Crossed dual-rotor unmanned aerial vehicle relative motion compensation landing control system and method

Abstract

The present application belongs to the technical field of unmanned aerial vehicle control, and particularly relates to a cross dual-rotor unmanned aerial vehicle relative motion compensation landing control system and method, comprising the following steps: acquiring unmanned aerial vehicle body state data and target platform multi-source sensing data; performing space-time alignment and fusion filtering on the multi-source sensing data to construct a platform state vector; constructing a relative pose state space model through coordinate system conversion to output relative pose measured values; outputting relative pose prediction values in a future time window and landing window evaluation results; generating multi-dimensional compensation control quantities based on state errors and landing window evaluation results; and mapping the multi-dimensional compensation control quantities into rotor motor control instructions to drive the cross dual-rotor actuator to complete landing actions. The present application can solve the problems of dynamic platform relative pose estimation error accumulation and lack of motion feedforward compensation leading to landing instability in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) control technology, specifically relating to a relative motion compensation landing control system and method for a cross-rotor UAV. Background Technology

[0002] With the widespread application of drones in logistics, power line inspection, marine monitoring, emergency rescue, and low-altitude economy, autonomous landing technology has become a crucial technology affecting flight safety and mission efficiency. Cross-rotor drones, due to their high lift efficiency, strong wind resistance, good hovering stability, and large payload capacity, have broad application prospects in complex environment missions. However, in practical applications, drones often need to complete autonomous landing tasks on moving platforms, non-fixed target platforms, or in complex environments.

[0003] Problems with existing technology: However, in scenarios where the target platform is in dynamic motion or subject to complex environmental disturbances, existing technologies lack the ability to dynamically adjust the covariance of multi-source sensor data and predict the relative motion time sequence, resulting in the accumulation of relative pose estimation errors over time. At the same time, the fixed-gain control law cannot perform feedforward compensation based on the relative motion trend, which can easily lead to trajectory deviation and attitude instability, ultimately resulting in landing failure. Summary of the Invention

[0004] The purpose of this invention is to provide a relative motion compensation landing control system and method for a cross-rotor unmanned aerial vehicle, which can solve the problems of dynamic platform relative pose estimation error accumulation and landing instability caused by lack of motion feedforward compensation in the prior art.

[0005] The specific technical solution adopted by this invention is as follows: A landing control method for relative motion compensation of a cross-rotor unmanned aerial vehicle (UAV) includes the following steps: Acquire drone body status data and multi-source sensor data from the target platform; Spatiotemporal alignment and fusion filtering are performed on multi-source sensor data to construct a platform state vector; Based on the ontology state data and platform state vector, a relative pose state space model is constructed through coordinate system transformation, and the measured values ​​of relative pose are output. The measured relative pose values ​​are input into the time series prediction model, and combined with the historical relative motion sequence, the predicted relative pose values ​​within the future time window and the landing window evaluation results are output. Calculate the state error between the measured and predicted relative pose values, and generate multidimensional compensation control variables based on the state error and landing window evaluation results. The multidimensional compensation control quantity is mapped to the control command of the rotor motor, which drives the cross-rotor actuator to complete the landing action.

[0006] According to another aspect of the present invention, spatiotemporal alignment and fusion filtering are performed on the multi-source sensor data to construct a platform state vector, including: Platform contour feature points are extracted from visual recognition data and combined with laser ranging data and millimeter-wave radar data to construct a multi-source observation vector; If visual recognition data is missing or the confidence level is lower than a preset threshold, dead reckoning is performed on the millimeter-wave radar data and the inertial measurement data in the UAV's body state data to generate a platform state prediction vector. An extended Kalman filter is used to update the state of the multi-source observation vector or the platform state prediction vector, and the output is a platform state vector containing the platform position, platform velocity, platform attitude, platform boundary and platform size.

[0007] According to another aspect of the present invention, based on the ontology state data and the platform state vector, a relative pose state space model is constructed through coordinate system transformation, and the measured relative pose values ​​are output, including: Construct the homogeneous transformation matrix from the UAV body coordinate system to the target platform coordinate system; Obtain vibration state information from the UAV's body state data, and dynamically calculate the observation noise covariance matrix based on the amplitude and frequency characteristics of the vibration state information; The homogeneous transformation matrix and the observation noise covariance matrix are input into the unscented Kalman filter algorithm to perform nonlinear state estimation of relative position, relative attitude and relative velocity, and output the measured values ​​of relative pose.

