Unmanned aerial vehicle precision landing method and system

Abstract

The present application relates to the technical field of unmanned aerial vehicle autonomous navigation, in particular to a kind of unmanned aerial vehicle precision landing method and system, the method is first by millimeter wave radar and multispectral imaging fusion perception module is scanned to landing area, identify the preset phase encoding active light beacon;Then construct UWB-IMU-barometer layered fusion positioning architecture, realize the coarse positioning of unmanned aerial vehicle position by federated Kalman filtering;Then the precise distance between unmanned aerial vehicle and beacon is calculated using phase difference ranging method, combined with the geometric relationship of beacon array, the high-precision three-dimensional coordinates of unmanned aerial vehicle are solved;Finally, based on the adaptive landing control algorithm of deep reinforcement learning, dynamically adjust control strategy according to real-time environmental disturbance, while cooperating with the active attitude adjustment mechanism of landing platform, precision landing is completed together.The present application can realize millimeter level all-weather precision landing, and is suitable for complex application scenarios such as offshore operation, border patrol, emergency rescue, etc.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation technology for unmanned aerial vehicles (UAVs), specifically a method and system for precise landing of UAVs. Background Technology

[0002] As drone technology matures, its applications are rapidly expanding from civilian consumer use to industrial and military applications. In specialized scenarios such as offshore oil platform inspections, border patrols, mountain emergency rescue, and maritime search and rescue, drones often need to perform missions under extremely complex and harsh environmental conditions. These scenarios place extremely high demands on the drones' precision landing capabilities, requiring not only accurate landings in clear daytime weather but also high reliability and accuracy in extreme environments such as nighttime, rain, fog, strong winds, and strong electromagnetic interference.

[0003] Currently, existing drone precision landing technologies have many limitations when facing extreme environments. While vision-based positioning technology can achieve high accuracy in clear daytime conditions, image quality deteriorates significantly in low-light conditions such as nighttime, rain, fog, and strong light or backlight, leading to beacon recognition failures. Although lidar-based positioning technology can operate at night, the laser beam is scattered by water droplets in rain and fog, resulting in decreased point cloud data quality and a significant reduction in positioning accuracy. Furthermore, lidar equipment is expensive, bulky, and heavy, limiting its application in small drones.

[0004] While satellite navigation-based positioning technology can achieve all-weather positioning, its accuracy is generally only at the meter level, which is insufficient for precise landing. Furthermore, satellite navigation signals are susceptible to electromagnetic interference, posing serious safety hazards in military and certain specialized industrial applications. In remote areas such as at sea and in mountainous regions, satellite signals may also attenuate or be lost.

[0005] UWB-based positioning technology has advantages such as high accuracy, strong anti-interference ability, and low power consumption. However, its effective measurement distance is limited, generally only achieving high-precision positioning within a range of tens of meters. Moreover, UWB signals are easily blocked and reflected by metal objects, and the reliability of positioning in complex environments needs to be improved.

[0006] Besides the issues with positioning technology, existing drone landing control algorithms also have shortcomings. Most drones still use traditional PID control algorithms, which perform well for linear systems, but are less effective for drones, which are nonlinear, strongly coupled, and multivariable systems. Especially in complex environments with wind interference and airflow disturbances, PID control algorithms are prone to overshoot and oscillation, leading to decreased landing accuracy and stability.

[0007] Furthermore, most existing precision landing technologies focus solely on the control of the drone itself, neglecting the collaborative role of the landing platform. When landing on dynamic platforms such as those used for maritime operations or moving vehicles, the platform itself undergoes attitude changes due to waves or vehicle movement, posing a significant challenge to precise drone landing. Relying solely on the drone's own control to adapt to the platform's motion is not only difficult to control but also compromises landing accuracy and reliability.

[0008] Therefore, developing a method and system for precise UAV landing that can achieve high accuracy and high reliability in extreme environments such as nighttime, rain, fog, and strong electromagnetic interference is of great practical significance and application value. Summary of the Invention

[0009] To address the problems in existing technologies, this invention provides a method and system for precise landing of unmanned aerial vehicles (UAVs), which solves problems such as positioning failure in extreme environments, poor anti-interference capabilities, and unsatisfactory control effects.

