An unmanned aerial vehicle mobile platform autonomous landing and automatic locking control method

Abstract

The application provides a kind of unmanned aerial vehicle mobile platform autonomous landing and automatic locking control method, it is related to unmanned aerial vehicle landing and locking control technical field, the motion state of mobile platform is filtered and estimated, the current motion state of mobile platform is output, according to the current motion state, the output predicted motion trajectory is carried out in rolling time domain prediction;Unmanned aerial vehicle identifies visual marker through airborne vision sensor, determines the real-time pose of unmanned aerial vehicle relative to mobile platform;Real-time pose is used as observation value to feedback correction filtering estimation, update predicted motion trajectory, unmanned aerial vehicle tracks updated predicted motion trajectory, jointly judges whether to meet landing condition, when meeting, executes landing unmanned aerial vehicle and triggers locking after falling to mobile platform, after confirming that clamping force meets the standard through force sensor feedback, output locking confirmation signal.The application can realize the full autonomous landing and adaptive reliable locking of unmanned aerial vehicle under the bumpy working condition of mobile platform, improve the landing success rate and locking safety.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) landing and locking control technology, and more specifically, to a method for autonomous landing and automatic locking control of a UAV mobile platform. Background Technology

[0002] Autonomous landing and locking technologies for UAVs and mobile platforms are crucial for achieving long-endurance autonomous operation. Existing solutions typically rely on GPS for waypoint navigation, switching to visual guidance mode upon approach. This involves calculating the relative attitude by recognizing cooperative markers (such as April Tags) on the platform and controlling the UAV to land on the helipad. For locking, mechanical latches or electromagnetic chucks are commonly used, with locking initiated by position triggering or manual remote control after the UAV lands on the helipad. Some solutions incorporate an inertial measurement unit (IMU) on the mobile platform to detect attitude changes and provide reference information to the UAV.

[0003] However, none of the above solutions fully consider the impact of continuous turbulence caused by uneven road surfaces or wind and waves on the landing of the mobile platform. The UAV struggles to predict the platform's future position, and tracking based solely on the current pose can easily lead to lag or even instability. In particular, visual observation involves an inherent delay of tens to hundreds of milliseconds from image acquisition to pose calculation and backhaul. Conventional filtering estimations are purely forward recursive, directly using historical observations to correct the current state without considering the platform's movement to a new position during this delay. This causes the estimation to deviate from the true value, severely impacting landing accuracy. Therefore, achieving fully autonomous landing and adaptive reliable locking of UAVs under turbulent conditions, and improving landing success rate and locking safety, has become a challenge for the industry. Summary of the Invention

[0004] This application provides a method for autonomous landing and automatic locking control of a mobile UAV platform, which can realize fully autonomous landing and adaptive reliable locking of the UAV under turbulent conditions, thereby improving the landing success rate and locking safety.

[0005] In a first aspect, this application provides a method for autonomous landing and automatic locking control of an unmanned aerial vehicle (UAV) mobile platform, the control method comprising the following steps: The motion state of the mobile platform is filtered and estimated, and the current motion state of the mobile platform is output. Based on the current motion state, a rolling time-domain prediction is performed, and the predicted motion trajectory of the mobile platform is output. The drone identifies the visual markers using its onboard visual sensors to determine its real-time pose relative to the mobile platform. The real-time pose is used as the observation value to correct the filtered estimate and update the predicted motion trajectory. The UAV tracks the updated predicted motion trajectory and determines whether the landing conditions are met based on the airborne altitude sensor and visual pose. If the conditions are met, the UAV performs the landing. After the drone lands on the mobile platform, it triggers a lock. Once the force sensor confirms that the clamping force is sufficient, a lock confirmation signal is output.

[0006] In this embodiment, the visual markers are set around the helipad on top of the mobile platform. The visual markers are AprilTag two-dimensional barcodes, and there are 6 of them, which are asymmetrically distributed.

[0007] In this embodiment, the filtering estimation incorporates observation delay compensation: historical visual observations and their timestamps are cached, the current filtered prediction is reverse-calculated to the corresponding historical time, and the residual between the prediction and the observation is determined and then injected into the current estimate.

[0008] In this embodiment, the process of performing rolling time-domain prediction based on the current motion state and outputting the predicted motion trajectory of the mobile platform specifically includes: Using the current motion state as the initial value, a trajectory sequence in the prediction time domain is recursively generated using a kinematic model. The optimal control quantity is solved with tracking error and control increment as objectives, and the predicted motion trajectory of the mobile platform is output in a rolling manner.

[0009] In this embodiment, the feedback correction using a fixed lag window for smooth replay specifically includes: caching the state estimates and visual observations of the most recent N moments; when a new observation arrives, a replay window is formed by extending L moments before and after the current moment as the center; filtering and updating are redone within the window; and the correction amount is allocated to the current moment.

[0010] In this embodiment, the rolling time-domain prediction includes execution delay compensation: measuring the total end-to-end delay from image acquisition to attitude calculation to flight control command issuance, and shifting the starting point of the predicted time-domain reference trajectory forward by this total delay.

[0011] In this embodiment, the landing conditions are: the height of the UAV relative to the mobile platform is ≤0.2m, the horizontal offset is ≤±15mm, and the heading deviation is ≤±2°. If any one of these conditions is not met, the UAV will attempt to go around.

[0012] Secondly, this application provides an autonomous landing and automatic locking control system for an unmanned aerial vehicle (UAV) mobile platform for executing an autonomous landing and automatic locking control method for an UAV mobile platform, wherein the autonomous landing and automatic locking control system for an UAV mobile platform includes: The trajectory prediction module is used to filter and estimate the motion state of the mobile platform, output the current motion state of the mobile platform, perform rolling time-domain prediction based on the current motion state, and output the predicted motion trajectory of the mobile platform. The pose calculation module is used by the UAV to identify the visual markers through the onboard visual sensor and determine the real-time pose of the UAV relative to the mobile platform. The landing decision module is used to use the real-time pose as an observation value to correct the filtered estimate, update the predicted motion trajectory, and the UAV tracks the updated predicted motion trajectory. Based on the airborne altitude sensor and visual pose, it jointly determines whether the landing conditions are met, and executes the landing if the conditions are met. The locking control module is used to trigger locking after the drone lands on the mobile platform. After confirming that the clamping force is sufficient through feedback from the force sensor, it outputs a locking confirmation signal.

[0013] Thirdly, this application provides a computer device, which includes a memory and a processor. The memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the computer device executes the above-described method for autonomous landing and automatic locking control of a UAV mobile platform.

[0014] Fourthly, this application provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the aforementioned autonomous landing and automatic locking control method for a UAV mobile platform.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The motion state of the mobile platform is filtered and estimated to output the current motion state of the mobile platform. Based on the current motion state, a rolling time-domain prediction is performed to output the predicted motion trajectory of the mobile platform. The UAV identifies the visual markers through its onboard vision sensor to determine the real-time pose of the UAV relative to the mobile platform. The real-time pose is used as an observation value to correct the filtered estimate and update the predicted motion trajectory. The UAV tracks the updated predicted motion trajectory and determines whether the landing conditions are met based on the onboard altitude sensor and the visual pose. If the conditions are met, the UAV performs a landing. After the UAV lands on the mobile platform, a locking mechanism is triggered. After the force sensor confirms that the clamping force is sufficient, a locking confirmation signal is output.

