A method for vehicle-mounted unmanned aerial vehicle (UAV) landing based on multi-source localization fusion
By using multi-source positioning fusion technology, combined with sensors such as the Global Positioning System, Inertial Measurement Unit, and Visual Camera, the accuracy and reliability issues of autonomous landing of UAVs on dynamic vehicles have been solved, achieving high-precision dynamic landing and safety control.
Patent Information
- Application Number
- CN202611310426.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-27
- Publication Date
- 2026-11-17
AI Technical Summary
When existing drones autonomously land on dynamic vehicles, the positioning accuracy of traditional single sensors is insufficient, making it difficult to meet the requirements of high-precision dynamic docking. Furthermore, visual guidance is easily affected by changes in lighting and motion blur, and static trajectory planning cannot adapt to changes in the position of the mobile platform.
A multi-source positioning fusion method is adopted, which collects data through the Global Positioning System, Inertial Measurement Unit, Visual Camera and LiDAR, performs coordinate system transformation, distortion correction and feature extraction, constructs UAV state vector, uses adaptive Kalman filtering for time update and measurement update, plans dynamic landing trajectory, and monitors relative attitude error in real time to trigger go-around procedure.
It significantly improves the robustness and reliability of relative pose estimation between UAVs and vehicle platforms in complex environments, enhances the accuracy and safety of dynamic landing, and reduces the risk of collision.
Smart Images

Figure CN123113973A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and more specifically to a vehicle-mounted UAV landing method based on multi-source positioning fusion. Background Technology
[0002] As drone applications expand to mobile platforms, such as aerial logistics and emergency inspection, the demand for autonomous landing of drones on dynamic vehicles is becoming increasingly urgent. Traditional single-sensor systems relying on the Global Positioning System (GPS) are prone to inaccuracies in environments with weak satellite signals or high-speed movement, making it difficult to meet the requirements for high-precision dynamic docking. Multi-source positioning fusion technology, by integrating data from GPS, inertial measurement units (INS), vision, and lidar, can significantly improve the robustness and accuracy of relative positioning in complex environments, and has become a key technological direction for realizing dynamic drone landing on vehicles.
[0003] With the rapid development of drone technology and the continuous expansion of its application scenarios, the demand for autonomous take-off and landing of drones on mobile platforms is becoming increasingly prominent. In tasks such as cargo delivery, emergency inspection, and marine recovery, drones need to be able to land precisely on dynamic carriers such as moving vehicles and ships, which places higher demands on positioning accuracy and flight control. Most existing drone landing technologies are designed for fixed take-off and landing points, relying on the cooperation of the Global Positioning System (GPS) and visual markers to complete the landing task.
[0004] However, traditional methods face numerous challenges in vehicle-mounted dynamic landing scenarios. Relying solely on the Global Positioning System (GPS) suffers a sharp decline in positioning accuracy when satellite signals are blocked or multipath effects occur, making it difficult to meet landing requirements. While visual guidance offers high accuracy at close range, it is susceptible to blurring caused by changes in lighting and rapid motion. Furthermore, the high-speed movement of the vehicle causes the landing point to change in real time, making it impossible for conventional static trajectory planning methods to achieve effective convergence between the UAV and the mobile platform. Therefore, a solution capable of achieving high-precision dynamic landing in complex environments is needed. Summary of the Invention
[0005] To address the aforementioned shortcomings of existing technologies, this invention provides a vehicle-mounted UAV landing method based on multi-source positioning fusion, which can effectively solve the problems of low dynamic landing accuracy and poor reliability in complex environments in the background technology.
[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a vehicle-mounted unmanned aerial vehicle landing method based on multi-source positioning fusion, comprising: Step 1: Collect raw data and perform coordinate system transformation, distortion correction, outlier removal, and feature extraction on the data from each sensor to generate a pre-fusion data package with timestamps.
[0007] Step 2: Through time synchronization and spatial synchronization, unify the data from all sensors to the same time reference and spatial coordinate system.
[0008] Step 3: Construct the UAV state vector, based on time updates of inertial measurement unit data and measurement updates based on multi-source positioning data.
[0009] Step 4: Based on the fused relative pose information between the UAV and the vehicle platform, predict the future trajectory of the vehicle and plan the dynamic landing trajectory of the UAV.
[0010] Step 5: Generate control commands based on the planned trajectory, execute UAV landing control, and monitor relative attitude error in real time. Trigger a go-around when the error exceeds the safety threshold.
[0011] Preferably, sensor deployment, data acquisition, and preprocessing are carried out, and the specific process is as follows: Configure and deploy a variety of positioning and sensing sensors on drones and vehicle platforms.
[0012] Collect raw data output from each sensor.
[0013] The raw data is preprocessed, including error correction, coordinate system unification, and feature information extraction for data from different types of sensors.
[0014] Add a unified timestamp to each of the preprocessed data sets to generate a data packet to be merged.
[0015] Preferably, sensor spatiotemporal synchronization and spatial synchronization analysis are performed, and the specific process is as follows: Spatiotemporal synchronization processing is performed on the preprocessed multi-source sensor data.