[0010] The technical solution adopted by this invention to solve its technical problem is: a method for precise landing of a drone, comprising the following steps: S1: The landing area is scanned by a sensing module that combines millimeter-wave radar and multispectral imaging on the UAV to identify a pre-set phase-coded active light-emitting beacon array. S2: Collect UWB ranging data, IMU data and barometer data of the UAV, construct a hierarchical fusion positioning architecture, and use the federated Kalman filter algorithm to coarsely locate the position of the UAV and obtain the initial three-dimensional coordinates of the UAV. S3: When the UAV enters the effective communication range of the beacon, the phase difference ranging module is activated to calculate the precise distance between the UAV and each beacon, and the high-precision three-dimensional coordinates of the UAV relative to the beacon array are calculated by combining the geometric relationship of the beacon array. S4: Establish a collaborative dynamics model between the UAV and the landing platform. Based on high-precision three-dimensional coordinates and the preset optimal landing trajectory, use an adaptive landing control algorithm based on deep reinforcement learning to generate UAV control quantities. S5: Sends the UAV control signals to the flight controller and simultaneously sends attitude adjustment commands to the landing platform, controlling the UAV and the landing platform to work together to complete a precise landing.

[0011] Specifically, the sensing module for fusing millimeter-wave radar and multispectral imaging in step S1 includes: acquiring point cloud data of the landing area using millimeter-wave radar to detect potential beacon targets; acquiring multispectral images of the beacon using the multispectral imaging module to extract the phase coding features of the beacon; and performing spatiotemporal registration of the radar point cloud data and the multispectral image to fuse the feature information of the two modes and achieve accurate beacon identification.

[0012] Specifically, the hierarchical fusion positioning architecture described in step S2 includes: a first layer is a UWB sub-filter that processes ranging data between the UWB base station and the UAV; a second layer is an IMU-barometer sub-filter that processes IMU and barometer data; and a third layer is a main filter that fuses the outputs of the two sub-filters to obtain the final coarse positioning result.

[0013] Specifically, the phase difference ranging method described in step S3 includes: each beacon in the beacon array emits a continuous sine wave signal of a specific frequency; the UAV receives signals from multiple beacons and measures the phase difference between different beacon signals; and the precise distance between the UAV and each beacon is calculated based on the phase difference and signal wavelength.

[0014] Specifically, the adaptive landing control algorithm based on deep reinforcement learning described in step S4 includes: constructing a Markov decision process that includes environmental state, UAV state, and control actions; using landing accuracy, landing time, and energy consumption as reward functions; training the control policy network using a deep deterministic policy gradient algorithm; and dynamically adjusting the control policy according to real-time environmental disturbances during the actual landing process.

[0015] Specifically, the active attitude adjustment of the landing platform described in step S5 includes: the landing platform is equipped with a three-axis attitude adjustment mechanism and a tilt sensor; the angle that the landing platform needs to adjust is calculated based on the real-time position and attitude information of the UAV; and the three-axis attitude adjustment mechanism is controlled to adjust the attitude of the platform so that the platform always remains perpendicular to the landing direction of the UAV.

[0016] A precision landing system for unmanned aerial vehicles (UAVs) includes: The multimodal sensing module is used to identify a pre-defined phase-coded active luminous beacon array by fusing millimeter-wave radar with multispectral imaging; The coarse positioning module is used to collect UWB ranging data, IMU data and barometer data, and to perform coarse positioning of the UAV using a federated Kalman filter algorithm; The precise positioning module is used to calculate the precise distance between the UAV and the beacon using the phase difference ranging method, and to solve the high-precision three-dimensional coordinates of the UAV. The cooperative control module is used to establish a cooperative dynamics model between the UAV and the landing platform, and uses an adaptive landing control algorithm based on deep reinforcement learning to generate control quantities. The flight execution module is used to receive control signals from the UAV and control the UAV's flight. The platform execution module is used to receive attitude adjustment commands and control the landing platform to adjust its attitude.

[0017] Specifically, the phase-coded active light-emitting beacon array consists of four beacons arranged in a square; each beacon integrates an infrared LED transmitter, a radio frequency signal transmitter, and a microcontroller; each beacon emits a unique phase-coded signal to distinguish different beacons.

[0018] Specifically, the precise positioning module includes: A signal receiving unit is used to receive radio frequency signals transmitted by multiple beacons; The phase difference measurement unit is used to measure the phase difference between different beacon signals; The distance calculation unit is used to calculate the precise distance between the UAV and each beacon based on the phase difference and signal wavelength. The coordinate calculation unit is used to calculate the high-precision three-dimensional coordinates of the UAV using the least squares method based on the known positions and measured distances of multiple beacons.