[0008] According to another aspect of the present invention, the measured relative pose value is input into a time-series prediction model, and combined with historical relative motion sequences, the predicted relative pose value within a future time window is output, including: Construct a historical relative motion sequence with a preset time step, the sequence including historical relative position, historical relative velocity and historical relative attitude; The measured relative pose values ​​are concatenated with historical relative motion sequences and then input into a temporal Transformer network that includes multi-head attention. The temporal-space dependent features are extracted by a temporal Transformer network and mapped through a fully connected layer to output the relative pose prediction values ​​for the next N control cycles, where N is an integer greater than 1.

[0009] According to another aspect of the present invention, the method for maintaining the historical relative motion sequence includes: A fixed-length sliding window is used to cache historical relative motion data; After each control cycle, the latest relative pose measurement value is appended to the end of the sliding window, and the oldest data at the beginning of the sliding window is removed. When a sudden acceleration change in historical relative motion data is detected due to a violent maneuver by the drone exceeding a preset threshold, the sliding window is cleared and data is reacquired.

[0010] According to another aspect of the present invention, generating the landing window assessment result and generating a multidimensional compensation control quantity based on the state error and the landing window assessment result includes: Based on the platform boundary and platform size in the platform state vector, a three-dimensional safe landing envelope is constructed. Calculate the minimum Euclidean distance between the predicted landing trajectory corresponding to the relative pose prediction value and the three-dimensional safe landing envelope; When the minimum Euclidean distance is greater than the preset safety threshold, and the angle between the terminal velocity vector of the predicted landing trajectory and the normal vector of the target platform is less than the preset angle threshold, a landing window evaluation result containing the landable time window and the optimal landing time is generated. Construct a cost function for model predictive control, which includes a position error weight term, a velocity error weight term, an attitude error weight term, and a control increment penalty term; Under the constraint of the landing time window, the optimal control sequence is solved by a quadratic programming algorithm with the goal of minimizing the cost function. The control quantity of the first control cycle in the optimal control sequence is extracted as a multi-dimensional compensation control quantity that includes position compensation, velocity compensation, and attitude compensation.

[0011] According to another aspect of the present invention, mapping the multidimensional compensation control quantity to control commands for a rotor motor includes: Calculate the expected total thrust and expected triaxial torque based on multidimensional compensation control variables; Obtain the rotor spacing and rotor crossing angle of the intersecting rotors, and construct a thrust distribution matrix that includes the aerodynamic interference coefficient between the rotors; The desired total thrust and desired triaxial torque are multiplied by the pseudo-inverse of the thrust distribution matrix to generate the desired speed command for each rotor motor. After the desired speed command is limited and smoothed, it is output to the motor drive unit.

[0012] According to another aspect of the present invention, during the process of driving the cross-rotor actuator to complete the landing action, the method further includes: Real-time calculation of the relative altitude and vertical descent rate between the drone and the target platform; When the relative height is less than the preset terminal threshold, the proportional gain and derivative gain of the attitude control loop are dynamically increased based on the ratio of the vertical descent rate to the relative height in order to suppress attitude divergence at the moment of landing.

[0013] According to another aspect of the present invention, the process of driving the cross-rotor actuator to complete the landing action further includes: Real-time monitoring of the actual speed and current feedback data of each rotor motor; If the deviation between the actual speed and the desired speed command exceeds the preset stall threshold, or if the current feedback data triggers overcurrent protection, an execution abnormal signal is generated. In response to the execution of abnormal signaling, the landing control flow is immediately cut off, and the go-around control is triggered to generate a maximum lift command to drive the UAV away from the target platform.

[0014] According to another aspect of the present invention, a relative motion compensation landing control system for a cross-rotor unmanned aerial vehicle is also provided, comprising: The data acquisition unit is used to acquire UAV body status data and multi-source sensor data of the target platform; The state fusion unit is used to perform spatiotemporal alignment and fusion filtering on multi-source sensor data to construct the platform state vector; The pose estimation unit is used to construct a relative pose state space model based on the ontology state data and platform state vector through coordinate system transformation, and output the measured value of relative pose. The motion prediction unit is used to input the measured relative pose values ​​into the time series prediction model, combine them with the historical relative motion sequence, and output the relative pose prediction values ​​within the future time window and the landing window evaluation results. The compensation calculation unit is used to calculate the state error between the measured relative pose value and the predicted relative pose value, and to generate multi-dimensional compensation control quantity based on the state error and the landing window evaluation result. The execution control unit is used to map multidimensional compensation control quantities into control commands for the rotor motors, driving the cross-rotor actuators to complete the landing action.