[0016] Therefore, this application demonstrates the following: First, by filtering and estimating the motion state of the mobile platform and predicting its future trajectory, the UAV can anticipate the mobile platform's motion trend in the future, providing a motion prediction basis for subsequent landing. Second, by using an onboard visual sensor to identify visual markers on the mobile platform and calculate the real-time relative pose, the positioning continuity under partial occlusion conditions is ensured, providing a stable and reliable relative navigation reference for landing. Third, the visual pose is used as an observation value to correct the filtered estimation and update the predicted trajectory, enabling the trajectory prediction of the mobile platform to integrate visual absolute positioning information and inertial relative perception information, avoiding erroneous landings caused by single-sensor misjudgment or environmental interference, and improving the safety and reliability of landing triggering. Finally, after the UAV lands on the mobile platform, the clamping force is fed back in real time by a force sensor, ensuring that the locking mechanism can reliably fix the UAV even in turbulent environments, improving the adaptability and long-term reliability of the locking mechanism.

[0017] In summary, the technical solution adopted in this application can realize fully autonomous landing and adaptive reliable locking of UAVs under turbulent conditions on mobile platforms, thereby improving the landing success rate and locking safety. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an exemplary flowchart of an autonomous landing and automatic locking control method for a mobile unmanned aerial vehicle platform provided in this application; Figure 2 This is a schematic diagram of the locking mechanism structure of the autonomous landing and automatic locking control system for a UAV mobile platform provided in this application; it includes a fixed base 100, a left gripper 101, a right gripper 102, a V-shaped clamping port 103, a left rotating shaft 104, a right rotating shaft 105, a left electromagnetic coil 106, a right electromagnetic coil 107, a left return spring 108, a right return spring 109, a left force sensor 110, a right force sensor 111, a displacement sensor 112, an elastic buffer pad 113, a guide cone surface 114, and a UAV landing gear main shaft 115; Figure 3 This is a modular structure diagram of an autonomous landing and automatic locking control system for an unmanned aerial vehicle (UAV) mobile platform provided in this application; it includes a trajectory prediction module 201, a pose calculation module 202, a landing decision module 203, and a locking control module 204. Figure 4This is a schematic diagram of the structure of a computer device for implementing an autonomous landing and automatic locking control method for a mobile unmanned aerial vehicle platform according to the present application; wherein, it includes a processor 501, a memory 502, an instruction set 503, a program 504, and a communication unit 505. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] This application provides a method for autonomous landing and automatic locking control of a UAV mobile platform. The core of this method is to filter and estimate the motion state of the mobile platform, outputting its current motion state. Based on this current motion state, a rolling time-domain prediction is performed to output the predicted motion trajectory of the mobile platform. The UAV identifies visual markers using an onboard vision sensor to determine its real-time pose relative to the mobile platform. This real-time pose is used as an observation value to correct the filtered estimate and update the predicted motion trajectory. The UAV tracks the updated predicted motion trajectory and, based on a combination of an onboard altitude sensor and visual pose, determines whether landing conditions are met. If met, landing is executed. After landing on the mobile platform, locking is triggered. A locking confirmation signal is output after the force sensor confirms that the clamping force meets the standard. This method enables fully autonomous landing and adaptive reliable locking of the UAV under turbulent conditions, improving landing success rate and locking safety.

[0022] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of an autonomous landing and automatic locking control method for a drone mobile platform according to this embodiment of the present application. The control method includes the following steps: In step S1, the motion state of the mobile platform is filtered and estimated, and the current motion state of the mobile platform is output. Based on the current motion state, a rolling time-domain prediction is performed, and the predicted motion trajectory of the mobile platform is output.

[0023] In practical implementation, the motion state of the mobile platform is filtered and estimated to output its current motion state. In this embodiment, the mobile platform is an unmanned vehicle or unmanned vessel. To fully describe the motion of the mobile platform in space, its motion state is defined to include twelve variables: position coordinates along the three axes of the global coordinate system, three attitude angles (roll, pitch, and yaw), linear velocity components along the three axes, and angular velocity components around the three axes. These twelve variables together constitute the motion state vector of the mobile platform at that moment. Based on the kinematic laws of the mobile platform, a state transition equation is established to represent the state change from the current moment to the next moment. The state transition equation takes the current state variables as input and combines the control commands issued by the mobile platform itself, including speed and steering commands, to derive the predicted state for the next moment through kinematic differential equations. Since the actual motion of the mobile platform is affected by various uncertainties such as air resistance, ground friction or water resistance, and wind and wave disturbances, a process noise term is introduced into the state transition equation to characterize the uncertainties caused by the unmodeled dynamics and external disturbances. The process noise is statistically assumed to be zero-mean Gaussian white noise, and its intensity is described by the process noise covariance matrix. The mobile platform is equipped with an odometer and an inertial measurement unit (IMU). The odometer provides information on the platform's forward speed and steering angle, while the IMU provides acceleration and angular velocity information along three axes. The observation model is a functional expression describing the mathematical relationship between the sensor outputs and the platform's current state. The observation model maps the current state vector to the space of physical quantities observable by the sensors. The sensors themselves have measurement errors, which are represented by the observation noise term, also statistically assumed to be zero-mean Gaussian white noise, and its intensity is described by the observation noise covariance matrix.

[0024] In addition, during the system power-on startup phase, filter initialization is performed. The initial position value is provided by the GPS receiver mounted on the mobile platform after the first successful positioning. In the initial attitude values, the roll and pitch angles are calculated by the inertial measurement unit (IMU) while the platform is stationary by detecting the direction of gravitational acceleration, and the yaw angle is provided by the dual-antenna GPS. The initial linear and angular velocities for each axis are set to zero. The initial covariance matrix is ​​set as a diagonal matrix, with each diagonal element having its initial variance value set according to the nominal accuracy of each sensor. The process noise covariance matrix and the observation noise covariance matrix are pre-calibrated according to the sensor model and platform dynamic characteristics and stored in the system parameter table. At the beginning of each filtering cycle, the time update step is performed, starting from the corrected optimal state estimate of the previous moment. That is, according to the unscented transformation principle, a set of sigma sampling points is constructed around the state estimate of the previous moment, with the total number of sampling points equal to twice the number of state variables plus one. Each sampling point is distributed around the estimated value according to a symmetrical sampling strategy, and its dispersion is controlled by a scaling factor, which is preset to an appropriate value based on the system's nonlinear strength. Substituting each sampling point into the state transition equation yields a new set of sampling points after one sampling period. The weighted summation of all these new sampling points yields the mean of the one-step predicted state at the current moment. This mean is the best prediction of the current motion state of the mobile platform before the introduction of sensor measurements at the current moment. A weighted outer product is calculated on the deviations of each new sampling point from the predicted mean, and this is superimposed with the process noise covariance matrix from the previous moment to obtain the one-step prediction covariance matrix. The covariance matrix quantifies the uncertainty of the predicted state; a larger value indicates lower prediction reliability.