[0019] Specifically, the collaborative control module includes: The model building unit is used to build a collaborative dynamics model of the UAV and the landing platform; The policy training unit is used to train the control policy network using a deep deterministic policy gradient algorithm. The control generation unit is used to generate UAV control quantities and platform attitude adjustment commands based on real-time status information. The disturbance compensation unit is used to estimate environmental disturbances in real time and compensate for control variables.

[0020] The beneficial effects of this invention are: By employing millimeter-wave radar and multispectral imaging fusion sensing, it can accurately identify phase-coded active luminous beacons in all weather conditions, effectively resisting environmental interference such as light, rain, and fog, and significantly improving the stability and environmental adaptability of beacon identification.

[0021] A hierarchical fusion positioning architecture of UWB-IMU-barometer is constructed, which, combined with federated Kalman filtering, achieves stable coarse positioning. The system has strong fault tolerance, and the failure of a single sensor does not affect the overall positioning, ensuring reliable positioning during the long-distance approach phase.

[0022] By using phase difference ranging combined with the geometric relationship of beacon arrays to calculate high-precision three-dimensional coordinates, the positioning accuracy is greatly improved, providing a reliable positional basis for precise landing.

[0023] The deep reinforcement learning adaptive landing control algorithm can dynamically adapt to environmental disturbances and the nonlinear dynamic characteristics of UAVs, improve the shortcomings of traditional control such as overshoot and oscillation, and enhance the stability of landing control.

[0024] Establish a collaborative control mechanism between the UAV and the landing platform, and coordinate with the platform to actively adjust its attitude to achieve collaborative operation between the UAV and the platform, thereby reducing the difficulty of UAV control and further improving landing accuracy and stability.

[0025] The system has strong resistance to electromagnetic interference and all-weather operation capabilities, and is suitable for complex scenarios such as maritime, mountainous, and emergency rescue and disaster relief, breaking through the environmental limitations of traditional precision landing technology.

[0026] Equipped with a fault-tolerant mechanism, the positioning mode can be automatically switched in case of beacon malfunction, ensuring a safe and continuous landing process; the equipment is reasonably priced, balancing high performance and practicality, and is easy to promote and apply to various types of drones. Attached Figure Description

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0028] Figure 1 A flowchart of a precise landing method for unmanned aerial vehicles provided by the present invention; Figure 2 This invention provides an architecture diagram of a precision landing system for unmanned aerial vehicles (UAVs). Figure 3 This is a schematic diagram of the arrangement of a phase-coded active light-emitting beacon array. Detailed Implementation

[0029] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0030] like Figure 1 and Figure 3 As shown, the present invention provides a method for precise landing of a drone, comprising the following steps: S1: The landing area is scanned by a sensing module that combines millimeter-wave radar and multispectral imaging on the UAV to identify a pre-set phase-coded active light-emitting beacon array. S2: Collect UWB ranging data, IMU data and barometer data of the UAV, construct a hierarchical fusion positioning architecture, and use the federated Kalman filter algorithm to coarsely locate the position of the UAV and obtain the initial three-dimensional coordinates of the UAV. S3: When the UAV enters the effective communication range of the beacon, the phase difference ranging module is activated to calculate the precise distance between the UAV and each beacon, and the high-precision three-dimensional coordinates of the UAV relative to the beacon array are calculated by combining the geometric relationship of the beacon array. S4: Establish a collaborative dynamics model between the UAV and the landing platform. Based on high-precision three-dimensional coordinates and the preset optimal landing trajectory, use an adaptive landing control algorithm based on deep reinforcement learning to generate UAV control quantities. S5: Sends the UAV control signals to the flight controller and simultaneously sends attitude adjustment commands to the landing platform, controlling the UAV and the landing platform to work together to complete a precise landing.

[0031] The sensing module for fusing millimeter-wave radar and multispectral imaging described in step S1 specifically includes: acquiring point cloud data of the landing area using millimeter-wave radar to detect potential beacon targets; acquiring multispectral images of the beacon using the multispectral imaging module to extract the phase coding features of the beacon; performing spatiotemporal registration of the radar point cloud data and the multispectral image, fusing the feature information of the two modes to achieve accurate beacon identification.

[0032] The layered fusion positioning architecture described in step S2 specifically includes: the first layer is a UWB sub-filter, which processes the ranging data between the UWB base station and the UAV; the second layer is an IMU-barometer sub-filter, which processes IMU and barometer data; and the third layer is the main filter, which fuses the outputs of the two sub-filters to obtain the final coarse positioning result.