[0015] According to another aspect of the present invention, an electronic device is also provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the cross-rotor unmanned aerial vehicle relative motion compensation landing control method described in any of the preceding embodiments.

[0016] According to another aspect of the present invention, a computer-readable storage medium is also provided, storing a computer program that, when executed by a processor, implements the cross-rotor unmanned aerial vehicle relative motion compensation landing control method described above.

[0017] The technical effects achieved by this invention are as follows: This invention employs spatiotemporal alignment and fusion filtering of multi-source sensor data, combined with the construction of a relative pose state space model based on the observation noise covariance matrix dynamically calculated from vibration amplitude and frequency characteristics. This reduces the variance of the relative pose estimation error under high-frequency vibration conditions of a cross-rotor system, thereby achieving high-precision relative state perception in complex motion platform environments. Simultaneously, by introducing a temporal Transformer network to predict future relative motion trends and generate multi-dimensional compensation control quantities, the control commands possess feedforward compensation capabilities, thereby eliminating trajectory lag and deviation caused by the relative motion of the dynamic platform.

[0018] This invention constructs a model predictive control cost function that includes position, velocity, attitude error weights, and control increment penalty terms. It then combines this function with a thrust allocation matrix that includes aerodynamic interference coefficients between rotors for pseudo-inverse calculation. This allows the aerodynamic coupling interference between the upper and lower rotors of a cross-rotor system to be explicitly canceled during the control allocation phase, thereby improving the attitude response accuracy under multi-rotor cooperative control. Furthermore, by dynamically increasing the proportional and differential gain based on the ratio of vertical descent rate to relative altitude during the terminal landing phase, the attitude divergence moment at the moment of touchdown is rapidly suppressed, thus ensuring attitude stability during the landing and touchdown process.

[0019] This invention, through the design of a millimeter-wave radar dead reckoning degradation strategy in the event of visual failure and a model degradation and lookup table interpolation mechanism under high load, enables the system to maintain the hard real-time performance of the basic control loop even when sensor data is lost or edge computing resources are limited, thereby avoiding control divergence caused by computing power blockage or perception interruption. At the same time, by establishing a physical thrust threshold cutoff and thrust deficit redistribution mechanism for the motor, the remaining thrust when a single rotor is saturated is proportionally transferred to the unsaturated rotor and the anti-torque is corrected synchronously, thereby preventing overall attitude instability caused by local actuator overload. Attached Figure Description

[0020] Figure 1 This is a block diagram of the overall system structure of the present invention; Figure 2 This is a flowchart of the target platform perception module in this invention; Figure 3 This is a schematic diagram of the working principle of the relative pose estimation module in this invention; Figure 4 This is a flowchart of the relative motion prediction process in this invention; Figure 5This is a schematic diagram of the flight status perception structure of the cross-rotor UAV in this invention; Figure 6 This is a schematic diagram of the working principle of the relative motion compensation module in this invention; Figure 7 This is a schematic diagram of the relative motion compensation between the UAV and the target platform in this invention; Figure 8 This is a schematic diagram of the relative motion compensation landing control principle in this invention. Detailed Implementation

[0021] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0022] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] According to an embodiment of the present invention, a method embodiment of a relative motion compensation landing control method for a cross-rotor unmanned aerial vehicle is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0024] In one optional embodiment, a relative motion compensation landing control system (hereinafter referred to as the system) for a cross-rotor unmanned aerial vehicle is provided, which is deployed on the airborne flight control computing platform of the cross-rotor unmanned aerial vehicle.

[0025] As a preferred approach, the flight control computing platform adopts a heterogeneous multi-core architecture, including an ARM microprocessor as the main controller and an FPGA (Field Programmable Gate Array) chip as a coprocessor.