[0025] In this embodiment, the filtering estimation incorporates observation delay compensation: historical visual observations and their timestamps are cached, the current filtered prediction is reverse-calculated to the corresponding historical time, and the residual between the prediction and the observation is determined and then injected into the current estimate.

[0026] In the specific implementation, a circular buffer is allocated in memory, storing the most recent frames of visual observations and their corresponding timestamps in a first-in, first-out (FIFO) queue. Each buffered visual observation contains the UAV's relative position coordinates in three directions and its three relative attitude angles relative to the mobile platform, i.e., complete six-DOF relative pose information, which is the UAV's real-time pose relative to the mobile platform. The timestamp for each frame is based on the onboard system time of the actual exposure moment of the UAV's onboard visual sensor. This timestamp is triggered by a hardware synchronization signal during image acquisition and transmitted back along with the observation to ensure the accuracy of the time reference. The buffer depth is preset to cover more than twice the maximum end-to-end latency, typically set to store all visual observation data within the last five to ten seconds. When the filter receives a new frame of visual observation transmitted back from the UAV, it extracts the timestamp information carried by the visual observation, compares this timestamp with the system time of the current filtering cycle, and calculates the time difference between the two. If the time difference is less than the preset delay threshold (usually set to ten milliseconds), the delay of the observation is considered negligible, and the observation is directly sent to the regular observation update step for processing. If the time difference is greater than or equal to the preset delay threshold, the observation is determined to have significant delay, triggering the following compensation process. Since wireless data transmission, visual pose calculation, and data backhaul links between the UAV and the mobile platform all introduce delay, the actual end-to-end delay is typically between tens and hundreds of milliseconds. Therefore, the aforementioned delay threshold can be dynamically adjusted based on the actual communication link quality. For visual observations determined to have significant delay, the historical time corresponding to the observation is read from the buffer. Using the one-step predicted state obtained after the current time update as a benchmark, the predicted state is recursively derived to the aforementioned historical time according to the kinematic state transition equation of the mobile platform. During the recursive process, the inverse kinematic model corresponding to the forward recursion is used, i.e., given the current state and control input, the state at the historical time is solved in reverse. Since the mobile platform's control inputs are recorded at every moment, the backpropagation can proceed by recalling the stored historical control input data frame by frame in chronological order, starting from the current moment and proceeding cycle by cycle until the historical moment corresponding to the visual observation is reached. After the backpropagation is completed, the predicted state value of the mobile platform at that historical moment is obtained. Because this predicted state value has undergone backpropagation, its timeline is perfectly aligned with the historical visual observation values ​​stored in the cache, and both correspond to the same historical moment. The predicted state at this historical moment is mapped to the observation space through the observation model to obtain the predicted observation value at that historical moment. The predicted observation value is subtracted from the actual visual observation value stored in the cache, and the residual vector between the two is calculated.The residual vector reflects the deviation between the filter's prediction based on the kinematic model and the actual visual measurement at that historical moment. This deviation does not contain any temporal mismatch components, thus possessing accurate temporal and spatial alignment.

[0027] Furthermore, in practical implementation, after obtaining the residual vector at a historical moment, it is necessary to transform this residual into an effective correction for the current state estimate. The Kalman gain at that historical moment is calculated based on the covariance matrix corresponding to that moment. The residual at that historical moment is multiplied by the Kalman gain to obtain the state correction for that historical moment. Starting from that historical moment, the state correction is recursively applied to the current moment according to the forward kinematic state transition equation. During the forward recursion, the state correction propagates with the system dynamics, and its influence gradually spreads to all relevant components of the state vector. After the recursion is completed, the equivalent value of the state correction at the current moment is obtained. This equivalent value is then superimposed onto the one-step predicted state at the current moment, thus achieving effective correction of the current state estimate by historical visual observations. Simultaneously with the forward injection of the residual, the covariance matrix also needs to be adjusted accordingly. That is, after the historical correction is injected, the covariance matrix at that moment is reduced according to the standard Kalman update formula to reflect the reduced uncertainty brought about by the observation information. The updated covariance matrix is ​​also recursively derived to the current time step according to the forward kinematics model, with the process noise covariance of each time step superimposed during the recursion process to ensure that the value of the covariance matrix at the current time step accurately reflects the comprehensive uncertainty of the corrected state estimate. After completing the entire process of the above reverse derivation, residual calculation and forward injection, an equivalent current time step observation value and the corresponding corrected state estimate are obtained after time alignment. This equivalent observation value is then sent to the observation update step and processed according to the conventional correction step method. This implementation also includes an online measurement and update mechanism for time delay parameters. In each filtering cycle, the image acquisition time of the UAV airborne vision sensor, the pose calculation completion time, and the time when the observation value arrives at the filter are recorded. The end-to-end actual time delay value is calculated through the above three time points. The actual time delay values ​​of multiple cycles are processed by moving average to obtain the average time delay estimate under the current operating conditions. This average time delay estimate is used to dynamically adjust the time span of the above reverse recursion so that the compensation accuracy can remain optimal even when the time delay changes due to changes in communication load or fluctuations in computing resources. When a sudden change in latency is detected, such as a sudden increase in latency caused by a switch in the communication link between the drone and the mobile platform, some old data in the current cache should be cleared immediately and the cache depth should be adjusted to quickly adapt to the new latency conditions.

[0028] In addition, in the specific implementation, after the time update step is completed, the odometry and inertial measurement unit output data of the mobile platform at the current moment are collected in real time. The odometry outputs the platform's instantaneous forward speed and turning angle at the current moment, and the inertial measurement unit outputs the acceleration values ​​and angular velocity values ​​of three axes. After the above sensor data is timestamped and validated for data validity, it is used as the input for observation update and sent to the filtering and correction step. The mean of the one-step predicted state is substituted into the observation model to calculate the predicted observation value that the sensor should measure in the current predicted state. A set of sigma sampling points is reconstructed for the one-step predicted state, and these sampling points are mapped to the observation space through the observation model to obtain a set of predicted observation sampling points. The weighted sum of the set of predicted observation sampling points is performed to obtain the mean of the predicted observation, and then the covariance matrix of the predicted observation and the cross-covariance matrix between the predicted state and the predicted observation are calculated. Based on the above cross-covariance matrix and the covariance matrix of the predicted observation, the Kalman gain is calculated. Each element of the Kalman gain reflects the weight with which each component of the observation residual should be used to correct each component of the predicted state. Subtracting the predicted observation from the actual sensor measurement yields the observation residual vector. Multiplying this residual vector by the Kalman gain matrix provides the state correction. This correction is then added to the mean of the predicted state over one step to obtain the posterior state estimate for the current moment, which is the optimal estimated state output after filtering and correction. The covariance matrix is ​​updated according to the standard formula to reduce its value, reflecting the improved accuracy of the state estimation after incorporating observation information. After the aforementioned time and observation updates, the filter outputs the posterior state estimate and its corresponding covariance matrix as the final result. The output includes the optimal estimates of twelve state components: the position coordinates in three directions, the attitude angles in three directions, the linear velocities in three directions, and the angular velocities in three directions, all within the global coordinate system. The diagonal elements of the covariance matrix correspond to the estimation variance of each state component, quantifying the estimation confidence of each component.