[0016] The spatiotemporal synchronization includes time synchronization and spatial synchronization.
[0017] Time synchronization unifies the sampling times of all sensors to the same time base and aligns the time of sensor data at different frequencies.
[0018] By spatial synchronization, the relative spatial positions of each sensor are pre-calibrated, and the observation data of all sensors are uniformly converted to the same spatial coordinate system to generate spatiotemporally aligned fused data.
[0019] Preferably, the state vector is constructed and initialized, and the specific process is as follows: Construct a vector describing the motion state of the UAV, wherein the state vector includes at least the UAV's position, velocity, attitude, and sensor bias parameters.
[0020] Initialize the filter state, including setting the initial value of the state vector and the corresponding initial value of the state covariance matrix. The initial value of the state covariance matrix reflects the uncertainty of the initial estimate of each state variable.
[0021] Preferably, the time update based on inertial measurement unit data follows this process: By using high-frequency sensor data and updating the current state vector based on the UAV's kinematic model, the state of the UAV at the next moment can be predicted.
[0022] At the same time, the state covariance matrix is updated to reflect the uncertainty introduced during the prediction process.
[0023] Preferably, measurement updates are performed based on multi-source positioning data, and the specific process is as follows: Acquire multi-source observation data after spatiotemporal synchronization.
[0024] Based on the type and characteristics of each observation data, corresponding observation models are constructed.
[0025] The predicted state vector is updated using the observation model, and the state estimate is corrected by fusing different observation data.
[0026] Update the state covariance matrix to reduce the uncertainty of state estimation.
[0027] Output the fused UAV status information, including the UAV's position, speed, and attitude.
[0028] Preferably, the future trajectory of the vehicle is predicted, and the specific process is as follows: Based on the fusion of the vehicle's historical motion state and current motion state, the vehicle's motion trajectory within a preset future time domain is dynamically predicted, generating a predicted trajectory that includes the vehicle's future position and speed information.
[0029] Preferably, the dynamic landing trajectory of the UAV is planned, and the specific process is as follows: Based on the current position and speed of the UAV, and combined with the predicted trajectory of the vehicle, a three-dimensional landing trajectory of the UAV is planned so that when the UAV reaches the landing point, its horizontal position coincides with the predicted position of the vehicle, and its horizontal speed is consistent with the instantaneous speed of the vehicle, thus achieving synchronous landing with the moving vehicle platform.
[0030] Preferably, the drone landing control is executed, and the specific process is as follows: Based on the expected trajectory planned by S4, obtain the expected position and expected speed of the drone at the current moment.
[0031] Collect the current actual position and speed of the drone, and calculate the deviation between the actual state and the expected state.
[0032] Based on the deviation, attitude control commands for the UAV are generated, driving the UAV to fly along the desired trajectory and achieving precise tracking of the dynamic landing point.
[0033] Preferably, the relative pose error is monitored in real time, and the specific process is as follows: During the landing process, the relative position error and relative speed error between the drone and the vehicle platform are monitored in real time.
[0034] The relative error is compared with a preset safety threshold.
[0035] When the relative error exceeds a preset threshold and continues for a preset time, it is determined to be a landing danger state, and the landing procedure is immediately aborted and a go-around procedure is executed.
[0036] The go-around procedure includes at least: controlling the drone to stop descending, climbing to a safe altitude, reacquiring positioning information, and deciding whether to attempt to land again.
[0037] The technical solution provided by this invention has the following advantages compared with the known prior art: 1. In the data acquisition and preprocessing process, this invention deploys multiple positioning and sensing sensors, including a GPS receiver, an inertial measurement unit, a visual camera, and a lidar, and performs preprocessing operations such as coordinate system transformation, distortion correction, outlier removal, and feature extraction on the raw data. This helps to eliminate the measurement errors and data noise of the sensors themselves from the source, and generates a high-quality pre-fusion data packet with a unified timestamp, laying a reliable data foundation for subsequent high-precision fusion calculation.
[0038] 2. In the spatiotemporal synchronization process of sensors, the embodiments of the present invention achieve time synchronization by combining hardware-triggered alignment with a precise time protocol, and use an offline calibration method to obtain the relative spatial position relationship between each sensor to achieve spatial synchronization. This is beneficial to unify multi-source heterogeneous data under the same spatiotemporal reference, eliminates fusion errors caused by time misalignment and coordinate system inconsistency, and significantly improves the accuracy and consistency of multi-source information fusion.
[0039] 3. In the state estimation and fusion calculation process, the embodiments of the present invention construct a UAV state vector that includes position, velocity, attitude and sensor zero bias, and use adaptive Kalman filtering for time updates based on inertial measurement units and measurement updates based on multi-source observation data. This is beneficial to make full use of the complementary characteristics of each sensor, and can maintain high-precision positioning even when the global positioning system signal is weak or interfered with, which significantly improves the robustness and reliability of relative pose estimation in complex environments.
