterior covariance matrix (final covariance at time k). .

[0036] Initialize at the start of filtering, i.e., when k=1. Set as a diagonal matrix. Its elements are based on the initial error variance of the sensor.

[0037] The updated As the fused positioning result, the vehicle fused positioning is obtained. The data collected from the aircraft is fed into S12 and the same process is performed to obtain the aircraft's fused positioning. .

[0038] S14: Environmental adaptive signal switching; Setting GNSS signal strength thresholds for trucks and aircraft = -130dBm (based on actual measurements in occluded scenarios within the work area, ensuring the minimum signal strength for stable RTK positioning), mode switching is achieved through a piecewise function: ,in The Mode of the signal strength output from the GNSS module of the truck / aircraft determines the subsequent signal processing path, ensuring that the positioning mode is adapted to the environment.

[0039] In Mode 1, RTK differential data is received in real time through the vehicle-to-everything (V2X) platform. Real-time differential signal with reference station (The base station is a fixed station around the work area.) For the pseudorange of the base station, Calculate the RTK differential correction amount using the base station timestamp. , where c (Speed ​​of light), through the formula Real-time correction of the original GNSS positioning to obtain corrected observations. Then re-execute the following filter update: ), to obtain the latest vehicle location results The vehicle positioning obtained from S12 is fused. Updated to the latest location results Soon The results are the latest ; The data collected by the aircraft is fed into the above process and processed in the same way to obtain the latest positioning result of the aircraft. The aircraft positioning obtained from S12 will be fused. Updated to the latest location results Soon The results are the latest ; In Mode 2, the following correction is based on the "carrier phase smoothing pseudorange" model (utilizing carrier phase stability to suppress noise): ); Where k is the phase-pseudorange conversion coefficient, and , For GNSS L1 band wavelength, The initial carrier phase; Calculated using weighted PPK: ,in GNSS weights = ( The lower, The smaller the value, the lower the GNSS weight. Correct parameters for the broadcast ionospheric / tropospheric structures; That is, the optimal PPK position of the vehicle; the GNSS observation vector in the original observation vector. Replace with the optimal position of vehicle PPK ,Right now ; Then update the vehicle state vector. , For the final fusion positioning of vehicles under Mode 2; at last The vehicle positioning obtained from S12 is fused. Updated to the latest location results Soon The results are the latest ; The data collected by the aircraft is fed into the above process and processed in the same way to obtain the latest positioning result of the aircraft. The aircraft positioning obtained from S12 will be fused. Updated to the latest location results Soon The results are the latest ; S1 eliminates static cumulative errors through IMU zero-bias calibration, dynamically adapts to truck vibration through adaptive filtering, and solves GNSS obstruction drift through lidar-assisted calibration. It addresses the deficiencies in positioning accuracy and environmental adaptability in dynamic and signal interruption scenarios. It achieves continuity in open and obstructed environments, improving the defects of positioning interruption or drift caused by satellite signal obstruction. Low-Earth orbit satellite low latency ensures PPK real-time performance and meets dynamic connection requirements.

[0040] S2: Design a low-latency transmission link between the vehicle and the aircraft and based on... and Low-latency transmission of truck positioning data was obtained. and low-latency transmission of aircraft positioning data ,according to Calculate the predicted location of the truck Construct a dynamic calibration feedback and communication anomaly handling mechanism, based on... The calibrated aircraft positioning is obtained based on a dynamic calibration feedback mechanism. Alternatively, the local location of the aircraft during a communication interruption can be obtained based on the communication anomaly handling mechanism. ; S21: Low-latency data transmission mechanism; The work area has preset edge node locations, and the set of edge node locations is also included. Furthermore, Unit: meters (m), with one edge node deployed every 500 meters; Truck fusion positioning based on S13 output Aircraft fusion positioning Communication module parameters of the hardware terminal in S11 (peak bandwidth of AG580 single module) And a preset set of edge node locations E, a low-latency transmission link is constructed through terminal-edge node distributed forwarding and TSN time-sensitive scheduling, specifically: First, perform the optimal selection of edge nodes, through... Calculate the signal strength between the truck T-BOX and each edge node. ,in This refers to the transmission power of the communication module (AG580 factory parameters). The horizontal distance between the truck and the edge node. To account for environmental degradation in the work area (based on actual measurements of mountain and vegetation obstruction), a signal strength threshold was set. =-85dBm, filtered out edge nodes, and then through Select the nearest node As a forwarding node.

[0041] Wireless signal strength decreases with distance; the closer the distance, the less path loss and the more stable the signal quality. Choosing the nearest node maximizes the received signal strength (RSSI) and reduces data packet loss. The speed of electromagnetic wave propagation is constant (approximately 3 × 10⁻⁶). 8 (m / s), distance directly determines transmission delay. In the operational area scenario, for every additional 100 meters of distance, the theoretical transmission delay increases by about 0.33 microseconds. Although the value is small, it will affect real-time performance in end-to-end transmission. Moreover, short-distance transmission is less affected by environmental interference (such as mountain obstruction and multipath effect), and can maintain a more stable connection.