[0026] Further, refer to the appendix Figure 5The sensor array employs a distributed spatial layout. A GNSS antenna is deployed on the top of the UAV to obtain global absolute position; an IMU (Inertial Measurement Unit) is deployed at the center of the fuselage to acquire high-frequency rigid body motion data; vibration sensors are deployed throughout the fuselage to capture the aerodynamic coupling high-frequency flutter unique to the crossed rotor system; a downward-looking laser rangefinder, a forward-looking vision camera, and a millimeter-wave radar are deployed at the bottom and front of the fuselage to construct a local environmental perception field; a barometric altimeter is deployed at the bottom, and rotor speed sensors are deployed on the front and rear beams. Data from these multi-source heterogeneous sensors is aggregated to the flight control computing platform via a CAN-FD bus and a high-speed SPI interface.

[0027] Furthermore, the high-frequency raw data acquired by the sensor is directly written into the FPGA's on-chip BRAM (Block RAM) via the DMA (Direct Memory Access) controller, bypassing the CPU cache and thus ensuring microsecond-level data throughput latency. The FPGA is responsible for performing low-level signal filtering and Fast Fourier Transform (FFT), and transmits the processed feature vectors to the ARM host's SRAM via the AXI bus for use by the upper-level timing prediction network and control law solving module.

[0028] like Figure 1 As shown, as an optional embodiment, the overall system structure is divided into a perception and estimation layer, a prediction and compensation layer, and a control and execution layer.

[0029] The perception and estimation layer includes: The flight status perception module includes an IMU inertial measurement unit, a GNSS positioning unit, a barometric altimeter, a flight status detection unit, and a vibration status detection unit, which are used to output the UAV's own status data. The target platform perception module includes a visual recognition unit, a laser ranging unit, a millimeter-wave radar unit, and a platform status detection unit, which is used to output target platform status information.

[0030] The prediction and compensation layer includes: The relative pose estimation module includes a coordinate transformation unit, a relative position calculation unit, a relative attitude calculation unit, and an error estimation unit. The relative motion prediction module includes a historical state database, a relative motion analysis unit, a motion trend prediction unit, and a landing window evaluation unit. The relative motion compensation module includes a position compensation unit, a velocity compensation unit, an attitude compensation unit, and a trajectory correction unit.

[0031] The control and execution layer includes: The dynamic landing control module includes a thrust distribution unit, an attitude control unit, a trajectory tracking unit, and an adaptive gain adjustment unit. The flight execution module includes a motor drive unit, a rotor actuator, and a flight control output unit.

[0032] The system data flow is as follows: after the external environment and flight status inputs are processed by the perception and estimation layer, the relative pose measured values ​​and predicted values ​​are generated. These values ​​are then input to the prediction and compensation layer for error calculation and compensation command generation. Finally, the control and execution layer outputs motor control commands to complete the autonomous landing mission of the cross-rotor UAV.

[0033] As an optional embodiment, refer to the appendix Figure 2 When the system performs target platform perception, in response to acquiring multi-source sensor data of the target platform, it initiates a multi-source fusion filtering process. The multi-source sensor data includes visual image frames acquired at a sampling rate of 30Hz, laser point cloud data, and target track point data from millimeter-wave radar.

[0034] For the aforementioned multi-source sensor data, when performing spatiotemporal alignment and fusion filtering, the system extracts platform contour feature points from the visual recognition data and combines them with laser ranging data and millimeter-wave radar data to construct a multi-source observation vector.

[0035] Furthermore, if visual recognition data is detected to be missing (e.g., feature point loss due to sudden changes in illumination) or the confidence level is lower than a preset threshold, a degraded perception strategy is triggered. At this point, the system pauses visual feature extraction, directly calls the target trajectory data output by the millimeter-wave radar, and combines it with inertial measurement data collected by the IMU to perform dead reckoning and generate a platform state prediction vector. Subsequently, an extended Kalman filter (EKF) is used to update the state of the multi-source observation vector or the platform state prediction vector, outputting a vector containing the platform's position. Platform speed Platform posture Platform Boundaries Platform dimensions The platform state vector.

[0036] As an optional embodiment, refer to the appendix Figure 3 The relative pose estimation module is used to establish the UAV coordinate system. With platform coordinate system The relative pose relationship between them.

[0037] When performing relative pose estimation, the system constructs a homogeneous transformation matrix from the UAV body coordinate system to the target platform coordinate system.

[0038] in, For relative rotation matrices, This is a relative position vector.