[0040] 4. In the dynamic trajectory planning process, this embodiment of the invention predicts the future trajectory of the vehicle based on the vehicle motion model and plans a three-dimensional landing trajectory in which the UAV and the dynamic landing point coincide and are synchronized in speed. This helps to solve the problem that traditional static planning methods cannot adapt to moving targets, and achieves the precise intersection of the UAV and the high-speed moving vehicle platform, which significantly improves the success rate and safety of dynamic landing.
[0041] 5. In the landing control and safety assurance process, the embodiments of the present invention use a closed-loop control algorithm to track the desired trajectory and monitor the relative attitude error between the UAV and the vehicle platform in real time. When the error exceeds the limit, the go-around procedure is triggered decisively, which helps to ensure that the landing process is always under control, avoids the risk of collision caused by excessive tracking deviation, and significantly enhances the safety and fault tolerance of the system in complex dynamic environments. Attached Figure Description
[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0043] Figure 1 This is a schematic diagram of the implementation steps of the present invention. Detailed Implementation
[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0045] The present invention will be further described below with reference to embodiments.
[0046] Please see Figure 1 As shown, the UAV vehicle-mounted landing method based on multi-source positioning fusion includes at least: Step 1: Collect raw data and perform coordinate system transformation, distortion correction, outlier removal, and feature extraction on the data from each sensor to generate a pre-fusion data package with timestamps.
[0047] In one specific embodiment, sensor deployment, data acquisition, and data preprocessing are performed, as follows: Multiple types of positioning and sensing sensors are distributedly installed on the drone's body, mainly including: A receiver that supports real-time dynamic differential is used to obtain the absolute geographical location of the drone in a wide area of space; a high-precision consumer-grade or industrial-grade inertial measurement unit is used to collect the drone's three-axis acceleration and three-axis angular velocity information at high frequencies (such as above 200Hz).
[0048] A downward-looking vision camera is fixedly installed on the bottom of the drone to collect image information of ground targets; and a single-line or multi-line lidar is installed on the drone's landing gear or gimbal to obtain high-precision distance information.
[0049] On the vehicle-mounted platform that serves as the landing target, a prominent cooperative identifier (such as a high-contrast ArUco code or AprilTag) is set at the center of the landing area on the roof, and auxiliary positioning beacons (such as UWB base stations or differential GNSS rover) can be optionally installed.
[0050] After acquiring the raw sensor data, preprocessing is performed immediately to eliminate measurement errors and standardize the data format.
[0051] For GNSS data, outlier removal is first performed to eliminate jump points caused by signal obstruction. Then, RTK correction information is used to convert latitude, longitude, and elevation coordinates into rectangular coordinates (X, Y, Z) in the local NED coordinate system.
[0052] For IMU data, zero-bias compensation and temperature drift correction are performed to obtain stable angular and velocity increments. For visual images, distortion correction is first performed to eliminate lens distortion, and then image recognition algorithms are used to detect cooperative markers in the image to calculate the four-degree-of-freedom relative pose (X, Y, Z, Yaw) of the UAV relative to the vehicle landing platform.
[0053] For the lidar point cloud, filtering processing (such as voxel filtering downsampling) is performed to reduce the computational load and to initially extract planar features.
[0054] Finally, a unified timestamp (usually based on GPS time or PTP precise time protocol) is added to all processed data, packaged into data units to be merged, and stored in a cache queue to await retrieval.
[0055] In the data acquisition and preprocessing process, this invention deploys multiple positioning and sensing sensors, including a GPS receiver, an inertial measurement unit, a vision camera, and a lidar, and performs preprocessing operations such as coordinate system transformation, distortion correction, outlier removal, and feature extraction on the raw data. This helps to eliminate the measurement errors and data noise of the sensors themselves from the source, and generates a high-quality pre-fusion data packet with a unified timestamp, laying a reliable data foundation for subsequent high-precision fusion calculation.
[0056] Step 2: Through time synchronization and spatial synchronization, unify the data from all sensors to the same time reference and spatial coordinate system.
[0057] In one specific embodiment, sensor spatiotemporal synchronization and spatial synchronization analysis are performed, and the specific process is as follows: Using the second pulse signal from the Global Positioning System (GPS) as the master clock source, this pulse signal is connected to the synchronization interfaces of the GPS receiver, inertial measurement unit (IMU), visual camera, and LiDAR via hardware circuitry, achieving hardware-triggered alignment of the sampling times of all sensors. For sensors that cannot directly receive the second pulse signal, a precise time protocol is used for network time synchronization, ensuring that the timestamps of all sensor data packets are traceable to the same atomic time reference. Before data fusion, the high-frequency sampling time of the IMU is used as the time reference axis, and linear interpolation or nearest-neighbor matching is performed on the data from other low-frequency sensors to generate a fused dataset with a unified time label. Through the above methods, the time registration error between multi-source sensors is controlled within one-tenth of the sampling interval.