[0029] It should be noted that the motion environment of the mobile platform may change during actual operation, such as the unmanned vehicle moving from a smooth road to a bumpy road, or the unmanned boat moving from calm water to a turbulent area. To maintain the filter's performance in changing environments, this implementation embeds an adaptive adjustment stage in each filtering cycle. The adaptive adjustment stage uses the observation residual sequence from the most recent few moments as a statistical sample to calculate the actual covariance of the residuals. This actual covariance is compared with the theoretical covariance. If the actual value is significantly greater than the theoretical value, it indicates that the currently used process noise covariance is too small and insufficient to reflect the actual disturbance amplitude, so the process noise covariance matrix is ​​increased proportionally. If the actual value is significantly less than the theoretical value, the process noise covariance matrix is ​​appropriately reduced to avoid the filter response being too slow.

[0030] In this embodiment, rolling time-domain prediction is performed based on the current motion state to output the predicted motion trajectory of the mobile platform. This can be achieved through the following steps: Using the current motion state as the initial value, a trajectory sequence in the prediction time domain is recursively generated using a kinematic model. The optimal control quantity is solved with tracking error and control increment as objectives, and the predicted motion trajectory of the mobile platform is output in a rolling manner.

[0031] In practical implementation, during each control cycle, the optimal state estimate and its covariance matrix of the mobile platform at the current moment are received. This embodiment includes an execution delay compensation mechanism in the rolling time-domain prediction, that is, measuring the end-to-end total delay from image acquisition to attitude calculation to flight control command issuance. The total delay is measured as follows: a first timestamp is recorded by hardware at the moment of image acquisition by the UAV's onboard visual sensor, a second timestamp is recorded when the visual attitude calculation is completed, and a third timestamp is recorded when the flight control command is actually issued by the mobile platform or UAV. The difference between the third timestamp and the first timestamp is taken as the end-to-end total delay. This measurement is performed once in each control cycle, and the most recent measurement values ​​are filtered by moving average to obtain a smoothed delay estimate. The end-to-end total delay obtained by the above measurement is used as the forward shift of the starting point of the predicted time-domain reference trajectory. That is, conventional rolling time-domain prediction uses the current moment as the starting moment of the prediction time domain, while this embodiment shifts the starting moment of the prediction time domain forward by an end-to-end total delay, so that the predicted reference trajectory is issued earlier on the time axis. Because there is an inherent delay between the completion of the prediction calculation and the actual application of the control command to the mobile platform's actuators, shifting the prediction starting point forward ensures that when the command reaches the actuators after the time delay, the actual motion state of the mobile platform corresponds precisely to the expected state of the predicted trajectory at that moment. This eliminates the impact of execution delay on tracking accuracy, enabling the UAV to obtain more accurate future position predictions when tracking the mobile platform. Using the estimated current motion state corresponding to the compensated prediction time domain starting point as the initial value, and based on the kinematic model of the mobile platform, a trajectory sequence is recursively generated moment-by-moment within the prediction time domain. The length of the prediction time domain is preset to a fixed value, typically twenty to thirty control cycles, with the specific value determined comprehensively based on the dynamic response speed of the mobile platform and the UAV's tracking response speed. At each prediction moment, the system calculates the state value for the next moment according to the kinematic model based on the current state and selectable control input combinations, thus forming a complete trajectory sequence starting from the initial state and covering the entire prediction time domain. Each point in this trajectory sequence contains the predicted position, predicted attitude, predicted velocity, and predicted angular velocity of the mobile platform at the corresponding moment.

[0032] Furthermore, in the specific implementation, during the generation of the predicted trajectory sequence, it is necessary to determine the optimal control input to be applied at each prediction time and construct an optimization objective function that includes tracking error and control increment. The tracking error term is the weighted sum of squared deviations between the state at each time step on the predicted trajectory and the state at the corresponding time step on the desired reference trajectory. This ensures that the predicted trajectory can effectively track the desired path of the mobile platform, enabling the UAV to accurately predict the future position of the mobile platform. The control increment term is the weighted sum of squared changes in control input between adjacent prediction times. This suppresses drastic fluctuations in control input and ensures the smoothness of control output. The above two terms are weighted and combined using preset weighting coefficients to form a complete quadratic objective function. The values ​​of the weighting coefficients are determined based on the actual system debugging results. The tracking error weight is usually set higher than the control increment weight to prioritize the accuracy of trajectory tracking. Using the control input at each time step in the prediction time domain as the decision variable and minimizing the above optimization objective function as the optimization direction, the optimal control sequence is solved under the premise of satisfying the kinematic constraints of the mobile platform itself. Kinematic constraints include maximum speed limits, maximum acceleration limits, maximum steering angle limits, and maximum steering angular velocity limits, which are determined by the physical performance of the mobile platform itself. The solution process employs a standard numerical optimization algorithm in rolling time-domain control, solving a finite-time-domain optimal control problem in each control cycle to obtain the optimal control input sequence at each moment in the prediction time domain. The first control variable in the solved optimal control sequence is applied to the mobile platform as the actual control command for the current moment. The entire predicted trajectory sequence, recursively generated using this optimal control sequence, is output to the UAV's tracking control module as the reference trajectory for the UAV to track the mobile platform. In the next control cycle, with a new filtered estimated state as the initial value, the above prediction and optimization process is repeated, solving for a new optimal control sequence and outputting a new predicted trajectory. This process is repeated cycle by cycle, forming a closed-loop predictive control. During the tracking process, the UAV continuously receives the continuously updated predicted trajectory and adjusts its flight path in real time according to the updated trajectory. Along with outputting the predicted trajectory, this implementation also outputs the uncertainty information at each moment along the trajectory. The uncertainty is obtained by propagating the covariance matrix of the filtered estimation output through the prediction step, reflecting the confidence level of each point on the predicted trajectory due to initial state estimation errors and model errors. After receiving this confidence level indicator, the UAV adaptively adjusts the tracking gain according to the confidence level of each predicted point during tracking control. A higher tracking gain is used for trajectory segments with higher confidence to ensure close tracking, while a lower tracking gain is used for trajectory segments with lower confidence to avoid overreaction.This implementation also includes a dynamic adjustment mechanism for the prediction time domain length. Specifically, when the mobile platform's speed is low, the prediction time domain can be set to a shorter value to reduce computational burden; when the mobile platform's speed is high, the prediction time domain is automatically extended to cover a wider prediction range, ensuring the UAV has sufficient time to respond to changes in the mobile platform's movement. The specific adjustment strategy is as follows: based on the instantaneous speed of the current mobile platform, the prediction time domain length is calculated according to a preset linear mapping relationship. Higher speeds result in a longer time domain, and lower speeds result in a shorter time domain. Simultaneously, upper and lower limits for the time domain length are set to ensure system stability and real-time performance.

[0033] In step S2, the UAV identifies the visual markers using its onboard vision sensors to determine its real-time pose relative to the mobile platform.