[0039] Furthermore, considering that the cross-rotor UAV inevitably generates high-frequency micro-vibrations during hovering and descent, these vibrations directly couple to the IMU and visual sensors, causing the observation noise to exhibit non-Gaussian characteristics. The error estimation unit acquires vibration state information from the UAV's body state data (obtaining acceleration time-domain signals through vibration sensors mounted on the fuselage, and extracting the dominant frequency amplitude through Fast Fourier Transform), and then uses the amplitude of the vibration state information... With frequency characteristics Dynamically calculate the observation noise covariance matrix :

[0040] in, The reference noise covariance matrix is... The vibration coupling coefficient, as defined in the formula, means that vibration energy is proportional to the square of the amplitude. Lower frequencies have a stronger cumulative effect on sensor integral drift, thus requiring the covariance matrix to increase nonlinearly. To prevent matrix singularity, this is typically achieved by... Add a minimum value to the diagonal elements (like The homogeneous transformation matrix and the dynamically adjusted observation noise covariance matrix are then used. The input is fed into the Unscented Kalman Filter (UKF) algorithm to perform nonlinear state estimation of relative position, relative attitude and relative velocity, and outputs the measured value of relative pose.

[0041] As an optional embodiment, refer to the appendix Figure 4 The relative motion prediction module predicts future relative motion based on historical data.

[0042] In one alternative embodiment, the historical state database maintains a fixed-length sliding window (e.g., 50) to cache historical relative motion data, including position. ,speed ,attitude At the end of each control cycle, the latest measured relative pose value is appended to the tail of the sliding window, and the oldest data at the head is removed. When a sudden acceleration change in historical relative motion data is detected due to a violent maneuver by the UAV exceeding a preset threshold, the sliding window is cleared and data is reacquired to prevent historical abrupt changes from interfering with future trend predictions.

[0043] The relative motion analysis unit concatenates the measured relative pose values ​​with historical relative motion sequences and inputs them into a temporal Transformer network incorporating multi-head attention. The motion trend prediction unit extracts spatiotemporal dependency features through this network, maps them via fully connected layers, and outputs predicted relative pose values ​​for the next N control cycles. .

[0044] The specific formula for calculating the multi-head self-attention mechanism is as follows:

[0045] in, These represent the query matrix, key matrix, and value matrix, respectively, obtained by linearly mapping relative pose measurements to historical sequences. The dimension of the key vector is used as the denominator of the scaling factor to prevent the dot product result from being too large, which would cause the Softmax function to enter the gradient saturation region. When the model is pruned and deployed on an airborne embedded platform, its empirical value is usually 64 or 32.

[0046] The landing window evaluation unit constructs a three-dimensional safe landing envelope based on the platform boundary and platform dimensions in the platform state vector. It calculates the minimum Euclidean distance between the predicted landing trajectory corresponding to the relative pose prediction value and the three-dimensional safe landing envelope. When the minimum Euclidean distance is greater than a preset safety threshold, and the angle between the terminal velocity vector of the predicted landing trajectory and the normal vector of the target platform is less than a preset angle threshold, a landing window evaluation result containing a landable time window and the optimal landing time is generated.

[0047] As an optional embodiment, refer to the appendix Figure 6 The relative motion compensation module generates compensation control quantities based on the prediction results.

[0048] When performing relative motion compensation, the system receives the measured values ​​of relative pose (position). ,speed ,attitude ) and relative motion prediction (position) ,speed ,attitude ), calculate the relative error: .

[0049] Furthermore, the compensation instruction generation unit constructs the cost function of model predictive control (MPC). :

[0050] in, , , These are the position, velocity, and attitude error weight matrices, respectively. To control the incremental penalty matrix, To control the increment, under the constraint of the landable time window, the goal is to minimize the cost function. To optimize the objective, a quadratic programming (QP) algorithm is used to solve for the optimal control sequence. The control input for the first control cycle in the optimal control sequence is extracted to generate the trajectory correction. Thrust correction Attitude correction amount The final output includes a multi-dimensional compensation control quantity that includes position compensation, velocity compensation, and attitude compensation.

[0051] like Figure 7 As shown, through the above relative motion compensation, the UAV can accurately align with the platform's actual trajectory along the compensated desired trajectory, eliminating the landing point deviation caused by the uncompensated actual trajectory and achieving precise landing.

[0052] As an optional embodiment, refer to the appendix Figure 8 The dynamic landing control module receives compensation commands to achieve autonomous landing control closed loop.