[0058] During the system integration phase, the Zhang Zhengyou calibration method was used to jointly calibrate the visual camera and the inertial measurement unit (IMU), obtaining the rotation matrix and translation vector from the camera coordinate system to the IMU coordinate system. The hand-eye calibration method was used to calibrate the lidar and IMU, obtaining the transformation parameters from the lidar coordinate system to the IMU coordinate system. After calibration, all sensor observation data were first transformed to a body coordinate system centered on the IMU. After the UAV took off, combined with the absolute position and attitude information obtained from the GPS, a real-time transformation relationship from the body coordinate system to the NE-GD navigation coordinate system was constructed. The position, velocity, and attitude data in the body coordinate system were synchronously transformed to a unified navigation coordinate system for fusion calculation.
[0059] It should be noted that the second pulse signal is an electrical signal output by a GPS receiver, consisting of one pulse per second. Its rising edge is strictly aligned with Coordinated Universal Time (UTC), achieving nanosecond-level accuracy. Precise Time Protocol (RTP) is a protocol that achieves time synchronization of distributed devices through network data packet exchange, reaching microsecond-level synchronization accuracy. The sampling frequency of an inertial measurement unit (IMU) is typically much higher than that of a visual camera; therefore, using it as a time reference axis can preserve the continuity of motion information to the greatest extent possible.
[0060] It should also be noted that Zhang Zhengyou's calibration method is a camera calibration method based on a planar checkerboard grid, which can solve for both intrinsic and extrinsic camera parameters simultaneously. Hand-eye calibration is used to solve for the transformation matrix between two fixedly connected sensor coordinate systems. It is usually solved by constructing a system of equations by collecting observation data under multiple attitude conditions. The NE-GYN navigation coordinate system is a rectangular coordinate system with the UAV takeoff point as the origin, the north as the positive X-axis, the east as the positive Y-axis, and the geocentric direction as the positive Z-axis. It conforms to the right-hand rule. The inertial measurement unit is abbreviated as IMU.
[0061] In the spatiotemporal synchronization process of sensors, this invention achieves time synchronization by combining hardware-triggered alignment with a precise time protocol, and uses an offline calibration method to obtain the relative spatial position relationship between each sensor to achieve spatial synchronization. This is beneficial for unifying multi-source heterogeneous data under the same spatiotemporal reference, eliminating fusion errors caused by time misalignment and coordinate system inconsistency, and significantly improving the accuracy and consistency of multi-source information fusion.
[0062] Step 3: Construct the UAV state vector. Through time and measurement updates using filtering algorithms, fuse spatiotemporally synchronized GPS data, inertial measurement unit data, visual relative pose data, and LiDAR data to calculate high-precision relative pose information between the UAV and the vehicle platform.
[0063] In one specific embodiment, the state vector is constructed and initialized, and the specific process is as follows: First, the state vector of the filtering system is constructed. This vector contains all the necessary parameters describing the motion state of the UAV. The state vector is defined in the N-E navigation coordinate system, and its specific form is: ,in The position of the UAV in the navigation coordinate system is expressed in meters. This represents the three-dimensional velocity of the UAV in the navigation coordinate system, expressed in meters per second. A unit quaternion representing the attitude of a drone, dimensionless. This is expressed as the three-axis zero bias of the gyroscope, in radians per second. Represented as the three-axis zero bias of the accelerometer, in meters per square second. When the state vector is initialized, the position and velocity are set by the first valid observations of the Global Positioning System, and the attitude is calculated by the accelerometer and magnetometer or the initial alignment algorithm. The initial values of each zero bias are set to zero, and the state covariance matrix is initialized at the same time to reflect the confidence of the initial state.
[0064] It should be noted that the NE-G coordinate system is a rectangular coordinate system with the UAV's takeoff point as the origin, north as the positive X-axis, east as the positive Y-axis, and the geocentric direction as the positive Z-axis. Quaternions are a hypercomplex number representation used to describe three-dimensional rotations. They have the advantages of no singularity and suitability for interpolation, and are widely used in inertial navigation. Gyroscope zero bias refers to the non-zero constant error output by the gyroscope when it is stationary; the same applies to accelerometer zero bias. The state covariance matrix is a diagonal matrix, with diagonal elements representing the variance of the state estimation errors. The initial values are set based on the sensor noise characteristics.
[0065] It should be noted that the state covariance matrix Initialization is a crucial step in starting the filter. It is a 16×16 square matrix, where the diagonal elements represent the variance of the initial estimates of each state variable, and the off-diagonal elements represent the covariance between different state variables. At the initial moment, it is usually assumed that each state variable is independent, so it is set as a diagonal matrix. The diagonal values are set according to the accuracy of the initial alignment and the noise characteristics of the sensor. The variances of position and velocity are determined by the initial positioning accuracy of the Global Positioning System, the attitude variance is determined by the accuracy of the initial alignment algorithm, and the zero-bias variance is set according to the zero-bias stability set by the factory calibration of the inertial measurement unit.