[0033] The phase difference ranging method described in step S3 specifically includes: each beacon in the beacon array emits a continuous sine wave signal of a specific frequency; the UAV receives signals from multiple beacons and measures the phase difference between different beacon signals; and the precise distance between the UAV and each beacon is calculated based on the phase difference and signal wavelength.

[0034] The adaptive landing control algorithm based on deep reinforcement learning described in step S4 specifically includes: constructing a Markov decision process that includes environmental state, UAV state, and control actions; using landing accuracy, landing time, and energy consumption as reward functions; training the control policy network using a deep deterministic policy gradient algorithm; and dynamically adjusting the control policy according to real-time environmental disturbances during the actual landing process.

[0035] The active attitude adjustment of the landing platform described in step S5 specifically includes: the landing platform is equipped with a three-axis attitude adjustment mechanism and a tilt sensor; the angle that the landing platform needs to adjust is calculated based on the real-time position and attitude information of the UAV; and the three-axis attitude adjustment mechanism is controlled to adjust the attitude of the platform so that the platform always remains perpendicular to the landing direction of the UAV.

[0036] As attached Figure 2 and Figure 3 As shown, the present invention also provides a precision landing system for unmanned aerial vehicles, comprising: The multimodal sensing module is used to identify a pre-defined phase-coded active luminous beacon array by fusing millimeter-wave radar with multispectral imaging; The coarse positioning module is used to collect UWB ranging data, IMU data and barometer data, and to perform coarse positioning of the UAV using a federated Kalman filter algorithm; The precise positioning module is used to calculate the precise distance between the UAV and the beacon using the phase difference ranging method, and to solve the high-precision three-dimensional coordinates of the UAV. The cooperative control module is used to establish a cooperative dynamics model between the UAV and the landing platform, and uses an adaptive landing control algorithm based on deep reinforcement learning to generate control quantities. The flight execution module is used to receive control signals from the UAV and control the UAV's flight. The platform execution module is used to receive attitude adjustment commands and control the landing platform to adjust its attitude.

[0037] The phase-coded active light-emitting beacon array consists of four beacons arranged in a square. Each beacon integrates an infrared LED transmitter, a radio frequency signal transmitter, and a microcontroller. Each beacon emits a unique phase-coded signal to distinguish different beacons.

[0038] The precise positioning module includes: A signal receiving unit is used to receive radio frequency signals transmitted by multiple beacons; The phase difference measurement unit is used to measure the phase difference between different beacon signals; The distance calculation unit is used to calculate the precise distance between the UAV and each beacon based on the phase difference and signal wavelength. The coordinate calculation unit is used to calculate the high-precision three-dimensional coordinates of the UAV using the least squares method based on the known positions and measured distances of multiple beacons.

[0039] The collaborative control module includes: The model building unit is used to build a collaborative dynamics model of the UAV and the landing platform; The policy training unit is used to train the control policy network using a deep deterministic policy gradient algorithm. The control generation unit is used to generate UAV control quantities and platform attitude adjustment commands based on real-time status information. The disturbance compensation unit is used to estimate environmental disturbances in real time and compensate for control variables.

[0040] Example 1: This example provides a method for precise landing of a UAV, applied to the autonomous and precise landing of a hexacopter UAV on an offshore oil platform. The hexacopter UAV is equipped with millimeter-wave radar, a multispectral imaging module, a UWB tag, an IMU, a barometer, and a high-performance onboard computer. The landing platform is equipped with a phase-coded active light-emitting beacon array, a UWB base station, a three-axis attitude adjustment mechanism, and a wireless communication module. The beacon array consists of four beacons arranged in a square at the four corners of the landing platform.

[0041] The specific steps of the precise drone landing method in this embodiment are as follows: S1: The landing area is scanned by a sensing module that combines millimeter-wave radar and multispectral imaging onboard the UAV to identify a pre-set phase-coded active light-emitting beacon array.

[0042] Millimeter-wave radar operates in the 77 GHz band, featuring the ability to penetrate rain and fog and is unaffected by ambient light, enabling it to function normally at night and in adverse weather conditions. The multispectral imaging module includes both visible and near-infrared spectral channels, allowing it to acquire rich spectral information. The phase-coded active luminescent beacon integrates an infrared LED transmitter and a radio frequency signal transmitter, capable of emitting infrared light and radio frequency signals at specific frequencies; each beacon has a unique phase code.