[0034] In practical implementation, the UAV uses an onboard vision sensor to identify visual markers and determine its real-time pose relative to the mobile platform. Specifically, a helipad is located on top of the mobile platform, and six AprilTag 2D barcodes are placed around the helipad, asymmetrically distributed. These six markers surround the edge of the helipad, forming a convex polygon enclosing the central area. The UAV, equipped with the onboard vision sensor, continuously acquires images of the helipad area at a preset frame rate, recording a precise timestamp for each frame. After preprocessing such as grayscale conversion, histogram equalization, and denoising, the acquired images are used to extract candidate marker regions. The internal encoding of each marker is decoded to determine its number. For each detected marker, the 2D coordinates of its four corner points in the image are extracted. Combined with the known 3D coordinates of the marker on the helipad plane and the camera's intrinsic parameters, an independent pose estimate of the UAV relative to the marker is calculated by solving a perspective n-point problem. After obtaining the independent poses of the six markers, nonlinear optimization is performed with the goal of minimizing the reprojection error, outputting the real-time six-DOF pose of the UAV relative to the center of the mobile platform's landing pad, including position offsets in three directions and attitude deviations in three directions. After pose calculation, the six-DOF relative pose, along with the image acquisition timestamp and pose calculation completion timestamp, are packaged into an observation data packet, which is transmitted via data link to the filtering estimation module of the mobile platform and the flight control module of the UAV. The filtering estimation module uses the timestamp information to perform observation delay compensation, and the flight control module uses the pose information to perform tracking control. When the visual sensor fails to detect a sufficient number of markers for multiple consecutive frames, visual recognition is deemed to have failed, the UAV pauses visual tracking and enters hovering mode, resuming the landing process after visual recognition recovers. For applications with high reliability requirements, the UAV can be configured with two visual sensors, a primary and a backup. Automatic switching occurs when the deviation between the primary and backup sensor calculation results exceeds a threshold, ensuring continuous availability of visual positioning.

[0035] In step S3, the real-time pose is used as the observation value to correct the filtered estimate and update the predicted motion trajectory. The UAV tracks the updated predicted motion trajectory and determines whether the landing conditions are met based on the airborne altitude sensor and visual pose. If the conditions are met, the UAV performs the landing.

[0036] In practical implementation, the real-time pose is used as an observation value to correct the filtering estimate and update the predicted motion trajectory. Specifically, after the UAV identifies visual markers on the mobile platform using its onboard vision sensor, it calculates the UAV's real-time pose relative to the mobile platform. This real-time pose is then transmitted back to the filtering estimation module on the mobile platform via a data link, serving as an external observation value input to the filtering estimator. It should be noted that the mobile platform's own filtering estimation originally relied on odometers and inertial measurement units for state prediction and updates. The real-time visual pose introduced here is a type of external absolute observation information. The visual observation reflects the absolute pose information of the mobile platform in the global coordinate system observed by the UAV, complementing the relative observation information from the odometer and inertial measurement unit. When the filtering estimation module receives the aforementioned visual pose observation, it performs time-series alignment processing on the observation according to the aforementioned observation delay compensation method. Specifically, it uses the image acquisition timestamp carried by the observation to match and align it with the current filtering time. Then, the aligned pose observation is used as the equivalent observation value for the current time and introduced into the filtering update equation. In each filtering cycle, when both observations from the mobile platform's own odometer / inertial measurement unit (IMU) and visual pose observations transmitted from the UAV are present simultaneously, the filtering estimation module processes these two types of observations sequentially according to the observation update steps. First, the filter performs an observation update using the odometer and IMU observations to obtain a pre-corrected state estimate. Using this pre-corrected state as the prediction benchmark, visual pose observations are introduced for a second observation update. Both updates calculate the residuals and Kalman gain according to the standard Kalman filter correction formula, and the two types of observation information are injected into the state estimate sequentially. The weight of visual observations in the observation update is automatically adjusted by the filter's Kalman gain. When the noise covariance of the visual observations is small, its corresponding Kalman gain is large, and the visual information has a stronger corrective effect on the state estimate; when the noise covariance of the visual observations is large, its gain is small, and the filter relies more on its own sensor's observation information. This weight allocation is adaptively completed by the filter, without the need for manually setting fixed weight coefficients. After the filtered estimation completes the fusion update, the corrected posterior state estimate and its covariance matrix are output to the rolling temporal prediction process. The rolling temporal prediction process uses the updated state estimate as the new initial prediction condition, re-executes trajectory recursion and optimal control variable solving within the prediction temporal domain, and generates an updated predicted motion trajectory starting from the current moment. Each time a new visual observation frame is acquired, the filtered estimation performs a fusion update, triggering the regeneration of the predicted trajectory. During the tracking landing process, the UAV continuously receives the aforementioned rolling updated predicted trajectory and adjusts its flight path in real time based on the latest trajectory.

[0037] It should be noted that the feedback correction in this embodiment also includes a fixed-lag window smoothing replay mechanism to further improve the smoothness and consistency of state estimation. Specifically, the state estimates and visual observations of the most recent N time moments are cached in memory, where N is preset as an appropriate positive integer based on the system's computing power and lag time. When a new visual observation arrives, it is not immediately used for state correction at the current time moment. Instead, a complete replay window is formed, extending L time moments before and after the time moment corresponding to the observation. L is less than N to ensure the replay window is always contained within the cached data range. Within the replay window, the visual observation at that moment, along with existing observation data from previous time moments within the window, is used as measurement input. Starting from the state estimate at the beginning of the window, a complete filtering update process is re-executed. Since the replay window contains data from multiple time moments before and after the newly arrived observation, the filtering update within the window can simultaneously consider historical and new information, resulting in a smoother state estimation result than single-point correction. After the filtering update within the replay window is completed, a sequence of re-smoothed state estimates for each time moment within the window is obtained. Subsequently, the corrected state estimate corresponding to the current time step in the sequence is extracted as the final state output for that time step. The correction values ​​at other time steps within the window are saved for recursion in subsequent windows, but the control output outside the current time step is not immediately changed to ensure the smoothness of the control action.

[0038] In practical implementation, the UAV tracks the updated predicted trajectory and, based on a combination of onboard altitude sensor and visual pose assessment, determines whether landing conditions are met. If they are met, landing is executed. Specifically, the rolling time-domain prediction process on the mobile platform outputs an updated predicted trajectory in each control cycle. This trajectory is presented as the predicted position, attitude, velocity, and angular velocity of the mobile platform at various points in the future, with each prediction point accompanied by an uncertainty indicator. The UAV receives the predicted trajectory in real time via its onboard communication link and uses it as a reference track for its own flight. The UAV's flight control system compares its current position with the expected position at the corresponding moment on the predicted trajectory, calculating position and velocity deviations. The selection principle for this corresponding moment is: using the current moment as a reference, plus the inherent response delay of the UAV's flight control system, to determine the predicted trajectory point corresponding to the actual control actions performed by the UAV. Based on the aforementioned deviations, the flight control system calculates the UAV's attitude and throttle commands according to a preset control law, controlling the UAV to fly along the predicted trajectory, gradually bringing the UAV's position closer to the predicted position of the center point of the mobile platform's landing pad. During the tracking process, the UAV continuously receives the rolling updates of the predicted trajectory from the mobile platform. Each time a new predicted trajectory is received, the flight control system recalculates the deviation and updates the control commands in the above manner, forming a closed-loop tracking.