[0066] The matrix form is as follows, showing only the non-zero elements on the diagonal: ,in This represents the initial position variance, with a typical value corresponding to the single-point positioning accuracy of the Global Positioning System. Let Variance be the initial velocity. The equivalent variance corresponding to the initial attitude quaternion is obtained by converting the attitude angle error into a quaternion error. Zero bias variance for the gyroscope To achieve zero bias variance for the accelerometer, all off-diagonal elements are initially set to zero, indicating that the initial estimates of each state are uncorrelated.
[0067] In one specific embodiment, the time update based on inertial measurement unit (IMU) data proceeds as follows: The time update utilizes high-frequency data from the IMU to predict the state vector, establishing a kinematic model of the state's evolution over time. The system state equation is established based on the inertial navigation mechanics equations, where the dynamic law governing the UAV's state evolution over time is: , , , , ,in and These are the three-dimensional position and velocity vectors, respectively. It is the rotation matrix from the body coordinate system to the navigation coordinate system, derived from the current attitude quaternion. The result of the conversion is It is the triaxial specific force vector measured by the inertial measurement unit accelerometer, with units of meters per square second, including the gravitational component. It is the zero bias vector of the accelerometer. It is the gravity vector in the navigation coordinate system, and is represented as in the northeast-northeast coordinate system. ,in Taking the local gravitational acceleration value, it is approximately 9.8 meters per second squared. To represent quaternion multiplication, This is represented as the three-axis angular velocity vector measured by the inertial measurement unit gyroscope, in radians per second. It is represented as the zero bias vector of the gyroscope. and These are the driving white noises for the gyroscope zero bias and the accelerometer zero bias, respectively, and they follow a Gaussian distribution.
[0068] Discretizing the above continuous-time equation yields the state prediction equation: ,in Indicates the use of The state at any given moment The predicted value of the state at time step. yes The filtered state estimate at time t. yes The inertial measurement unit input at time t, including and , It is the sampling interval of the inertial measurement unit, in seconds, function The specific discretization form is implemented using the fourth-order Runge-Kutta integral or the median integral method.
[0069] Simultaneously predict the state covariance matrix: ,in It is the predicted state covariance matrix. yes The state covariance matrix at time t, It is the state transition matrix, which is the Jacobian matrix obtained by taking the partial derivative of the system state equations with respect to the state vectors. Linearization is obtained at the point of linearization. It is the process noise covariance matrix, reflecting the contribution of inertial measurement unit noise and zero-bias random walk to prediction uncertainty. It is calculated based on the noise density and zero-bias stability parameters of the inertial measurement unit.
[0070] It should be noted that the body coordinate system is fixed to the UAV, with its origin located at the UAV's center of gravity. The X-axis points towards the nose, the Y-axis points to the right side of the fuselage, and the Z-axis points downwards from the fuselage, conforming to the right-hand rule. Specific force is an inertial navigation term referring to the resultant force of inertial and gravitational forces per unit mass, as measured by the accelerometer. Quaternion multiplication follows specific rules and is used for combined rotations and discretization step sizes. Matching the output frequency of the inertial measurement unit, a typical value of 0.005 seconds corresponds to a sampling rate of 200 Hz, and the Jacobian matrix... The derivation is based on the partial derivatives of the state equation with respect to the state variables, and is the core of the extended Kalman filter linearization. Process noise covariance matrix. It is a diagonal or block diagonal matrix, and its value is set according to the random walk coefficient and angular random walk coefficient in the inertial measurement unit datasheet.
[0071] In one specific embodiment, measurement updates are performed based on multi-source positioning data. The specific process is as follows: Measurement updates correct the predicted state using observations from external sensors. When observation data from different sensors arrive, updates are performed sequentially or batch fusion processing is performed. The general form of the measurement equation is: ,in yes Sensor observation vector at time 10:00 Represented as an observation function, it maps the state vector to the observation space. The observation noise is represented as a zero-mean Gaussian distribution, and its covariance matrix is... .
[0072] Specifically, it includes three measurement sources: Global Positioning System (GPS) measurement update. When GPS outputs valid positioning data, the observation vector is a three-dimensional position: , The observation function is: The observation noise covariance matrix is a three-dimensional diagonal matrix. The diagonal elements are adaptively adjusted according to the GPS solution status. When in a real-time dynamic difference fixed solution, the position noise variance is set to... When the solution is in floating point mode, the variance increases to When the location is single-point, the variance increases further.
[0073] Visual relative pose measurement update. When the UAV detects the cooperative identifier on the vehicle platform and calculates the relative pose, the observation vector is the UAV's three-dimensional position and yaw angle relative to the vehicle: ,in This refers to the relative position of the drone in the vehicle coordinate system, expressed in meters. This is the relative yaw angle of the UAV relative to the vehicle, in radians. The observation function needs to convert the UAV's position in the state vector and the vehicle's position in the navigation system into a relative relationship, involving vehicle pose information. The vehicle pose is synchronously acquired through the onboard GPS inertial measurement unit combined system. The observation noise covariance matrix... Based on the accuracy settings of the visual recognition algorithm, the position noise variance is set to... Angular noise variance set to .