[0053] In one optional embodiment, the thrust distribution unit calculates the desired total thrust and desired triaxial torque based on multidimensional compensation control variables. Considering the aerodynamic interference between the upper and lower rotors of the crossed dual rotors, the rotor spacing and rotor crossing angle of the crossed dual rotors are obtained, and a system incorporating the inter-rotor aerodynamic interference coefficient is constructed. Thrust distribution matrix :

[0054] in, For rotor arm length, The rotor crossing angle is calculated by multiplying the desired total thrust and desired triaxial torque by the pseudo-inverse of the thrust distribution matrix. The desired speed commands for each rotor motor are generated. After being limited and smoothed by filtering, the desired speed commands are output to the motor drive unit.

[0055] Furthermore, during the landing maneuver driven by the cross-rotor actuator, the relative altitude and vertical descent rate between the UAV and the target platform are calculated in real time. When the relative altitude is less than a preset terminal threshold (e.g., 0.5 meters), the proportional gain and derivative gain of the attitude control loop are dynamically increased based on the ratio of the vertical descent rate to the relative altitude to suppress attitude divergence at the moment of landing.

[0056] Given the complexity of the actual flight environment, the system must have a robust anomaly handling mechanism.

[0057] In one optional embodiment, for motor saturation anomalies: if any rotor thrust value in the thrust distribution control command exceeds the motor's physical maximum thrust threshold (e.g., PWM duty cycle reaches 95% or higher), thrust reallocation is triggered. The rotor thrust value exceeding the threshold is truncated to the physical maximum thrust threshold, and the thrust deficit is calculated; based on the lever arm ratio of the cross-rotor, the thrust deficit is proportionally allocated to the other unsaturated rotor, and the anti-torque compensation coefficient of the other rotor is adjusted synchronously, updating the thrust distribution control command.

[0058] Furthermore, for sensor data packet loss or timeout degradation handling: when continuous packet loss is detected from multiple sensor sources, and the packet loss duration exceeds a preset timeout threshold (e.g., 50 milliseconds), the inference of the time series prediction network is paused to save computing power. An extended Kalman filter is invoked, and based on the state vector and system state transition matrix of the last valid moment before packet loss, the relative pose at the current moment is recursively estimated; the gain parameter matrix of the flight controller is downgraded to a fixed parameter mode to maintain basic flight stability.

[0059] Furthermore, to address computing bottlenecks caused by limited edge computing resources or high loads: when the utilization rate of the flight control main CPU exceeds 85% for three consecutive control cycles, model degradation is triggered. The system automatically switches the Transformer model containing a multi-head self-attention mechanism to a gated recurrent unit (GRU) network with fewer parameters, or directly calls the "relative state-control gain" three-dimensional lookup table generated offline based on historical relative motion data for interpolation calculation, sacrificing a slight amount of prediction accuracy to ensure that the control loop meets the hard real-time requirements of over 100Hz.

[0060] As a preferred application scenario, this invention can be applied to the task of recovering unmanned aerial vehicles on the deck of a marine research vessel.

[0061] In one optional embodiment, the target platform is the deck of a research vessel that oscillates with the waves in six degrees of freedom. Millimeter-wave radar and a visual camera work together to overcome wave reflections and salt spray interference, acquiring real-time features of the deck edges. A relative motion prediction module predicts the periodic trends of the hull's roll and pitch in advance, and the MPC control module begins attitude feedforward compensation two seconds before the UAV touches the ground, offsetting the relative pose error caused by the deck's tilt. Even when the UAV encounters gusts causing severe hull vibrations, the dynamic observation noise covariance matrix effectively filters out vibration interference, ensuring the UAV lands accurately and smoothly within the safe envelope of the moving deck.

[0062] In an alternative embodiment, refer to the appendix Figure 1 A relative motion compensation landing control system for a cross-rotor unmanned aerial vehicle is also provided, including: The data acquisition unit is used to acquire UAV body status data and multi-source sensor data of the target platform; The state fusion unit is used to perform spatiotemporal alignment and fusion filtering on multi-source sensor data to construct the platform state vector; The pose estimation unit is used to construct a relative pose state space model based on the ontology state data and platform state vector through coordinate system transformation, and output the measured value of relative pose. The motion prediction unit is used to input the measured relative pose values ​​into the time series prediction model, combine them with the historical relative motion sequence, and output the relative pose prediction values ​​within the future time window and the landing window evaluation results. The compensation calculation unit is used to calculate the state error between the measured relative pose value and the predicted relative pose value, and to generate multi-dimensional compensation control quantity based on the state error and the landing window evaluation result. The execution control unit is used to map multidimensional compensation control quantities into control commands for the rotor motors, driving the cross-rotor actuators to complete the landing action.