[0042] Next, high-priority data transmission will be implemented: and Data marked as high priority (needs to be used for real-time calibration) is allocated 70% of the peak bandwidth. =0.7 The remaining 30% of bandwidth is used to transmit low-priority data such as historical trajectories.

[0043] A time-triggered scheduling mechanism is employed to allocate specific time periods for the transmission of high-priority data. Within each transmission cycle, according to a preset schedule, high-priority data is allowed to exclusively occupy the communication channel within a designated time window, avoiding conflicts with other data streams and thus reducing data transmission latency.

[0044] The preset schedule refers to a transmission plan pre-defined based on task priority and data type. This schedule divides the communication cycle into multiple time slots, each allocated to a specific type of data transmission. The criteria for setting the preset schedule include the urgency of the task, the importance of the data, and the periodicity of the task.

[0045] The specific time period refers to a continuous time interval specifically reserved in the time schedule for high-priority data. Its setting is based on the bandwidth and latency requirements of data transmission, combined with the actual situation of the communication link. Generally, the length of the time period should ensure that high-priority data can be transmitted within the allowable delay range.

[0046] The designated time window refers to a short time interval within a specific time period, further divided for the transmission of a single data packet or a specific data stream. The purpose of setting a time window is to achieve finer flow control and data scheduling, avoiding multiple data streams competing for transmission resources within the same time period. Its setting can be based on factors such as data packet size, transmission rate, and the system's scheduling precision.

[0047] In implementing time-triggered scheduling, the communication module uses a precise time synchronization protocol to accurately timestamp and schedule transmission tasks. Each transmission task is assigned a predetermined execution time, and the system strictly follows the time sequence to execute data transmission operations, ensuring that critical data can be transmitted on time and reducing latency fluctuations.

[0048] The predetermined execution time refers to the specific start time of each transmission task in the schedule, which is usually determined by the scheduling algorithm based on the task priority, the urgency of the data, and the current system state. The predetermined execution time must ensure that the task can start execution within the specified time window and complete the data transmission within the stipulated time.

[0049] Edge nodes are directly connected to the cloud via fiber optic cables, passing through the aforementioned forwarding nodes. Truck location data is forwarded and transmitted with high priority bandwidth to obtain low-latency transmission. Maintain with The same parameter format; after processing by the same transmission mechanism in this step, low-latency transmission of aircraft positioning data is obtained. .

[0050] This step reduces transmission distance by replacing direct transmission over the public network with distributed edge forwarding, avoids data conflicts by time-triggered scheduling, and adapts dynamic bandwidth allocation to data update needs. It solves the unexpected problem of data latency fluctuations caused by public network congestion, and provides highly timely and stable positioning data input for subsequent dynamic calibration, preventing the accumulation of alignment deviations due to high collaborative data latency.

[0051] S22: Dynamic calibration feedback; The aircraft achieves precise tracking of the truck through spatiotemporal synchronization and dynamic deviation calibration. The specific steps are as follows: To achieve cross-terminal time standardization between trucks and aircraft, two key timestamps from the aircraft side need to be accessed: one is the positioning timestamp output by the GNSS module. Used with trucks The first is alignment; the second is the timestamp td of the visual sensor, used to calibrate the time consistency of multi-source data within the aircraft.

[0052] The specific operation is as follows: First, calculate the internal time deviation of the aircraft. Since the aircraft DBOX integrates GNSS, IMU, and vision sensors, even if synchronized via CAN bus, there may still be slight deviations. The positioning timestamp output by the aircraft's GNSS module is used as the basis for this calculation. Based on this, the timestamps of all sensor data inside the aircraft are unified to its own GNSS. A baseline is established to ensure internal data time consistency; then cross-terminal time synchronization is performed, extracting the GNSS timestamps from the aircraft's internal memory. Calculate the deviation from the truck's GNSS timestamp. = | - |; like >1ms, calculate uniform correction amount = - The timestamps of all sensors on the aircraft that have been uniformly calibrated will be uniformly corrected to... = + As the latest timestamp; By aligning the times as described above, we can avoid connection errors caused by time mismatches between visual data and positioning data.

[0053] Next, position prediction and deviation calculation are performed: the truck's position is predicted 1 second later based on the S12 adaptive filtering model, and the current truck fusion positioning is then used. Through the state transition equation Calculate the predicted location of the truck = ; Next, the alignment deviation vector between the predicted positions of the aircraft and the truck is calculated. (Unit: m) and deviation amplitude (Unit: m); Finally, calibration coefficients are introduced. ,when When ≤2cm, =0.1+0.2 ; when When >2cm, (The larger the deviation, the closer the coefficient is to 1, and the larger the calibration range.) Calculate the aircraft positioning after calibration. This ensures that the aircraft's takeoff and landing points converge toward the center of the truck's cargo hold.

[0054] After calibration, the aircraft is positioned. It directly reflects the spatial coordinates that the aircraft needs to adjust, providing a real-time correction benchmark for subsequent docking.

[0055] S23: Communication anomaly handling; Furthermore, a 5G communication interruption flag is set based on 5G communication strength and interruption duration to indicate a communication anomaly; specifically, when the 5G signal strength RSSI < -85dBm and the interruption duration is... If the time exceeds 100ms, it is determined that 5G communication is interrupted and communication is abnormal.