[0074] LiDAR measurement updates: When the drone descends to a height of less than 5 meters from the landing platform, the LiDAR precise positioning mode is activated. The LiDAR obtains the drone's altitude and horizontal offset relative to the landing platform through point cloud matching. ,in This is represented as the horizontal offset of the drone within the plane of the landing platform. These are relative heights, all in meters. The observation noise covariance matrix is set as a diagonal matrix. The location noise variance is set to .
[0075] Calculate the Kalman gain for each measurement update: ,in Represented as the Kalman gain matrix, It is the observation function The Jacobian matrix obtained by taking the partial derivative with respect to the state vector is used in predicting the state. Calculation at point, It is the observation noise covariance matrix corresponding to the current measurement.
[0076] State update equation: ,in, Represented as the updated state estimate, It is the observation residual or information, representing the difference between the actual observation and the predicted observation.
[0077] Covariance update equation: ,in, Represented as the updated state covariance matrix, It is the identity matrix with the same dimension as the state.
[0078] Through the alternating iterations of time updates and measurement updates, the filtering algorithm continuously outputs high-precision, high-frequency UAV state estimates, including the UAV's position, velocity, and attitude in the navigation coordinate system. Combined with the vehicle pose information synchronously calculated by the onboard platform, the precise relative pose of the UAV with respect to the vehicle landing platform is finally calculated for subsequent trajectory planning and landing control.
[0079] It should be noted that the vehicle coordinate system is defined similarly to the UAV's body coordinate system, with the origin located at the vehicle's center of gravity or the center of the landing platform. The X-axis points towards the front of the vehicle, the Y-axis points to the right side of the vehicle, and the Z-axis points to the ground. The derivation of the observation function Jacobian matrix H is based on the specific measurement equations, measured by the Global Positioning System. For constant matrices, H represents the partial derivatives of coordinate transformations involved in visual and lidar measurements, and the observation noise covariance matrix is also present. This reflects the sensor's measurement accuracy; the smaller the value, the higher the confidence in the measurement. Adaptability is reflected in... Real-time dynamic differential is a differential positioning technology for GPS that dynamically adjusts fusion weights as the GPS solution status changes and as other sensor measurements become valid.
[0080] In the state estimation and fusion calculation process, this invention constructs a UAV state vector that includes position, velocity, attitude, and sensor zero bias, and uses adaptive Kalman filtering for time updates based on inertial measurement units and measurement updates based on multi-source observation data. This helps to fully utilize the complementary characteristics of each sensor and maintain high-precision positioning even when the GPS signal is weak or interfered with, significantly improving the robustness and reliability of relative pose estimation in complex environments.
[0081] Step 4: Based on the fused relative pose information between the UAV and the vehicle platform, predict the future trajectory of the vehicle and plan the dynamic landing trajectory of the UAV to achieve precise intersection with the mobile platform.
[0082] In one specific embodiment, predicting the future trajectory of the vehicle involves the following process: defining the vehicle's motion state vector as: ,in and This refers to the vehicle's horizontal position in the navigation coordinate system, expressed in meters. This is the vehicle's heading angle, defined as the angle between the vehicle's heading and the north direction of the navigation coordinate system, expressed in radians. It is the instantaneous speed of the vehicle, measured in meters per second, and directed along the heading angle. It is the yaw rate of the vehicle, that is, the rate of change of the heading angle, and the unit is radians per second. A positive value indicates a left turn.
[0083] Based on the current time The vehicle's motion state, for the future Predicting trajectories within seconds, predicting in the time domain. The predicted step size is 2 seconds, covering the time required for the drone to complete the landing maneuver. Set the value to 0.1 seconds to match the cycle of the control system.
[0084] For any prediction time ,in For positive integers, the recursive formula for predicting the state is as follows: When the absolute value of the yaw rate is greater than a set threshold, the vehicle moves in a circular arc: , , Through the above recursion, we can obtain the vehicle's future time: A series of predicted locations: This constitutes the vehicle's predicted trajectory, which will serve as a dynamic sequence of target points for the UAV to plan its landing path.
[0085] It should be noted that the constant turning rate and speed model is a commonly used motion prediction model, and the navigation coordinate system in this invention refers to the northeast-northeast coordinate system. The heading angle is defined as the angle from north to the velocity direction, either clockwise or counterclockwise; the specific definition depends on the coordinate system convention. This invention uses north-northeast as positive. The yaw rate can be directly measured by the inertial measurement unit installed on the vehicle or estimated by a fusion filtering algorithm. The prediction time domain... The selection of the time should take into account the drone's maneuverability and computational load, with 2 seconds being a typical value.
[0086] In one specific embodiment, the dynamic landing trajectory of the drone is planned, and the specific process is as follows: Determine the desired landing time. The landing time should be within the predicted time domain, and it must meet the condition that the UAV can reach the predicted point from its current position with maximum maneuverability. Calculate the UAV's current position. The horizontal distance to each point on the predicted vehicle trajectory, combined with the maximum horizontal acceleration of the drone. Estimate the time required to reach each point and select the time that is closest to the predicted time and is feasible. As the target landing time, the corresponding landing point coordinates are: ,in The height of the vehicle landing platform is considered a constant.