[0039] Furthermore, in practical implementation, as the UAV continuously tracks the predicted trajectory and gradually approaches the mobile platform, it is necessary to accurately determine whether the UAV has entered a safe landing state. This implementation adopts a joint judgment method using an airborne altitude sensor and visual pose. Specifically, the airborne altitude sensor measures the absolute altitude of the UAV relative to the ground. In the landing scenario, the altitude value output by the altitude sensor is denoised and stabilized to determine whether the UAV has descended to the safe touchdown height range. The UAV's airborne visual sensor continuously identifies visual markers on the mobile platform and obtains the relative altitude, horizontal offset, and heading deviation of the UAV relative to the center of the mobile platform's landing pad through pose calculation. The above two types of sensor information are jointly judged: the visual pose provides the accurate relative altitude and horizontal position deviation of the UAV relative to the mobile platform, while the airborne altitude sensor provides the absolute altitude of the UAV as a redundancy check. When the deviation between the altitude values ​​measured by the two sensors is less than a preset consistency threshold, the data are considered consistent and reliable, and the weighted average of the two is taken as the final altitude determination basis. When the altitude values ​​measured by the two sensors deviate beyond a threshold, it indicates that one of the sensors may be interfered with or malfunctioning. The relative altitude information provided by the visual pose is prioritized, and a sensor data inconsistency alarm is issued. The landing conditions are determined using the following criteria: the UAV's vertical height relative to the center of the mobile platform's landing pad is no greater than 0.2 meters, the horizontal offset is no greater than ±15 millimeters, and the heading deviation is no greater than ±2 degrees. All three conditions must be met simultaneously; failure to meet any one condition results in the failure to meet the landing conditions. The three determination parameters are obtained as follows: vertical height is determined jointly by the onboard altitude sensor and the visual pose; horizontal offset is determined by the two horizontal position offset components in the visual pose calculation, i.e., the distance of the UAV relative to the center of the landing pad in the forward / backward and left / right directions; and heading deviation is determined by the yaw angle deviation component in the visual pose calculation, i.e., the angle difference between the UAV's nose direction and the preset orientation of the landing pad. The values ​​in the above judgment conditions are determined by comprehensively considering the UAV landing gear structural dimensions, the apron area, the residual sway amplitude of the mobile platform in the predicted trajectory, and the capture range of the locking mechanism. The accuracy requirements of ±15 mm for horizontal offset and ±2 degrees for heading deviation ensure that the UAV landing gear accurately falls within the capture range of the locking mechanism on the apron during landing. When all three conditions are met simultaneously, a landing permission signal is output. Upon receiving this signal, the UAV flight control system executes the landing maneuver: gradually reducing the throttle value in the current hovering state according to the preset descent rate, causing the UAV to move vertically downwards at a descent speed not exceeding 0.3 meters per second. During the descent, the visual and altitude sensors continuously operate, monitoring changes in relative altitude, horizontal offset, and heading deviation in real time. If any landing condition indicator is detected to exceed the allowable range, the landing process is immediately interrupted and a go-around procedure is initiated.When the drone descends to the point where its landing gear contacts the tarmac surface, the contact or pressure sensors on the drone's landing gear detect a sudden change in contact force, determining that the drone has physically landed on the mobile platform. At this point, the drone shuts down its motors or enters idle mode, completing the landing maneuver.

[0040] In addition, in practical implementation, if any of the following situations occur during the UAV's tracking and landing process, a go-around procedure will be triggered: First, during the landing determination phase, the three conditions of vertical altitude, horizontal deviation, and heading deviation are not simultaneously met; second, during the descent, any landing condition indicator exceeds the allowable range; third, visual recognition suddenly fails during the descent and cannot be recovered within a preset time; fourth, the uncertainty of the predicted trajectory by the mobile platform exceeds the safety threshold. When the go-around procedure is triggered, the UAV's flight control system immediately stops the descent and increases the throttle value to above the hovering power, causing the UAV to climb vertically at a speed of not less than one meter per second. During the climb, the UAV maintains its current heading until it reaches a preset safe altitude, which is usually set to be no less than five meters relative to the mobile platform. After reaching the safe altitude, the UAV re-enters tracking mode, continues to receive the predicted motion trajectory from the mobile platform, and re-executes the tracking approach and landing condition determination process. If three consecutive go-arounds fail to land successfully, it is determined that the current environmental conditions are unsuitable for landing, and the UAV switches to return-to-home mode, flying back to the preset takeoff point or alternate landing point.

[0041] In step S4, after the drone lands on the mobile platform, it triggers locking. After confirming that the clamping force is sufficient through feedback from the force sensor, a locking confirmation signal is output.

[0042] In practical implementation, the drone triggers locking upon landing on the mobile platform. After force sensors confirm that the clamping force is sufficient, a locking confirmation signal is output. Specifically, a drone locking mechanism is installed on the landing pad at the top of the mobile platform, distributed within the landing area on the pad surface, with its placement matching the structural dimensions of the drone's landing gear. The locking mechanism can employ any of the following: electromagnetically driven grippers, electrically operated push-rod latches, or pneumatic calipers, with electromagnetically driven grippers being preferred due to their advantages of fast response, controllable locking force, and self-holding even after power failure. The locking mechanism is initially in the open or released position, without interfering with the drone's normal landing contact. Each gripper's moving parts integrate a force sensor for real-time detection of the contact pressure between the gripper and the drone's landing gear. When the UAV descends to the landing gear and contacts the apron surface according to the aforementioned landing procedure, a contact detection is triggered. This detection is achieved in parallel using two methods: a contact sensor or pressure switch is installed on the UAV's landing gear. When the supporting force on the landing gear undergoes a step change, the contact sensor outputs a ground contact signal; a force sensor on the apron locking mechanism also outputs a ground contact signal when it detects a sudden pressure change from above. The presence of either of these ground contact signals indicates that the UAV has physically landed on the moving platform. These two detection methods are redundant to prevent a single sensor from missing a detection, which could prevent the locking action from triggering. Simultaneously with the ground contact signal triggering, the UAV's flight control system detects that the vertical velocity has dropped to zero and the motor current has decreased to near its no-load value, which also serves as an auxiliary basis for ground contact determination. If any two of the three determination methods simultaneously confirm ground contact, the UAV is considered to have landed stably on the apron and immediately enters the locking process. Upon confirmation of ground contact, a locking command is immediately sent to the locking mechanism. Upon receiving the command, the locking mechanism changes from an open state to a closed state, with each gripper or latch moving to its corresponding position on the UAV's landing gear. The locking mechanism operates in two phases. The first phase is the rapid feed phase, where the grippers approach the landing gear at a high speed to shorten the locking response time. The second phase is the clamping phase, where the grippers switch to a low-speed, high-torque mode after contacting the landing gear, gradually applying clamping force until a preset force threshold is reached. The switching between these two phases is controlled by the clamping force value fed back in real time by a force sensor. When the contact force detected by the force sensor reaches the preset switching threshold, the drive system automatically switches from high-speed mode to low-speed clamping mode to prevent high-speed impact damage to the UAV landing gear or apron structure.