[0056] Truck Predicted Location Based on S22 Output Aircraft calibration and positioning The communication interruption flag and the aircraft's visual sensor image I are used to ensure positioning continuity during communication interruptions. The specific steps are as follows: First, extract the truck's location history sequence for the past 5 seconds. The standard deviation of the angular velocity calculated by combining the IMU data collected from the truck over 5 seconds (Unit: rad / s) Construct a prediction model, and the prediction formula is: Where i is the number of samples corresponding to the interrupt duration, based on Calculated with a 10ms sampling interval (if interrupt duration) =0.15s, then i=ceil( (15 times); t represents the start time of the communication interruption, a discrete-time index in units of IMU sampling periods. and The two numbers represent the truck fusion positioning data from the one and two sampling times preceding the start time of the interruption, respectively, and are derived from historical sequences. middle; The predictive model coefficients are dynamically adjusted, specifically... , (The stronger the vibration, the better) The larger the coefficient, the better it adapts to dynamic changes.

[0057] Furthermore, the aforementioned The core definition is the real-time three-dimensional position vector of a truck based on its historical location and IMU vibration intensity prediction when communication is interrupted. It refers to the locally predicted position of a commercial vehicle, with the parameter format similar to the commercial vehicle fusion positioning discussed earlier. Completely identical, but the calculation only focuses on the three-dimensional position component. ), attitude angular components ( Since the primary focus during the interruption is on positional alignment, it is not included in the calculation of visual correction. Essentially, it uses a predictive model to utilize the commercial vehicle's positioning history before the interruption to predict the real-time position during the interruption, providing a baseline prediction value for visual correction and avoiding drift caused by image noise in pure visual positioning. The following three formulas are used in the calculation: The predicted information in the X, Y, and Z directions is obtained separately, and then combined to obtain the predicted center position coordinates, i.e. .

[0058] Next, the truck cargo compartment image I captured by the aircraft was preprocessed, and the rectangular edge corners of the cargo compartment were extracted using the Canny edge detection algorithm (thresholds th1=50, th2=150). (Pixel coordinates), using pixel-to-physical conversion coefficients calibrated by a visual sensor. =0.001m (1 pixel corresponds to 0.001m physical distance), according to Convert the corner points of the rectangle to physical coordinates ; Then according to Calculate the physical coordinates of the cargo hold center ,pass Obtain visual correction amount .

[0059] Finally, combining the predicted and corrected values, through Obtain the aircraft's local location when communication is interrupted This ensures that the aircraft can still track the truck's location during the interruption.

[0060] Furthermore, the aforementioned It is a temporary replacement in case of communication failure. Degraded location data, i.e., data used in subsequent steps when communication is abnormal. Temporary replacement Its core function is to maintain the system's most basic positioning and tracking capabilities when the data link is interrupted, preventing the reconnection process from completely failing.

[0061] S3: Establish a path optimization model and algorithm, based on the vehicle... With aircraft or The actual path of the vehicle is obtained based on the optimization model and algorithm. With the optimal trajectory of the aircraft ; S31: Multi-objective parameter modeling; According to the S22 calibration, the aircraft is positioned... Truck Predicted Location S23's local positioning of aircraft (In case of communication failure) and work area scenario constraint parameters Construct a multi-objective optimization function and constraints, where The upper limit of road gradient and ≤ ; The set of coordinates of the no-fly zone for aircraft and ; The minimum turning radius for a truck (a physical characteristic parameter determined by the vehicle's wheelbase, steering system design, etc., and in practical applications, can be determined based on the specific parameters of the commercial vehicle). The specific steps are as follows: First, we introduce the mathematical expressions for the scene constraints, including but not limited to the following constraints: Road slope constraints: ; No-fly zone constraints: ; Minimum turning radius constraint: ( For the truck acceleration in S12, (This refers to the steering angle).

[0062] Furthermore, the steering angle Based on heading angular velocity ( for (components), truck synthesis speed And the wheelbase L is calculated. Specifically, when a truck turns, the vehicle satisfies the basic kinematic relationships. By transforming the formula, the steering angle can be directly solved. .

[0063] Next, define the multi-objective optimization function J. ,in + + =1 ( , , The three weight values ​​were derived from optimization experiments based on actual scenarios in the work area, ensuring a balance between different objectives. The distance (in meters) between adjacent nodes on the preset path is specifically the distance from the previous node k-1 to the current node k. The UTM back-projection algorithm is used to convert the previous node k-1 and the current node k into geographic latitude and longitude coordinates. , )and( ), according to The calculation yields the result, where R is the Earth's radius; when k=1, ( , )for plane coordinates ( Converted to geographic latitude and longitude coordinates, ( , Predicted location for trucks plane coordinates ( Converted to geographic latitude and longitude coordinates; Simultaneously preset truck planning routes and aircraft path planning ; (k) represents the waiting time (in seconds), specifically the time it takes for the aircraft / truck to reach the corresponding node on the pre-planned path after arriving at the current node of the pre-planned path, measured by the truck's real-time speed in the east and north directions. Through formula The synthesis rate was calculated. (k) (unit: m / s) and the cruising speed of the aircraft ( (Based on the performance parameters of the aircraft), according to the formula Seek; For path feasibility parameters, specifically, = , where m is the number of path nodes from the starting point of the preset path to the current node k; m_feasible is the number of nodes in the path from the starting point to the current node k that satisfy the constraints.