[0087] Horizontal trajectory planning: Design a trajectory from the current horizontal position of the UAV to the horizontal position of the landing point The path is determined, requiring the drone to arrive at a horizontal speed matching the vehicle's speed, achieving relative stillness. A polynomial programming method is used to generate a smooth trajectory. (Time...) Define the horizontal position command as the independent variable. The function is a fifth-degree polynomial function in time, satisfying the boundary conditions: , ,in The current horizontal velocity component of the UAV is represented by the fused state. By solving the polynomial coefficients, the expected horizontal position and velocity at any time between the current time and the landing time can be obtained.
[0088] Vertical height planning: A segmented strategy is adopted, in the horizontal following stage ( to The drone should maintain its current altitude or descend slowly to avoid approaching the ground prematurely. During the final descent phase (final... Within seconds, the drone descends vertically at a constant speed to the height of the vehicle platform; typical descent time... Taking 1 second, the descent speed is calculated based on the difference between the current height and the platform height: To ensure safety, the maximum descent speed is set at 1 meter per second. During the descent, the vertical position is expected to change linearly. The desired position command for the UAV in each control cycle is: The desired speed is given by the trajectory derivative. These instructions will be sent to the underlying controller to perform tracking. If the desired position is detected to deviate too much from the vehicle's predicted position at any time during the trajectory planning process (e.g., the horizontal deviation exceeds 0.5 meters), the system will replan or trigger a go-around.
[0089] It should be noted that the fifth-order polynomial trajectory ensures continuity of position, velocity, and acceleration, making it suitable for smooth UAV tracking. Maintaining altitude or employing a slow descent strategy during the horizontal following phase helps conserve energy and reduce airflow interference. A descent speed limit of 1 meter per second is a typical value that comprehensively considers safety and efficiency. A go-around refers to the process where the UAV abandons its current landing attempt, climbs back to a safe altitude, and awaits the next landing instruction.
[0090] In the dynamic trajectory planning process, this invention predicts the future trajectory of the vehicle based on the vehicle motion model and plans a three-dimensional landing trajectory in which the UAV and the dynamic landing point coincide and are synchronized in speed. This helps to solve the problem that traditional static planning methods cannot adapt to moving targets, and achieves precise intersection between the UAV and the high-speed moving vehicle platform, significantly improving the success rate and safety of dynamic landing.
[0091] Step 5: Generate control commands based on the planned trajectory, execute UAV landing control, and monitor relative attitude error in real time. Trigger a go-around when the error exceeds the safety threshold.
[0092] In one specific embodiment, the drone landing control is executed as follows: The controller takes the deviation between the drone's current actual position and speed and the desired position and speed as input, calculates the desired attitude angle, and finally converts it into motor speed commands to drive the drone to fly along the planned trajectory. The core formula of the control law is: ,in, and This represents the desired position and desired velocity output by the planning module. and This represents the current actual position and speed of the drone. and For the control gain matrix, It is the feedforward acceleration term, the calculated desired acceleration. The attitude calculation is converted into the desired roll and pitch angles, which are then fed into the attitude controller to generate the final control inputs, enabling the UAV to accurately track the dynamic landing point.
[0093] It should be noted that the control gain matrix in the first part... and The value of the feedforward acceleration term is usually determined through debugging or simulation. Its magnitude directly affects the control's response speed and stability. Excessive gain can cause system oscillations, while insufficient gain leads to tracking hysteresis. The second derivative from the trajectory planning module, when added to the feedforward, can significantly reduce dynamic tracking errors, enabling the UAV to have better phase-keeping capabilities when chasing moving targets at high speeds.
[0094] In one specific embodiment, the relative pose error is monitored in real time, and the specific process is as follows: Define the relative position error of the UAV relative to the vehicle landing platform. and relative velocity error ,in and The vehicle's current position and speed, setting safety thresholds, and horizontal position error thresholds. Meters, vertical height error threshold meters, when or If the situation persists for more than the set time, it is deemed a dangerous state and an immediate go-around is triggered. The go-around procedure includes: emergency abort of descent, vertical climb to a safe altitude, repositioning, and deciding whether to attempt landing again.
[0095] It should be noted that the setting of the safety threshold needs to take into account the sensor noise level, the drone's maneuverability, and the vehicle's speed. Typical values are as follows. and This applies to scenarios where vehicle speed does not exceed 30 km / h. The judgment of the continuous over-limit time is to prevent false triggering caused by instantaneous sensor jumps. It is usually set to 0.2 to 0.5 seconds. The repositioning process after the go-around takes a certain amount of time. Therefore, the go-around decision should be executed decisively to avoid missing the safety opportunity due to hesitation.