[0043] When approaching the landing platform from a distance, the millimeter-wave radar is first activated to scan the landing area and acquire point cloud data. Potential beacon targets are detected from the point cloud data using point cloud clustering and target detection algorithms. Then, based on the target location detected by the radar, the multispectral imaging module is guided to image the target area. After acquiring the multispectral image of the beacon, the multispectral imaging module extracts the beacon's infrared emission features and phase-coded features.

[0044] To improve the accuracy and robustness of beacon recognition, spatiotemporal registration is performed between millimeter-wave radar point cloud data and multispectral images. First, temporal registration is achieved by synchronizing the data from the two sensors using timestamps. Then, spatial registration is achieved by projecting the radar point cloud onto the image plane using the calibrated extrinsic parameter matrix between the radar and camera. Finally, the radar range information and the spectral feature information from the multispectral images are fused to achieve accurate beacon recognition. Beacon recognition is considered successful when at least three beacons are identified.

[0045] S2: Collect UWB ranging data, IMU data, and barometer data from the UAV, construct a hierarchical fusion positioning architecture, and use the federated Kalman filter algorithm to coarsely locate the UAV's position and obtain the UAV's initial three-dimensional coordinates.

[0046] The UWB positioning system consists of four UWB base stations mounted on a landing platform and one UWB tag on the drone. The UWB base stations and the tag communicate via ultra-wideband pulse signals, and the distance is calculated by measuring the signal's time-of-flight. The IMU (Integrated Measurement Unit) measures the drone's three-axis acceleration and angular velocity; integration yields the drone's position and attitude. A barometer measures the drone's altitude.

[0047] To improve the accuracy and reliability of coarse positioning, a three-layer federated Kalman filter architecture was constructed. The first layer is a UWB sub-filter that processes ranging data between four UWB base stations and the tag, calculates the UAV's position using trilateration, and estimates the covariance matrix of the position error. The second layer is an IMU-barometer sub-filter that fuses the IMU's acceleration and angular velocity data with the barometer's altitude data to obtain the UAV's position and attitude estimates. The third layer is the master filter, which optimally fuses the outputs of the two sub-filters to obtain the final coarse positioning result.

[0048] The federated Kalman filter algorithm has the advantages of good fault tolerance and low computational cost. When a sensor malfunctions or its data is abnormal, only the corresponding sub-filter is affected, without causing the entire positioning system to fail. Through this hierarchical fusion architecture, a coarse positioning result with an accuracy of about 0.5 meters can be obtained, meeting the requirements of UAVs approaching landing platforms from a distance.

[0049] S3: When the UAV enters the effective communication range of the beacon, the phase difference ranging module is activated to calculate the precise distance between the UAV and each beacon, and the high-precision three-dimensional coordinates of the UAV relative to the beacon array are calculated by combining the geometric relationship of the beacon array.

[0050] When the UAV is approximately 30 meters from the landing platform, it enters the effective communication range of the phase-coded beacons. At this point, the phase difference ranging module is activated. Each beacon in the beacon array simultaneously transmits a continuous sinusoidal radio frequency signal at a specific frequency, and each beacon's signal has a unique phase offset. The signal receiving unit on the UAV simultaneously receives the signals from all four beacons and amplifies, filters, and digitizes them.

[0051] The phase difference measurement unit calculates the phase difference between signals from different beacons. Since signals experience phase delay during propagation, the phase difference is proportional to the distance traveled. Based on the phase difference and signal wavelength, the distance difference between the UAV and two beacons can be calculated. By measuring the phase difference between multiple beacon pairs, multiple distance difference equations can be obtained.

[0052] By combining the known position coordinates of four beacons and the measured distance differences, the three-dimensional coordinates of the UAV can be calculated using the least squares method. Phase difference ranging is characterized by high accuracy and strong anti-interference capability, achieving millimeter-level ranging accuracy. This method can yield high-precision three-dimensional coordinates with an accuracy of approximately 5 millimeters, providing reliable position information for precise landing.

[0053] S4: Establish a collaborative dynamics model between the UAV and the landing platform. Based on high-precision three-dimensional coordinates and a preset optimal landing trajectory, use an adaptive landing control algorithm based on deep reinforcement learning to generate UAV control quantities.

[0054] First, a six-degree-of-freedom nonlinear dynamic model of the hexarotor UAV is established, considering factors such as the UAV's mass, moment of inertia, air resistance, and gyroscopic effects. Then, a dynamic model of the landing platform is established, considering the platform's mass, moment of inertia, and the dynamic characteristics of the three-axis attitude adjustment mechanism. Finally, the two models are coupled to obtain a cooperative dynamic model of the UAV and the landing platform.