[0064] This step yields a set of multi-objective optimization functions and constraints, providing a quantitative optimization benchmark for subsequent path correction.

[0065] S32: Real-time path correction algorithm; Using the multi-objective optimization function J and constraints of S31, a truck planning path is preset. Aircraft planning path As input, dynamic path correction is achieved through a real-time path correction algorithm. The specific steps are as follows: The following steps are performed on the truck side: (1) Detailed processing of heuristic functions and input parameters; Plan routes for trucks and integrate truck positioning with the output of S13. The format is consistent; it is a sequence containing m consecutive nodes, that is... ; The node spacing is preset to 5m / node based on the road characteristics of the work area. and horizontal distance =5m); Target connection point The End ,in The target coordinates of the heuristic function, (Horizontal connection requirements).

[0066] The complete heuristic function is as follows: , ; The three-dimensional coordinates (east, north, and vertical, in meters) of the current search node n are derived from candidate nodes generated in real time. The standard deviation (in rad / s) of the angular velocity calculated by the truck's IMU over the last 5 seconds. The vibration influence coefficient is determined through experiments in the work area, and its value ranges from [0.1, 0.5]. The formula for the total cost of a node is: ;in, These are the weighting coefficients; This represents the cumulative actual travel cost from the starting point of the path to the current node n, obtained by weighted summation of travel distance, waiting time, and path feasibility parameters between adjacent nodes. Specifically, + + ; The value of the multi-objective optimization function corresponding to the current node n; Better Set it to 0.1 to ensure that the impact of J on the cost does not exceed 10% of g(n)+h(n).

[0067] (2) Iteration of path search; Before the iteration begins, initialize two core lists: OpenList (List of Nodes to be Expanded): Stores candidate nodes that have been generated but whose constraints have not been verified. Initially, it only contains the starting point. (Preset path start point), each node records: coordinates Parent node index , , ; CloseList (Expanded Node List): Stores nodes whose constraints have been verified and whose costs have been calculated. It is initially empty to avoid duplicate searches.

[0068] Then, each 500ms iteration is divided into the following 4 sub-steps: (2.1) Select the node with the lowest cost in the OpenList; Extract from OpenList The smallest node is used as the current node. and will Move it from OpenList to CloseList and mark it as "expanded".

[0069] (2.2) Generate the candidate neighboring nodes of the current node; Based on the road width of the work area (preset 3m), Four adjacent candidate nodes are generated in the east, north, northeast, and northwest directions (to avoid expansion beyond the road). The coordinates are calculated as follows:

[0070] Horizontal step size The step size is 2m, which is less than the node interval of 5m to ensure a smooth path. Vertical step length ; (2.3) Verify the scenario constraints of adjacent nodes; For each adjacent candidate node, both the slope constraint and the minimum turning radius constraint in document S31 must be satisfied simultaneously; otherwise, it is marked as "infeasible" and discarded. Calculate the current node With neighboring nodes road gradient The formula is: ;like If the condition is met, the constraint is satisfied; otherwise, it is discarded.

[0071] (2.4) Calculate the cost of feasible nodes and update OpenList; For neighboring nodes that pass the constraint verification, calculate their g(n), h(n), and f(n): g(n) is the actual cost from the starting point to this node, which is equal to the cost from the parent node. of Add the horizontal distance between the two nodes, that is ; Calculate according to the heuristic function in step (1); Will If a node is already in the OpenList and the new f(n) is smaller, then update its cost and the index of its parent node.

[0072] Note that iteration should stop when any of the following conditions are met: 1. Target connection point Added to CloseList (i.e., the search has reached its end); 2. OpenList is empty (no feasible path, preset path needs to be reinitialized). ; 3. If the iteration time reaches 500ms and the endpoint is not found, the current OpenList is retained and the search continues in the next cycle.

[0073] (3) Actual path Output; Backtrack the target node from CloseList. By analyzing the parent node chain, we can obtain the complete path sequence: Each node attitude angle (Constraining road flatness) Calculated from the direction vectors of adjacent nodes ( ).

[0074] The following steps are performed on the aircraft side: (1) Population initialization, generating candidate solutions for the aircraft trajectory: The trajectory is not randomly generated; the aircraft plans its path. Actual route of the truck The correlation is used to generate initial candidate solutions, ensuring that the initial solutions meet the constraints. The specific operation is as follows: Read the actual route of the truck (10 nodes, 5m apart) The initial waypoint of the aircraft must correspond one-to-one with the waypoint of the truck to avoid trajectory misalignment.

[0075] For each initial candidate solution (100 in total) at the i-th waypoint Generate according to the following rules: Position components via preset paths waypoints Based on this, a random deviation of ±10cm (simulating the initial alignment error) is added, i.e.: in (Randomly generated) and must meet no-fly zone constraints. These are the three-dimensional position parameters of the no-fly zone in the UTM coordinate system; if violated, they will be regenerated. .

[0076] The current attitude of the aircraft is referenced from the attitude angular components output by S13. The random bias of ±0.01 rad is added (to ensure initial attitude stability), i.e.: ,in .