[0063] Furthermore, the data flow and connection relationships between the above units are as follows: the output of the data acquisition unit is communicatively connected to the inputs of the state fusion unit and the pose estimation unit, respectively, to distribute the acquired raw data to downstream processing nodes; the output of the state fusion unit is connected to the input of the pose estimation unit, to transmit the constructed platform state vector to the pose estimation unit; the output of the pose estimation unit is connected to the inputs of the motion prediction unit and the compensation calculation unit, respectively, to synchronously send the relative pose measurement values ​​to the prediction and compensation nodes; the output of the motion prediction unit is connected to the input of the compensation calculation unit, to input the relative pose prediction values ​​and landing window evaluation results to the compensation calculation unit; the output of the compensation calculation unit is connected to the input of the execution control unit, to send the generated multidimensional compensation control quantity to the execution control unit, and finally the execution control unit outputs control commands to the motor drive unit.

[0064] This application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the cross-rotor UAV relative motion compensation landing control method of any of the foregoing embodiments.

[0065] This application also provides a computer-readable storage medium storing computer instructions for causing a computer to execute the relative motion compensation landing control method for any of the foregoing embodiments of the cross-rotor UAV.

[0066] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0067] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A landing control method for relative motion compensation of a cross-rotor unmanned aerial vehicle, characterized in that, Includes the following steps: Acquire drone body status data and multi-source sensor data from the target platform; Spatiotemporal alignment and fusion filtering are performed on multi-source sensor data to construct a platform state vector; Based on the ontology state data and platform state vector, a relative pose state space model is constructed through coordinate system transformation, and the measured values ​​of relative pose are output. The measured relative pose values ​​are input into the time series prediction model, and combined with the historical relative motion sequence, the predicted relative pose values ​​within the future time window and the landing window evaluation results are output. Calculate the state error between the measured and predicted relative pose values, and generate multidimensional compensation control variables based on the state error and landing window evaluation results. The multidimensional compensation control quantity is mapped to the control command of the rotor motor, which drives the cross-rotor actuator to complete the landing action.

2. The method for relative motion compensation landing control of a cross-rotor unmanned aerial vehicle according to claim 1, characterized in that, The multi-source sensor data is spatiotemporally aligned and fused, and filtered to construct a platform state vector, including: Platform contour feature points are extracted from visual recognition data and combined with laser ranging data and millimeter-wave radar data to construct a multi-source observation vector; If visual recognition data is missing or the confidence level is lower than a preset threshold, dead reckoning is performed on the millimeter-wave radar data and the inertial measurement data in the UAV's body state data to generate a platform state prediction vector. An extended Kalman filter is used to update the state of the multi-source observation vector or the platform state prediction vector, and the output is a platform state vector containing the platform position, platform velocity, platform attitude, platform boundary and platform size.

3. The method for relative motion compensation landing control of a cross-rotor unmanned aerial vehicle according to claim 1, characterized in that, Based on the aforementioned ontology state data and platform state vectors, a relative pose state space model is constructed through coordinate system transformation, and the measured relative pose values ​​are output, including: Construct the homogeneous transformation matrix from the UAV body coordinate system to the target platform coordinate system; Obtain vibration state information from the UAV's body state data, and dynamically calculate the observation noise covariance matrix based on the amplitude and frequency characteristics of the vibration state information; The homogeneous transformation matrix and the observation noise covariance matrix are input into the unscented Kalman filter algorithm to perform nonlinear state estimation of relative position, relative attitude and relative velocity, and output the measured values ​​of relative pose.

4. The method for relative motion compensation landing control of a cross-rotor UAV according to claim 1, characterized in that, The measured relative pose values ​​are input into the time-series prediction model, and combined with historical relative motion sequences, the predicted relative pose values ​​within future time windows are output, including: Construct a historical relative motion sequence with a preset time step, the sequence including historical relative position, historical relative velocity and historical relative attitude; The measured relative pose values ​​are concatenated with historical relative motion sequences and then input into a temporal Transformer network that includes multi-head attention. The temporal-space dependent features are extracted by a temporal Transformer network and mapped through a fully connected layer to output the relative pose prediction values ​​for the next N control cycles, where N is an integer greater than 1.