[0043] In addition, in practical implementation, after the locking mechanism enters the clamping phase, the force sensors integrated on each gripper continuously detect the contact pressure between the gripper and the landing gear at a sampling frequency of no less than 200 times per second. The real-time readings of all force sensors are summarized, and their average value is taken as the representative value of the current clamping force. Simultaneously, the difference between the readings of each force sensor is monitored to determine whether the locking is uniform. When the representative value of the clamping force reaches the preset locking force threshold, a clamping force compliance confirmation timer is initiated. The locking force threshold is determined comprehensively based on the UAV's own weight, landing gear structural strength, maximum acceleration of the mobile platform, and the minimum friction required to prevent overturning. It is typically set to three to five times the UAV's weight to ensure that the UAV remains reliably fixed even under the most severe motion conditions. If the clamping force remains above the compliance threshold without falling for a preset duration, the clamping force is considered to be stably compliant. This duration is typically set to 200 to 500 milliseconds. During the period when the clamping force remains compliant, the difference between the readings of each force sensor is monitored simultaneously. If the difference is less than the preset equalization threshold, it indicates that the clamping force of each gripper in the locking mechanism is uniform and the locking state is good. If the difference exceeds the equalization threshold, it indicates that there is skewness or unevenness in the locking. Differential adjustment is made by adjusting the drive current of each gripper to make the clamping force of each gripper tend to be consistent until the difference falls back to within the equalization threshold. When the clamping force is stable and meets the standard within the preset duration and the force of each gripper is balanced, it is determined that the UAV has been reliably locked on the mobile platform helipad, and a locking confirmation signal is generated. The locking confirmation signal is a high-level digital signal or a specific communication protocol message. The locking confirmation signal is sent to the UAV's flight control system via the vehicle-mounted or ship-mounted communication link of the mobile platform, and simultaneously sent to the remote ground station or mother ship control center. After receiving the locking confirmation signal, the UAV flight control system confirms that the motors are completely shut down and in braking state, switches its own state to locked mode, and stops all active flight control outputs. After receiving the signal, the remote ground station or mother ship control center displays a status indicator indicating that the UAV has been safely locked on the operation interface, allowing the operator to confirm that the landing mission has been completed. After the locking confirmation signal is output, the monitoring of the locking status does not stop. During the locking period, the force sensor continues to detect the clamping force at a reduced sampling frequency of no less than fifty times per second. If, at any time after locking, the clamping force is detected to drop below the target force threshold and persists for more than the preset unlocking judgment time, a lockout is determined to have occurred, an unlocking alarm signal is immediately output, and the locking process is retried. If three consecutive re-locking attempts fail to stabilize the clamping force to the target value, a fault alarm is output, prompting the operator to intervene. Under conditions of high-speed movement or severe vibration of the mobile platform, the locking mechanism may be subjected to additional dynamic loads from the UAV, causing instantaneous spikes in the force sensor readings. To avoid misjudging the clamping force abnormality due to dynamic load spikes, a sliding window filtering method is used to process the force sensor data during continuous monitoring. The average value within the window is used as the basis for judging the clamping force. Instantaneous spikes are smoothed out in the window average and do not affect the judgment result of the locking status.When the drone needs to be released from the mobile platform for takeoff, it receives an unlocking command from the ground station or the drone. Upon receiving the unlocking command, the locking mechanism drives each gripper to move from the clamped state to the open state, while force sensors continuously monitor changes in the clamping force. When all force sensor readings return to zero, it indicates that the grippers have completely disengaged from the drone's landing gear, and an unlocking completion signal is output, confirming that the drone can take off safely.

[0044] refer to Figure 2 The diagram shows the locking mechanism structure of the autonomous landing and automatic locking control system for a UAV mobile platform provided in this application. The locking mechanism includes a fixed base 100, a left gripper 101, a right gripper 102, a V-shaped clamping port 103, a left rotation shaft 104, a right rotation shaft 105, a left electromagnetic coil 106, a right electromagnetic coil 107, a left return spring 108, a right return spring 109, a left force sensor 110, a right force sensor 111, a displacement sensor 112, an elastic buffer pad 113, a guide cone surface 114, and a UAV landing gear main shaft 115. The left gripper 101 and right gripper 102 are pivotally connected to the left and right sides of the fixed base 100 via the left rotation shaft 104 and right rotation shaft 105, respectively. The free ends of the two are arranged opposite each other and together form a V-shaped clamping opening 103. The opening width of the V-shaped clamping opening 103 gradually narrows from the outside to the inside to form a guide structure. The entrance ends of the left gripper 101 and the right gripper 102 are respectively provided with guide cone surfaces 114. The left force sensor 110 and the right force sensor 111 are respectively embedded on the clamping surfaces. An elastic buffer pad 113 is attached to the inside of the force sensor. The fixed base 100 is provided with a left electromagnetic coil 106 and a right electromagnetic coil 107, as well as a left return spring 108 and a right return spring 109. A displacement sensor 112 is provided at the left rotation axis 104 or the right rotation axis 105. The UAV landing gear main shaft 115 is vertically inserted into the V-shaped clamping opening 103, and its axis coincides with the symmetrical center line of the V-shaped clamping opening 103.

[0045] Therefore, this application demonstrates the following: First, by filtering and estimating the motion state of the mobile platform and predicting its future trajectory, the UAV can anticipate the mobile platform's motion trend in the future, providing a motion prediction basis for subsequent landing. Second, by using an onboard visual sensor to identify visual markers on the mobile platform and calculate the real-time relative pose, the positioning continuity under partial occlusion conditions is ensured, providing a stable and reliable relative navigation reference for landing. Third, the visual pose is used as an observation value to correct the filtered estimation and update the predicted trajectory, enabling the trajectory prediction of the mobile platform to integrate visual absolute positioning information and inertial relative perception information, avoiding erroneous landings caused by single-sensor misjudgment or environmental interference, and improving the safety and reliability of landing triggering. Finally, after the UAV lands on the mobile platform, the clamping force is fed back in real time by a force sensor, ensuring that the locking mechanism can reliably fix the UAV even in turbulent environments, improving the adaptability and long-term reliability of the locking mechanism.

[0046] In summary, the technical solution adopted in this application can realize fully autonomous landing and adaptive reliable locking of UAVs under turbulent conditions on mobile platforms, thereby improving the landing success rate and locking safety.

[0047] Example 2: This application provides an autonomous landing and automatic locking control system for an unmanned aerial vehicle (UAV) mobile platform, referencing... Figure 3 As shown in the figure, this is a block structure diagram of the autonomous landing and automatic locking control system for an unmanned aerial vehicle (UAV) mobile platform according to this embodiment of the present application. The control system includes: The trajectory prediction module 201 is used to filter and estimate the motion state of the mobile platform, output the current motion state of the mobile platform, perform rolling time-domain prediction based on the current motion state, and output the predicted motion trajectory of the mobile platform. The pose calculation module 202 is used by the UAV to identify the visual markers through the onboard visual sensor and determine the real-time pose of the UAV relative to the mobile platform. The landing decision module 203 is used to use the real-time pose as an observation value to correct the filtered estimate, update the predicted motion trajectory, and the UAV tracks the updated predicted motion trajectory. The UAV jointly determines whether the landing conditions are met based on the airborne altitude sensor and visual pose, and executes the landing if the conditions are met. The locking control module 204 is used to trigger locking after the UAV lands on the mobile platform. After confirming that the clamping force is sufficient through feedback from the force sensor, it outputs a locking confirmation signal.