[0077] Each candidate solution is a sequence of 10 waypoints. Generate 100 such candidate solutions to form the initial population. . (2) Calculation and operation of dynamic crossover probability; Input alignment deviation magnitude of S22 Determined according to the following rules : like At this point, the alignment deviation is large, requiring accelerated evolution and forced change. 8; like 2cm: At this point, the alignment deviation is small, and stability needs to be maintained. 0.5; Two parent candidates are randomly selected from the population. Generate random numbers ,like Then perform crossover: Randomly select two intersection points; Swap the waypoint sequence between the crossover points of the two parent generations to generate two child solutions:

[0078]

[0079] After crossover, it is necessary to verify whether the child solution meets the constraints. If it does not, the child solution is discarded and a new parent solution is selected for crossover.

[0080] Next, the mutation operation is performed, with a fixed mutation probability. = 0.1, generate random numbers for each waypoint of each candidate solution (parent + child). ,like Then the following mutations will be performed: A random deviation of ±2 cm (a small perturbation to avoid excessive variation) is added to the waypoint position components. ); Add a random bias of ±0.005 rad to the waypoint attitude angle components, i.e. .

[0081] After mutation, ensure that the waypoints still meet all constraints; otherwise, mutate again.

[0082] Non-dominated sorting and crowding calculation; Based on the multi-objective optimization function defined in S31 Filter all candidate solutions (parent generation + offspring generation, a total of 200) in the population: For each candidate solution traversing its 10 waypoints (i=1,2...10), waypoints corresponding to the truck Matching, calculating the truck's travel distance for each waypoint. (i) Aircraft waiting time (i) Path feasibility parameters ; Then calculate ; Candidate solution combined objective function value 10 waypoints The average value (reflecting the overall optimization effect of the entire trajectory), i.e. ; Next, non-dominated sorting (hierarchical screening) is performed: if the multi-objective optimization function value of candidate solution A is... If A has shorter driving distance and waiting time, higher loading and unloading efficiency, and lower energy consumption, and A is better than B in at least one objective dimension, then A dominates B. The solution that is not dominated by any other solution is selected from the population to form the non-dominated layer 1 (optimal layer); then the solution that is not dominated is selected from the remaining solutions to form the non-dominated layer 2, and so on, until all solutions are layered. We prioritize retaining solutions from non-dominated layer 1 (approximately 50 in total) and discard solutions from non-dominated layer 3 and beyond (as they have poorer objective function values).

[0083] Then, the crowding degree is calculated (to maintain solution diversity): For solutions in non-dominated layer 1, the "crowding degree" of each solution in the objective function space is calculated to avoid solutions concentrating in local regions. Specifically, for each component of J, the solutions in non-dominated layer 1 are sorted in ascending order of that component value; the distance between each solution and its neighboring solutions is calculated. The greater the crowding, the sparser the distribution of the solution in the solution set. It should be retained first to avoid the homogenization of the optimal solution.

[0084] Then, select the 100 solutions with the highest crowding from the non-dominated layer 1 to form the next generation population. (Maintain a population size of 100).

[0085] (4) Iteration and optimal trajectory selection; Repeat steps (2)-(3) for a total of 50 iterations. After the 50th iteration, take the solution of the non-dominated layer 1 as the optimal solution set.

[0086] Then filter to meet the requirements. The trajectory: for each candidate trajectory in the optimal solution set Perform the following verification: Visit its 10 waypoints Calculate the waypoints for each waypoint and the waypoints corresponding to the trucks. Alignment deviation ; If all waypoints satisfy If so, then the trajectory is the feasible optimal trajectory; From the feasible optimal trajectories, the trajectory with the smallest J is selected as the final optimal trajectory for the aircraft. Each waypoint it contains That is .

[0087] The output of this step is the actual truck route. With the optimal trajectory of the aircraft Both meet the scenario constraints and accuracy requirements, providing dynamic path guidance for collaborative connection. Together, they ensure the spatiotemporal alignment of the vehicle path with the aircraft take-off and landing point.

[0088] Through all the above steps, by fusing data from GNSS, IMU, lidar, and visual sensors and employing an adaptive filtering algorithm, the system outputs centimeter-level positioning results for trucks and aircraft in real time, providing a precise location reference for dynamic rendezvous.

[0089] A low-latency data transmission link is established, and the alignment deviation between the aircraft and the truck is calculated in real time through spatiotemporal synchronization and position prediction. When the deviation exceeds 2cm, a calibration mechanism is triggered to ensure that the aircraft accurately tracks the moving truck cargo compartment.

[0090] An optimization model is constructed that comprehensively considers driving distance, waiting time, loading and unloading efficiency, and energy consumption, and incorporates scenario constraints such as slope and no-fly zones to provide quantitative targets for route planning, ensuring that the route is both efficient and meets the requirements for connection accuracy.

[0091] The present invention has been described in the above-described embodiments; however, these embodiments are merely examples for implementing the present invention. It must be noted that the disclosed embodiments do not limit the scope of the present invention. Conversely, any modifications and refinements made without departing from the spirit and scope of the present invention are within the scope of patent protection of the present invention.