[0055] The preset optimal landing trajectory employs a segmented design. At higher altitudes from the platform, a straight descent trajectory is used to shorten landing time; at lower altitudes, a hovering-slow descent trajectory is used to improve landing accuracy. The trajectory parameters can be adjusted according to the specific application scenario.

[0056] An adaptive landing control algorithm based on the Deep Deterministic Policy Gradient (DDPG) algorithm is adopted. A state space is constructed, including the UAV's position, velocity, attitude angles, angular velocities, and platform attitude angles. An action space is constructed, including the four motor speeds and the adjustment angles of the platform's three axes. The reward function is designed as a weighted sum of landing accuracy, landing time, and energy consumption, with landing accuracy having the highest weight.

[0057] During the offline training phase, a large amount of training data is generated using a simulation environment to train the control policy network and the value network. During the actual landing process, real-time state information is input into the trained policy network, directly outputting the optimal control input. Simultaneously, the algorithm can dynamically adjust the control policy based on real-time environmental disturbances, such as wind interference and platform motion, achieving adaptive control.

[0058] S5: Sends the UAV control signals to the flight controller and simultaneously sends attitude adjustment commands to the landing platform, controlling the UAV and the landing platform to work together to complete a precise landing.

[0059] After receiving control signals from the UAV, the flight controller converts them into PWM signals for the four motors, controlling their speed and thus the UAV's flight attitude and position. Simultaneously, the onboard computer sends attitude adjustment commands to the landing platform via a wireless communication module. Upon receiving these commands, the controller on the landing platform calculates the required angle adjustments for the three-axis attitude adjustment mechanism based on the platform's current attitude information.

[0060] The three-axis attitude adjustment mechanism consists of three servo motors, capable of adjusting the platform's roll, pitch, and yaw angles independently. By adjusting the platform's attitude in real time, it ensures the platform remains perpendicular to the drone's descent direction, thereby reducing the drone's control complexity and improving landing accuracy and stability. When the drone's height above the platform is less than 0.2 meters, the platform stops attitude adjustment, and the drone slowly lands at the platform's center.

[0061] Throughout the landing process, the system continuously updates the UAV's high-precision three-dimensional coordinates and recalculates the control quantities and platform attitude adjustment commands based on the new coordinates, forming a closed-loop collaborative control system.

[0062] Example 2 differs from Example 1 in that the multimodal sensing module employs an 80GHz frequency-modulated continuous wave millimeter-wave radar and an 8-channel multispectral imaging module. The 80GHz millimeter-wave radar offers higher range and angular resolution, enabling more precise detection of the beacon's position and shape. The 8-channel multispectral imaging module covers the spectral range from visible light to mid-infrared, acquiring richer spectral information and further improving the accuracy of beacon identification.

[0063] Furthermore, multi-frequency phase difference ranging technology is employed in the phase difference ranging module. By transmitting multiple signals at different frequencies and measuring the phase difference at different frequencies, the phase ambiguity problem is solved, and the effective range of phase difference ranging is expanded. Compared with single-frequency phase difference ranging technology, multi-frequency phase difference ranging technology expands the effective measurement range by more than 10 times while maintaining millimeter-level ranging accuracy.

[0064] The other steps in this embodiment are exactly the same as in Embodiment 1. By employing higher-performance millimeter-wave radar, a multispectral imaging module, and multi-frequency phase difference ranging technology, the beacon identification accuracy of this embodiment reaches 99.8%, the positioning accuracy reaches 3 mm, and the landing accuracy reaches 5 mm, representing a further improvement over Embodiment 1. Especially under adverse weather conditions such as heavy rain and dense fog, it can still maintain high performance.

[0065] Example 3 differs from Example 1 in that it introduces an attention mechanism and transfer learning into the deep reinforcement learning-based adaptive landing control algorithm. The attention mechanism allows the algorithm to focus more on state variables that significantly impact landing accuracy, such as position and velocity errors, thereby improving control accuracy and efficiency. Transfer learning enables the rapid transfer of models trained in simulation environments to real-world environments, greatly reducing training time and costs in real-world scenarios.

[0066] Furthermore, this embodiment incorporates a fault-tolerance mechanism. When a beacon malfunctions or is obstructed, the system can automatically detect the fault and utilize the remaining beacons for positioning. When fewer than three beacons remain, the system automatically switches to UWB-IMU fusion positioning mode to ensure the continuity and safety of the landing process.