[0048] The foregoing detailed an example of an autonomous landing and automatic locking control method for a UAV mobile platform provided in this application. It is understood that the corresponding device, in order to achieve the above functions, includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0049] In embodiment three, this application also provides a computer device, the computer device including a memory and a processor, the memory for storing computer programs, and the processor for calling and running the computer programs from the memory, so that the computer device executes the above-described method for autonomous landing and automatic locking control of a UAV mobile platform.

[0050] In this embodiment, reference Figure 4 The dashed lines in the figure indicate that the unit or module is optional. This figure is a schematic diagram of the structure of a computer device for an autonomous landing and automatic locking control system for a UAV mobile platform according to an embodiment of this application. The autonomous landing and automatic locking control method for a UAV mobile platform in the above embodiment can be achieved through… Figure 4 The computer device shown is used to implement this, and the computer device includes at least one processor 501, a memory 502 and at least one communication unit 505. The computer device may be a terminal device, a server or a chip.

[0051] Processor 501 can be a general-purpose processor or a special-purpose processor. For example, processor 501 can be a central processing unit (CPU), which can be used to control computer devices, execute software programs, and process data from software programs. The computer device may also include a communication unit 505 for receiving and transmitting signals.

[0052] For example, the computer device may be a chip, the communication unit 505 may be the input and / or output circuit of the chip, or the communication unit 505 may be the communication interface of the chip, and the chip may be a component of a terminal device, network device or other device.

[0053] For example, the computer device may be a terminal device or a server, and the communication unit 505 may be a transceiver of the terminal device or the server, or the communication unit 505 may be a transceiver circuit of the terminal device or the server.

[0054] The computer device may include one or more memories 502 storing a program 504. The program 504 can be executed by a processor 501 to generate instructions 503, causing the processor 501 to perform the methods described in the above method embodiments according to the instructions 503. Optionally, the memory 502 may also store data (such as a target audit model). Optionally, the processor 501 may also read data stored in the memory 502, which may be stored at the same storage address as the program 504, or the data may be stored at a different storage address than the program 504.

[0055] The processor 501 and memory 502 can be configured separately or integrated together, for example, integrated on the system-on-chip (SOC) of the terminal device.

[0056] It should be understood that each step of the above method embodiment can be completed by hardware logic circuits or software instructions in processor 501. Processor 501 can be a central processing unit, digital signal processor (DSP), application specific integrated circuit (ASIC), field programmable gate array (FPGA), or other programmable logic device, such as discrete gate, transistor logic device, or discrete hardware component.

[0057] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0058] In embodiment four, this application also provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the above-described method for autonomous landing and automatic locking control of a UAV mobile platform.

[0059] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0060] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of the invention. Therefore, if these modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A method for autonomous landing and automatic locking control of an unmanned aerial vehicle (UAV) mobile platform, characterized in that, The mobile platform is an unmanned vehicle or unmanned vessel, and visual markers are installed on the mobile platform. The control method includes: The motion state of the mobile platform is filtered and estimated, and the current motion state of the mobile platform is output. Based on the current motion state, a rolling time-domain prediction is performed, and the predicted motion trajectory of the mobile platform is output. The drone identifies the visual markers using its onboard visual sensors to determine its real-time pose relative to the mobile platform. The real-time pose is used as the observation value to correct the filtered estimate and update the predicted motion trajectory. The UAV tracks the updated predicted motion trajectory and determines whether the landing conditions are met based on the airborne altitude sensor and visual pose. If the conditions are met, the UAV performs the landing. After the drone lands on the mobile platform, it triggers a lock. Once the force sensor confirms that the clamping force is sufficient, a lock confirmation signal is output.

2. The method for autonomous landing and automatic locking control of a UAV mobile platform as described in claim 1, characterized in that, The visual markers are set around the helipad on top of the mobile platform. The visual markers are AprilTag two-dimensional barcodes, and there are 6 of them, which are asymmetrically distributed.

3. The method for autonomous landing and automatic locking control of a UAV mobile platform as described in claim 1, characterized in that, The filtering estimation incorporates observation delay compensation: historical visual observations and their timestamps are cached, the current filtered prediction is reverse-calculated to the corresponding historical moment, and the residual between the prediction and the observation is determined and then injected into the current estimate.

4. The method for autonomous landing and automatic locking control of a UAV mobile platform as described in claim 1, characterized in that, Based on the current motion state, a rolling time-domain prediction is performed, and the predicted motion trajectory of the mobile platform is output, specifically including: Using the current motion state as the initial value, a trajectory sequence in the prediction time domain is recursively generated using a kinematic model. The optimal control quantity is solved with tracking error and control increment as objectives, and the predicted motion trajectory of the mobile platform is output in a rolling manner.

5. The method for autonomous landing and automatic locking control of a UAV mobile platform as described in claim 1, characterized in that, The feedback correction using a fixed lag window for smooth replay specifically includes: caching the state estimates and visual observations of the most recent N moments; when a new observation arrives, forming a replay window with L moments before and after the current moment as the center; redoing the filtering update within the window and allocating the correction amount to the current moment.

6. The method for autonomous landing and automatic locking control of a UAV mobile platform as described in claim 1, characterized in that, The rolling time-domain prediction includes execution delay compensation: the total end-to-end delay from image acquisition to attitude calculation to flight control command issuance is measured, and the starting point of the predicted time-domain reference trajectory is shifted forward by this total delay.

7. The autonomous landing and automatic locking control method for a UAV mobile platform as described in claim 1, characterized in that, The landing conditions are: the height of the UAV relative to the mobile platform is ≤0.2m, the horizontal offset is ≤±15mm, and the heading deviation is ≤±2°. If any one of these conditions is not met, the UAV will attempt to go around.

8. A control system for autonomous landing and automatic locking of a mobile unmanned aerial vehicle (UAV) platform, used to execute the autonomous landing and automatic locking control method for a mobile UAV platform as described in any one of claims 1 to 7, characterized in that, The autonomous landing and automatic locking control system for a mobile UAV platform includes: The trajectory prediction module is used to filter and estimate the motion state of the mobile platform, output the current motion state of the mobile platform, perform rolling time-domain prediction based on the current motion state, and output the predicted motion trajectory of the mobile platform. The pose calculation module is used by the UAV to identify the visual markers through the onboard visual sensor and determine the real-time pose of the UAV relative to the mobile platform. The landing decision module is used to use the real-time pose as an observation value to correct the filtered estimate, update the predicted motion trajectory, and the UAV tracks the updated predicted motion trajectory. Based on the airborne altitude sensor and visual pose, it jointly determines whether the landing conditions are met, and executes the landing if the conditions are met. The locking control module is used to trigger locking after the drone lands on the mobile platform. After confirming that the clamping force is sufficient through feedback from the force sensor, it outputs a locking confirmation signal.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store computer programs, and the processor is used to call and run the computer programs from the memory, so that the computer device executes the autonomous landing and automatic locking control method for a UAV mobile platform according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions or code that, when executed on a computer, cause the computer to implement the autonomous landing and automatic locking control method for a UAV mobile platform as described in any one of claims 1 to 7.

Images