[0096] In the landing control and safety assurance process, this invention employs a closed-loop control algorithm to track the desired trajectory and monitors the relative attitude error between the UAV and the vehicle platform in real time. When the error exceeds the limit, a go-around procedure is decisively triggered, which helps to ensure that the landing process is always under control, avoids the risk of collision caused by excessive tracking deviation, and significantly enhances the safety and fault tolerance of the system in complex dynamic environments.
[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.
Claims
1. A UAV vehicle-mounted landing method based on multi-source positioning fusion, characterized in that, include: Step 1: Collect raw data and perform coordinate system transformation, distortion correction, outlier removal and feature extraction on the data from each sensor to generate a data package before fusion with timestamps; Step 2: Through time synchronization and spatial synchronization, unify the data from all sensors to the same time reference and spatial coordinate system; Step 3: Construct the UAV state vector, based on time updates of inertial measurement unit data and measurement updates based on multi-source positioning data; Step 4: Based on the fused relative pose information between the UAV and the vehicle platform, predict the future trajectory of the vehicle and plan the dynamic landing trajectory of the UAV. Step 5: Generate control commands based on the planned trajectory, execute UAV landing control, and monitor relative attitude error in real time. Trigger a go-around when the error exceeds the safety threshold.
2. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 1, characterized in that, The specific process of sensor deployment, data acquisition, and data preprocessing is as follows: Configure and deploy a variety of positioning and sensing sensors on drones and vehicle platforms; Collect raw data output from each sensor; The raw data is preprocessed, including error correction, coordinate system unification, and feature information extraction for data from different types of sensors. Add a unified timestamp to each of the preprocessed data sets to generate a data packet to be merged.
3. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 2, characterized in that, The specific process for performing sensor spatiotemporal synchronization and spatial synchronization analysis is as follows: Spatiotemporal synchronization processing is performed on the preprocessed multi-source sensor data; The spatiotemporal synchronization includes time synchronization and spatial synchronization; By synchronizing time, the sampling times of each sensor are unified to the same time base, and the time of sensor data at different frequencies is aligned. By spatial synchronization, the relative spatial positions of each sensor are pre-calibrated, and the observation data of all sensors are uniformly converted to the same spatial coordinate system to generate spatiotemporally aligned fused data.
4. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 3, characterized in that, The state vector is constructed and initialized, and the specific process is as follows: Construct a vector describing the motion state of the UAV, wherein the state vector includes at least the UAV's position, velocity, attitude, and sensor bias parameters; Initialize the filter state, including setting the initial value of the state vector and the corresponding initial value of the state covariance matrix. The initial value of the state covariance matrix reflects the uncertainty of the initial estimate of each state variable.
5. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 4, characterized in that, The time update based on inertial measurement unit data is as follows: Using high-frequency sensor data, the current state vector of the UAV is updated over time according to the UAV's kinematic model to predict the UAV's state at the next moment. At the same time, the state covariance matrix is updated to reflect the uncertainty introduced during the prediction process.
6. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 5, characterized in that, Measurement updates are performed based on multi-source positioning data, and the specific process is as follows: Acquire multi-source observation data after spatiotemporal synchronization; Based on the type and characteristics of each observation data, a corresponding observation model is constructed. The predicted state vector is updated by measuring the observation model described above, and the state estimate is corrected by fusing different observation data. Update the state covariance matrix to reduce the uncertainty of state estimation; Output the fused UAV status information, including the UAV's position, speed, and attitude.
7. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 6, characterized in that, The specific process for predicting the future trajectory of a vehicle is as follows: Based on the fusion of the vehicle's historical motion state and current motion state, the vehicle's motion trajectory within a preset future time domain is dynamically predicted, generating a predicted trajectory that includes the vehicle's future position and speed information.
8. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 7, characterized in that, The specific process for planning the dynamic landing trajectory of a drone is as follows: Based on the current position and speed of the UAV, and combined with the predicted trajectory of the vehicle, a three-dimensional landing trajectory of the UAV is planned so that when the UAV reaches the landing point, its horizontal position coincides with the predicted position of the vehicle, and its horizontal speed is consistent with the instantaneous speed of the vehicle, thus achieving synchronous landing with the moving vehicle platform.
9. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 8, characterized in that, The specific process for executing drone landing control is as follows: Based on the expected trajectory planned by S4, obtain the expected position and expected speed of the drone at the current moment; Collect the current actual position and speed of the drone, and calculate the deviation between the actual state and the expected state; Based on the deviation, attitude control commands for the UAV are generated, driving the UAV to fly along the desired trajectory and achieving precise tracking of the dynamic landing point.
10. The UAV vehicle-mounted landing method based on multi-source positioning fusion according to claim 9, characterized in that, Real-time monitoring of relative pose error, the specific process is as follows: During the landing process, the relative position error and relative speed error between the drone and the vehicle platform are monitored in real time. The relative error is compared with a preset safety threshold. When the relative error exceeds a preset threshold and continues for a preset time, it is determined to be a landing danger state, the landing procedure is immediately terminated and the go-around procedure is executed. The go-around procedure includes at least: controlling the drone to stop descending, climbing to a safe altitude, reacquiring positioning information, and deciding whether to attempt to land again.