5. The method for relative motion compensation landing control of a cross-rotor unmanned aerial vehicle according to claim 1, characterized in that, The methods for maintaining the historical relative motion sequence include: A fixed-length sliding window is used to cache historical relative motion data; After each control cycle, the latest relative pose measurement value is appended to the end of the sliding window, and the oldest data at the beginning of the sliding window is removed. When a sudden acceleration change in historical relative motion data is detected due to a violent maneuver by the drone exceeding a preset threshold, the sliding window is cleared and data is reacquired.

6. The relative motion compensation landing control method for a cross-rotor unmanned aerial vehicle according to claim 1, characterized in that, The landing window evaluation result is generated, and a multi-dimensional compensation control quantity is generated based on the state error and the landing window evaluation result, including: Based on the platform boundary and platform size in the platform state vector, a three-dimensional safe landing envelope is constructed. Calculate the minimum Euclidean distance between the predicted landing trajectory corresponding to the relative pose prediction value and the three-dimensional safe landing envelope; When the minimum Euclidean distance is greater than the preset safety threshold, and the angle between the terminal velocity vector of the predicted landing trajectory and the normal vector of the target platform is less than the preset angle threshold, a landing window evaluation result containing the landable time window and the optimal landing time is generated. Construct a cost function for model predictive control, which includes a position error weight term, a velocity error weight term, an attitude error weight term, and a control increment penalty term; Under the constraint of the landing time window, the optimal control sequence is solved by a quadratic programming algorithm with the goal of minimizing the cost function. The control quantity of the first control cycle in the optimal control sequence is extracted as a multi-dimensional compensation control quantity that includes position compensation, velocity compensation, and attitude compensation.

7. The method for relative motion compensation landing control of a cross-rotor unmanned aerial vehicle according to claim 1, characterized in that, Mapping the multidimensional compensation control quantity to control commands for the rotor motor includes: Calculate the expected total thrust and expected triaxial torque based on multidimensional compensation control variables; Obtain the rotor spacing and rotor crossing angle of the intersecting rotors, and construct a thrust distribution matrix that includes the aerodynamic interference coefficient between the rotors; The desired total thrust and desired triaxial torque are multiplied by the pseudo-inverse of the thrust distribution matrix to generate the desired speed command for each rotor motor. After the desired speed command is limited and smoothed, it is output to the motor drive unit.

8. The method for relative motion compensation landing control of a cross-rotor unmanned aerial vehicle according to claim 1, characterized in that, The process of driving the cross-rotor actuator to complete the landing action also includes: Real-time calculation of the relative altitude and vertical descent rate between the drone and the target platform; When the relative height is less than the preset terminal threshold, the proportional gain and derivative gain of the attitude control loop are dynamically increased based on the ratio of the vertical descent rate to the relative height in order to suppress attitude divergence at the moment of landing.

9. The method for relative motion compensation landing control of a cross-rotor unmanned aerial vehicle according to claim 1, characterized in that, The process of driving the cross-rotor actuator to complete the landing action also includes: Real-time monitoring of the actual speed and current feedback data of each rotor motor; If the deviation between the actual speed and the desired speed command exceeds the preset stall threshold, or if the current feedback data triggers overcurrent protection, an execution abnormal signal is generated. In response to the execution of abnormal signaling, the landing control flow is immediately cut off, and the go-around control is triggered to generate a maximum lift command to drive the UAV away from the target platform.

10. A relative motion compensation landing control system for a cross-rotor unmanned aerial vehicle, operating the method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition unit is used to acquire UAV body status data and multi-source sensor data of the target platform; The state fusion unit is used to perform spatiotemporal alignment and fusion filtering on multi-source sensor data to construct the platform state vector; The pose estimation unit is used to construct a relative pose state space model based on the ontology state data and platform state vector through coordinate system transformation, and output the measured value of relative pose. The motion prediction unit is used to input the measured relative pose values ​​into the time series prediction model, combine them with the historical relative motion sequence, and output the relative pose prediction values ​​within the future time window and the landing window evaluation results. The compensation calculation unit is used to calculate the state error between the measured relative pose value and the predicted relative pose value, and to generate multi-dimensional compensation control quantity based on the state error and the landing window evaluation result. The execution control unit is used to map multidimensional compensation control quantities into control commands for the rotor motors, driving the cross-rotor actuators to complete the landing action.

Images