Claims

1. A positioning calibration method for the dynamic docking of the take-off and landing point of a low-altitude logistics aircraft with a vehicle, characterized in that, Includes the following steps: S1: The hardware terminal collects multi-source data and preprocesses the collected multi-source data to obtain processed multi-source data. A multi-source data fusion algorithm is designed, and the processed multi-source data is fused using the multi-source data fusion algorithm to obtain the truck fusion positioning. Integrated positioning with aircraft ; The hardware terminals include a GNSS module, a 9-axis IMU, and a lidar integrated into the truck-mounted terminal T-BOX; and a GNSS module and a 9-axis IMU integrated into the aircraft DBOX terminal. The truck's onboard terminal T-BOX integrates the following hardware: a GNSS module that collects latitude and longitude (B, L) and elevation. 9-axis IMU acquires three-axis acceleration. Triaxial angular velocity Sampling frequency 100Hz; LiDAR collects point cloud data. The unit is m, where These are three-dimensional coordinates in the vehicle coordinate system. The hardware integrated into the aircraft's DBOX terminal includes: a GNSS module for collecting latitude and longitude coordinates (…). ), elevation 9-axis IMU acquires three-axis acceleration. =[ , , Triaxial angular velocity , Next, data preprocessing is performed: mean filtering is used to remove impulse noise from the GNSS module data of the truck and aircraft, and static initialization calibration is used to eliminate acceleration zero bias from the IMU data of the truck and aircraft. With zero bias of angular velocity ; Then, a multi-source data fusion algorithm is constructed: a filtering model and a state transition matrix are built using the preprocessed data. The control matrix compensates for gravity interference, and the vehicle fusion positioning is obtained through the state transition equation. Integrated positioning with aircraft ; S2: Design a low-latency transmission link between the vehicle and the aircraft and based on... and Low-latency transmission of truck positioning data was obtained. and low-latency transmission of aircraft positioning data ,according to Calculate the predicted location of the truck Construct a dynamic calibration feedback and communication anomaly handling mechanism, based on... The calibrated aircraft positioning is obtained based on a dynamic calibration feedback mechanism. Alternatively, the local location of the aircraft during a communication interruption can be obtained based on the communication anomaly handling mechanism. ; Calculate the set of preset edge node locations of trucks and work areas. How to determine the signal strength of each node? ,filter Set signal strength threshold The node and through Select the nearest node As a forwarding node; in This refers to the transmit power of the 5G industrial module it is equipped with. The horizontal distance between the truck and the edge node. The environmental damage in the work area was measured based on actual measurements of mountain and vegetation obstruction. Will and Data marked as high priority is allocated 70% of peak bandwidth for transmission, while the remaining 30% of bandwidth is used to transmit low priority data such as historical trajectories; edge nodes are directly connected to the cloud via fiber optic cables; via the aforementioned forwarding nodes Truck location data is forwarded and transmitted with high priority bandwidth to obtain low-latency transmission. After processing using the same transmission mechanism as in this step, low-latency transmission of aircraft positioning data is obtained. ; Next, the timestamps of the aircraft and the truck are corrected, and the truck's position is predicted 1 second later based on the S1 filtering model. The predicted position of the truck is then calculated using the state transition equation. = ; Then based on Calculate the alignment deviation vector between the predicted positions of the aircraft and the truck. and deviation amplitude And obtain the calibrated aircraft positioning ; Specifically, when the 5G signal strength RSSI < -85dBm and the interruption duration is... If the time exceeds 100ms, a 5G communication interruption is determined, indicating a communication anomaly; at this time, the truck's predicted position is used...

2. Aircraft positioning after calibration The aircraft's local positioning during communication interruption is obtained from the visual sensor image I output by the visual sensor. As a temporary alternative in case of communication failure Downgraded location data; S3: Establish a path optimization model and algorithm, based on the vehicle... With aircraft or The actual path of the vehicle is obtained based on the optimization model and algorithm. With the optimal trajectory of the aircraft ; according to , or Road slope constraints and minimum turning radius constraints are introduced, and a multi-objective optimization function J is constructed. Preset truck route planning and aircraft path planning ; For trucks, based on preset routes With the multi-objective optimization function J, vibration weights are introduced into the heuristic function, which is: , This represents the standard deviation (in rad / s) of the angular velocity calculated by the truck's IMU over the last 5 seconds. The traditional Euclidean distance heuristic value, , ( , ) represents the three-dimensional coordinates of the current node n, , , () represents the three-dimensional coordinates of the target connection point, derived from the preset path. The end point; The vibration impact coefficient, calibrated through experiments in the work area, has a value range of [0.1, 0.5]. The path search iterates every 500ms, combining slope constraints and minimum turning radius constraints to select feasible path nodes, based on the actual cost from the starting point to the node. Computation node cost The node with the lowest cost is selected for expansion until the target point is reached; finally, the actual path of the truck is obtained. ; For the aircraft, based on the actual truck route The initial population is the candidate solution for the aircraft trajectory. Each solution in the population is a sequence containing 10 waypoints. Next, design the dynamic crossover probability. The mutation probability is fixed at 1. =0.1; Finally, based on the multi-objective optimization function J, non-dominated sorting and crowding degree calculation are performed to select the Pareto optimal solution. After 50 iterations, the solution that satisfies the condition is selected from the optimal solution set. The optimal trajectory for the final aircraft .

2. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and the vehicle as described in claim 1, characterized in that, The data preprocessing in step S1 also includes: processing the point cloud data output by the lidar. First remove those that satisfy the condition. The conditions are set for points to avoid occlusion by the vehicle body itself; and the distance from all points to the initial plane is calculated. All of the point cloud data The points are marked as non-ground points. ; =0.1m is the preset ground point determination threshold.

3. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and the vehicle as described in claim 1, characterized in that, In step S1, the specific process of the multi-source data fusion algorithm obtaining the fused localization is as follows: Define the truck state vector , middle,( The three-dimensional position of the truck is given by its eastward coordinate (X), northward coordinate (Y), and elevation (Z), which is subsequently updated iteratively through filtering. These represent the vehicle's three-dimensional velocities (in m / s) in the east, north, and vertical directions, respectively. Initial values ​​were obtained from GNSS Doppler velocity measurements, and updated dynamically via IMU acceleration integration. ) represents the vehicle's attitude angles, namely the roll angle, pitch angle, and yaw angle (in rad) to describe the vehicle's rotation angles around the X-axis, Y-axis, and Z-axis, respectively. These angles are updated dynamically through IMU angular velocity integration. State transition matrix The dimension is 9×9, constructed based on the Euler angle kinematic model, reflecting the time evolution of each parameter in the state vector, and the control matrix. Compensate for gravitational interference; Through the state transition equation Predicted state, combined with GNSS observation vectors Calculate the filter gain renew ;in g is the acceleration due to gravity; process noise It is the uncertainty error in the vehicle state evolution process, which comes from IMU measurement noise and vehicle dynamic disturbances, and is related to the state vector. One-to-one correspondence, each row corresponds to each noise component, following a mean of 0 and a covariance of . Gaussian distribution; When the truck's GNSS observation residual Then, the preprocessed non-terrestrial point cloud C' is matched with the offline constructed work area environment map using ICP to obtain the location correction amount. Update the state vector Where σ is the noise standard deviation, and σ = That is, take The square root of the diagonal element; The data collected from the aircraft is also fed into the S1 process to obtain the aircraft's fused positioning. The That is, the vehicle after the update The That is, the updated aircraft .

4. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and the vehicle as described in claim 3, characterized in that, The filter gain , It is the prior covariance matrix at the previous time step k, and , describes the prior covariance at time k based on time k-1, the diagonal elements are the error variance of each state variable, and the off-diagonal elements are the error correlation between variables; For each time step k>1 It is the posterior covariance matrix of the filtering algorithm at time k-1, which is the output of the filtering update at the previous time step k-1 and serves as the input to the next time step k, describing the state vector at time k-1. The estimation error; finally, update the posterior covariance matrix. Initialize at the start of filtering, i.e., when k=1. Set as a diagonal matrix. Its elements are based on the initial error variance of the sensor.

5. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and the vehicle as described in claim 1, characterized in that, In step S1, GNSS signal strength thresholds for trucks and aircraft are also set. Mode switching is achieved through piecewise functions: ,in Signal strength collected by the GNSS module for trucks / aircraft; in Mode 1, RTK differential data is received in real time via the vehicle networking platform. Real-time differential signal with reference station , For the pseudorange of the base station, Calculate the RTK differential correction amount using the base station timestamp. , where c Through formula Real-time correction of the original GNSS positioning to obtain corrected observations. Then re-execute the following filter update: ), to obtain the latest vehicle location results The vehicle positioning obtained from S1 is fused. Updated to the latest location results Soon The results are the latest ; The data collected by the aircraft is fed into the above process and processed in the same way to obtain the latest positioning result of the aircraft. The aircraft positioning obtained from S1 is fused. Updated to the latest location results Soon The results are the latest ; In Mode 2, the following correction is made based on the "carrier phase smoothing pseudorange" model: ); where k is the phase-to-pseudorange conversion coefficient, and , For GNSS L1 band wavelength, The initial carrier phase; Calculated using weighted PPK: ,in GNSS weights = , Correct parameters for the broadcast ionospheric / tropospheric structures; That is, the optimal position of the vehicle's PPK; GNSS observation vectors in the original observation vector Replace with the optimal position of vehicle PPK ,Right now ; Then update the vehicle state vector. , For the final fusion positioning of vehicles under Mode 2; at last The vehicle positioning obtained from S1 is fused. Updated to the latest location results Soon The results are the latest ; The data collected by the aircraft is fed into the above process and processed in the same way to obtain the latest positioning result of the aircraft. The aircraft positioning obtained from S1 is fused. Updated to the latest location results Soon The results are the latest .

6. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and a vehicle according to claim 1, characterized in that, In step S2, the timestamps of the corrected aircraft and truck, and the calculated calibrated aircraft positioning are... The specific process is as follows: using the positioning timestamp output by the aircraft's GNSS module Based on this, the timestamps of all sensor data inside the aircraft are unified to its own GNSS. A baseline is established to ensure internal data time consistency; then cross-terminal time synchronization is performed, extracting the GNSS timestamps from the aircraft's internal memory. Calculate the deviation from the truck's GNSS timestamp. =| - |; like > 1ms, calculate uniform correction amount = - The timestamps of all sensors on the aircraft that have been uniformly calibrated will be uniformly corrected to... = + As the latest timestamp; Calculate the magnitude of the deviation between the predicted positions of the aircraft and the truck. ; Finally, calibration coefficients are introduced. ,when When ≤2cm, =0.1+0.2 ; when When >2cm, Calculate the aircraft positioning after calibration .

7. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and the vehicle as described in claim 1, characterized in that, In step S2, the local location of the aircraft when communication is interrupted is obtained. The specific process is as follows: First, extract the truck's location history sequence for the last 5 seconds. The standard deviation of the angular velocity calculated by combining the data collected by the truck's IMU within 5 seconds Construct a prediction model, and the prediction formula is as follows: Where i is the number of samples corresponding to the interrupt duration, based on Calculated with a 10ms sampling interval; t represents the start time of the communication interruption, a discrete-time index in units of IMU sampling periods. and The two numbers represent the truck fusion positioning data from the one and two sampling times preceding the start time of the interruption, respectively, and are derived from historical sequences. In the model, the coefficients are dynamically adjusted; specifically... , The When communication is interrupted, the real-time three-dimensional position vector of the truck is calculated based on the truck's historical location and IMU vibration intensity prediction. The following three formulas are used in the calculation: The predicted information in the X, Y, and Z directions is obtained separately, and then combined to obtain the predicted center position coordinates, i.e. Next, the truck cargo compartment image I captured by the aircraft is preprocessed, and the rectangular edge corners of the cargo compartment are extracted using the Canny edge detection algorithm. Using the pixel-to-physical conversion coefficient calibrated by the sensor =0.001m, according to Convert the corner points of the rectangle to physical coordinates ; Then according to Calculate the physical coordinates of the cargo hold center ,pass Obtain visual correction amount ; Finally passed Obtain the aircraft's local location when communication is interrupted .

8. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and the vehicle as described in claim 1, characterized in that, In step S3, the road slope constraint is: ; the minimum turning radius constraint is: , The turning angle is determined by the heading angular velocity. The synthesis speed of trucks And the wheelbase L is calculated; specifically, when a truck turns, the vehicle satisfies the basic kinematic relationships. By transforming the formula, the steering angle can be directly solved. .

9. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and the vehicle as described in claim 1, characterized in that, The specific formula for constructing the multi-objective optimization function J is as follows: ,in + + =1, the three weight values ​​are derived from optimization experiments based on the actual scenario of the work area, to ensure the balance between different objectives; The preset path is defined as the distance (in meters) between adjacent truck nodes, specifically the distance from the previous node k-1 to the current node k. The UTM back-projection algorithm is used to convert the previous node k-1 and the current node k into geographic latitude and longitude coordinates (lat(k-1), lon(k-1)) and (lat(k), lon(k)) respectively. The calculation yields the result, where R is the Earth's radius; (k) represents the waiting time (in seconds), specifically the time it takes for the aircraft / truck to reach the corresponding node on the pre-planned path after arriving at the current node of the pre-planned path, measured by the truck's real-time speed in the east and north directions. Through formula The synthesis rate was calculated. (k) (unit: m / s) and the cruising speed of the aircraft According to the formula Seek; For path feasibility parameters, specifically, = , where m is the number of path nodes from the starting point of the preset path to the current node k; m_feasible is the number of nodes in the path from the starting point to the current node k that satisfy the constraints.

10. The positioning calibration method for the dynamic connection between the take-off and landing point of a low-altitude logistics aircraft and a vehicle according to claim 1, characterized in that, In step S3, the key processing for generating the optimal trajectory at the aircraft end includes: initializing a trajectory population of size 100, each solution being a sequence of 10 waypoints, and the position components following a preset path. waypoints Based on the baseline, a random deviation of ±10cm is added, that is: in And must comply with no-fly zone restrictions. These are the three-dimensional position parameters of the no-fly zone in the UTM coordinate system; if violated, they will be regenerated. ; The current attitude of the aircraft is referenced by the attitude angular components output by S1. The random bias of ±0.01 rad is added, that is: ,in ; Using dynamic crossover probability Perform crossover operation to fix mutation probability A small perturbation is performed with a value of 0.1; where hour ,otherwise ; The specific process of the non-dominated sorting is as follows: for each candidate solution... traversing its 10 waypoints (i=1,2...10), waypoints corresponding to the truck Matching, calculating the truck's travel distance for each waypoint. (i) Aircraft waiting time (i) Path feasibility parameters ; Then calculate ; Candidate solution combined objective function value 10 waypoints The average value (reflecting the overall optimization effect of the entire trajectory), i.e. ; If the multi-objective optimization function value of candidate solution A is If at least one objective dimension A is superior to B, then A dominates B; The solution that is not dominated by any other solution is selected from the population to form the non-dominated layer 1; then the solution that is not dominated is selected from the remaining solutions to form the non-dominated layer 2, and so on, until all solutions are layered. Prioritize retaining solutions from non-dominated layer 1, and discard solutions from non-dominated layer 3 and beyond.

Images