[0067] The other steps in this embodiment are exactly the same as in Embodiment 1. By introducing attention mechanisms, transfer learning, and fault tolerance mechanisms, the control algorithm in this embodiment has better adaptability and robustness. It can maintain high landing accuracy and reliability even under complex and changing environmental conditions. At a wind speed of 8 m / s, the landing accuracy can still be controlled within 1 cm.

[0068] Comparative Example 1: This comparative example uses the traditional monocular vision positioning method. It uses a regular RGB camera to acquire ground images, uses the original YOLOv8 algorithm to identify preset RGB beacons, calculates the position of the UAV through monocular vision ranging, and finally uses a PID control algorithm to control the UAV to land.

[0069] Under the same test conditions (clear weather, unobstructed, and windless), the beacon recognition accuracy of this control example was 85%, the positioning accuracy was 0.5 meters, and the landing accuracy was 1.0 meter. In low-light conditions at night, the beacon recognition accuracy dropped below 30%, making landing almost impossible. In light rain, due to lens fogging and image blurring, the beacon recognition accuracy was almost zero.

[0070] Comparative Example 2: This comparative example uses a GPS and IMU fusion positioning method. It uses a high-precision GPS module and IMU, performs data fusion through an extended Kalman filter algorithm, and then uses a PID control algorithm to control the drone landing.

[0071] Under the same test conditions (open field, unobstructed, good GPS signal), the positioning accuracy of this control example was 1.5 meters, and the landing accuracy was 2.5 meters. In the presence of electromagnetic interference, the GPS signal may be lost, causing the drone to lose control. In remote areas such as at sea, the GPS signal is unstable, and the positioning accuracy drops to 5-10 meters, making precise landing impossible.

[0072] Comparative Example 3: This comparative example uses a lidar positioning method, employing a 32-line lidar to scan the ground, achieving positioning through a point cloud matching algorithm, and then using a PID control algorithm to control the drone's landing.

[0073] Under the same test conditions, the positioning accuracy of this comparative example was 0.1 meters, and the landing accuracy was 0.3 meters. In light rain, due to the scattering of the laser beam by water droplets, the quality of the point cloud data deteriorated, and the positioning accuracy decreased to 0.5 meters, while the landing accuracy decreased to 1.0 meter. In heavy rain, the lidar was almost inoperable. Furthermore, the lidar equipment was very expensive, costing approximately 8,000 yuan, which is more than three times the total cost of the system of this invention.

[0074] Test Results and Analysis: To verify the performance of the present invention, comprehensive tests were conducted on Examples 1-3 and Comparative Examples 1-3 under different environmental conditions. The tests included beacon recognition accuracy, positioning accuracy, landing accuracy, nighttime operation capability, rain and fog operation capability, electromagnetic interference resistance, and cost. The test results are shown in the table below:

[0075] The test results show that Embodiments 1-3 of the present invention are significantly superior to Comparative Examples 1 and 2 in all performance indicators. The advantages of the present invention are particularly pronounced in extreme environments such as nighttime, rainy / foggy weather, and strong electromagnetic interference. Compared to Comparative Example 3, the present invention has higher positioning and landing accuracy, stronger performance in rainy / foggy weather, and lower cost.

[0076] In summary, this invention effectively solves the problems existing in the prior art by organically combining millimeter-wave radar and multispectral imaging fusion perception technology, phase difference high-precision ranging technology, federated Kalman filter hierarchical fusion positioning technology, and machine-platform cooperative control technology based on deep reinforcement learning. It achieves high-precision and high-reliability accurate landing of UAVs in extreme environments and has broad application prospects.

[0077] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for precise landing of a drone, characterized in that, Includes the following steps: S1: The landing area is scanned by a sensing module that combines millimeter-wave radar and multispectral imaging on the UAV to identify a pre-set phase-coded active light-emitting beacon array. S2: Collect UWB ranging data, IMU data and barometer data of the UAV, construct a hierarchical fusion positioning architecture, and use the federated Kalman filter algorithm to coarsely locate the position of the UAV and obtain the initial three-dimensional coordinates of the UAV. S3: When the UAV enters the effective communication range of the beacon, the phase difference ranging module is activated to calculate the precise distance between the UAV and each beacon, and the high-precision three-dimensional coordinates of the UAV relative to the beacon array are calculated by combining the geometric relationship of the beacon array. S4: Establish a collaborative dynamics model between the UAV and the landing platform. Based on high-precision three-dimensional coordinates and the preset optimal landing trajectory, use an adaptive landing control algorithm based on deep reinforcement learning to generate UAV control quantities. S5: Sends the UAV control signals to the flight controller and simultaneously sends attitude adjustment commands to the landing platform, controlling the UAV and the landing platform to work together to complete a precise landing.

2. The method for precise landing of a drone according to claim 1, characterized in that: The sensing module for fusing millimeter-wave radar and multispectral imaging described in step S1 specifically includes: acquiring point cloud data of the landing area using millimeter-wave radar to detect potential beacon targets; acquiring multispectral images of the beacon using the multispectral imaging module to extract the phase coding features of the beacon; performing spatiotemporal registration of the radar point cloud data and the multispectral image, fusing the feature information of the two modes to achieve accurate beacon identification.

3. The method for precise landing of a drone according to claim 1, characterized in that: The layered fusion positioning architecture described in step S2 specifically includes: the first layer is a UWB sub-filter, which processes the ranging data between the UWB base station and the UAV; the second layer is an IMU-barometer sub-filter, which processes IMU and barometer data; and the third layer is the main filter, which fuses the outputs of the two sub-filters to obtain the final coarse positioning result.

4. The method for precise landing of a drone according to claim 1, characterized in that: The phase difference ranging method described in step S3 specifically includes: each beacon in the beacon array emits a continuous sine wave signal of a specific frequency; the UAV receives signals from multiple beacons and measures the phase difference between different beacon signals; and the precise distance between the UAV and each beacon is calculated based on the phase difference and signal wavelength.

5. The method for precise landing of a drone according to claim 1, characterized in that: The adaptive landing control algorithm based on deep reinforcement learning described in step S4 specifically includes: constructing a Markov decision process that includes environmental state, UAV state, and control actions; using landing accuracy, landing time, and energy consumption as reward functions; training the control policy network using a deep deterministic policy gradient algorithm; and dynamically adjusting the control policy according to real-time environmental disturbances during the actual landing process.

6. The method for precise landing of a drone according to claim 1, characterized in that: The active attitude adjustment of the landing platform described in step S5 specifically includes: the landing platform is equipped with a three-axis attitude adjustment mechanism and a tilt sensor; the angle that the landing platform needs to adjust is calculated based on the real-time position and attitude information of the UAV; and the three-axis attitude adjustment mechanism is controlled to adjust the attitude of the platform so that the platform always remains perpendicular to the landing direction of the UAV.

7. A precision landing system for unmanned aerial vehicles (UAVs), characterized in that, include: The multimodal sensing module is used to identify a pre-defined phase-coded active luminous beacon array by fusing millimeter-wave radar with multispectral imaging; The coarse positioning module is used to collect UWB ranging data, IMU data and barometer data, and to perform coarse positioning of the UAV using a federated Kalman filter algorithm; The precise positioning module is used to calculate the precise distance between the UAV and the beacon using the phase difference ranging method, and to solve the high-precision three-dimensional coordinates of the UAV. The cooperative control module is used to establish a cooperative dynamics model between the UAV and the landing platform, and uses an adaptive landing control algorithm based on deep reinforcement learning to generate control quantities. The flight execution module is used to receive control signals from the UAV and control the UAV's flight. The platform execution module is used to receive attitude adjustment commands and control the landing platform to adjust its attitude.

8. A precision landing system for unmanned aerial vehicles according to claim 7, characterized in that: The phase-coded active light-emitting beacon array consists of four beacons arranged in a square. Each beacon integrates an infrared LED transmitter, a radio frequency signal transmitter, and a microcontroller. Each beacon emits a unique phase-coded signal to distinguish different beacons.

9. A precision landing system for unmanned aerial vehicles according to claim 7, characterized in that: The precise positioning module includes: A signal receiving unit is used to receive radio frequency signals transmitted by multiple beacons; The phase difference measurement unit is used to measure the phase difference between different beacon signals; The distance calculation unit is used to calculate the precise distance between the UAV and each beacon based on the phase difference and signal wavelength. The coordinate calculation unit is used to calculate the high-precision three-dimensional coordinates of the UAV using the least squares method based on the known positions and measured distances of multiple beacons.

10. A precision landing system for unmanned aerial vehicles according to claim 7, characterized in that: The collaborative control module includes: The model building unit is used to build a collaborative dynamics model of the UAV and the landing platform; The policy training unit is used to train the control policy network using a deep deterministic policy gradient algorithm. The control generation unit is used to generate UAV control quantities and platform attitude adjustment commands based on real-time status information. The disturbance compensation unit is used to estimate environmental disturbances in real time and compensate for control variables.

Images