High-speed collision avoidance and falling control method for vector rotor unmanned aerial vehicle based on bidirectional motor

Abstract

This invention relates to a high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor, belonging to the field of flight control technology. Addressing the problem in existing technologies where severe turbulence in the airflow around the aircraft occurs at the moment the rotor switches from forward to reverse rotation, compounded by the influence of random environmental wind fields, easily leading to attitude oscillations and trajectory deviations in the UAV, this application utilizes a wind speed estimation method and a linear quadratic controller to accurately capture real-time wind field changes. This data is then used to compensate for the severe turbulence in the airflow around the aircraft and the combined effects of random environmental wind fields, thereby resolving the issues of attitude oscillations and trajectory deviations in the UAV. This application can significantly improve attitude stability in complex wind environments, greatly enhancing mission accuracy and reliability.

Description

Technical Field

[0001] This application relates to the field of flight control technology, specifically a high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor. Background Technology

[0002] With the inclusion of the national low-altitude economy in the future industrial cluster development plan, drones, as the core carrier of the low-altitude economy, have rapidly penetrated from their initial status as consumer entertainment tools into various industries such as logistics and distribution, emergency rescue, infrastructure inspection, and agricultural plant protection, becoming key equipment for activating new productivity and promoting the transformation and upgrading of traditional industries. Currently, the demand for civilian drones continues to upgrade. Whether it's material transportation in remote mountainous areas, safety inspections of high-altitude buildings, or emergency response at disaster sites, higher demands are being placed on the flight performance, safety, reliability, and scenario adaptability of drones. The technological bottlenecks of traditional multi-rotor drones are gradually becoming apparent, making it difficult to meet the needs of efficient and safe operations in complex civilian scenarios.

[0003] Traditional quadcopter drones primarily rely on rotor speed changes for lift adjustment, and their descent modes are mostly passive deceleration or slow hovering descent, exhibiting significant performance limitations. However, vector quadcopter drones, with their ability to instantly change lift direction by reversing rotor rotation, possess a core advantage in rapidly switching from a stable high-altitude cruise state to a reverse descent mode. However, due to airflow interference and dynamic instability issues, the airflow around the drone experiences severe turbulence at the instant the rotor switches from clockwise to counterclockwise rotation. This, combined with the influence of random environmental wind fields, can easily lead to attitude oscillations and trajectory deviations in the drone. Summary of the Invention

[0004] The purpose of this invention is to address the problem in existing technologies where, at the moment the rotor switches from forward to reverse rotation, the airflow around the aircraft becomes severely turbulent, and this, combined with the influence of random wind fields in the environment, easily leads to attitude oscillations and trajectory deviations in the UAV. This invention provides a high-speed collision avoidance and landing control method for vector rotor UAVs based on bidirectional motors.

[0005] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0006] A high-speed collision avoidance descent control method for vector rotor UAVs based on bidirectional motors includes the following steps:

[0007] Step 1: Obtain the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter is rotating forward, and the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter is rotating in reverse, so as to construct a dynamic model of the vector rotor drone;

[0008] Step 2: Based on the dynamic model of the vector rotor UAV and combined with the interference observer, obtain the interference force of wind on the quadcopter UAV;

[0009] Step 3: Utilize the wind's interference force on the quadcopter drone to obtain the compensation correction amount at the moment of rotor switching. ;

[0010] Step 4: Utilize The observer is optimized, and the optimized observer is used to obtain the force generated by the wind acting on the UAV. and torque ;

[0011] Step 5: Acquire an environmental image and input the image into the neural network to obtain the virtual tension output by the neural network;

[0012] Step 6: Construct the system state equations and base them on the forces generated by the wind acting on the UAV. Torque And virtual thrust, to obtain the desired servo command when the rotor reverses. and and the expected rotational speed of the propeller. This is used to control the propeller. and These represent the desired steering angles for the inner and outer servos, respectively.

[0013] Furthermore, the specific process of obtaining the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter UAV rotates forward in step 1 is as follows:

[0014] Step 11: Select the northeast reference coordinate system The inertial coordinate system is used as the initial coordinate system, and then the body coordinate system is established at the rotation centers of the four vector thrust units of the quadcopter UAV. ;

[0015] Step 12: In Establish a local coordinate system at the rotation center of the vector thrust unit. , The origin of the coordinate system is at the end of the boom and does not rotate with the tilting mechanism; it is a radial tilting coordinate system. With tangential tilt coordinate system As sub-coordinate systems, they correspond to the motion planes of the two-degree-of-freedom steering mechanism of the thrust unit:

[0016] Step 13: Define the first Coordinate system of vector thrust unit With body coordinate system The difference between them is the rotation angle around the Z-axis of the machine body. ,but Compared to rotation matrix for:

[0017] ,

[0018] in, , The standard rotation matrix function for rotation about a single Z-axis is expressed as:

[0019] ,

[0020] Will Rotate twice to obtain Rotation matrix for:

[0021] ,

[0022] , These are the standard rotation matrix functions for single-axis rotation about the X and Y axes of each vector element, respectively. for The internal servo angle corresponding to the number of vector thrust units for The external servo angle corresponding to the thrust vectoring unit:

[0023] ,

[0024] The roll angle is defined as the angle by which a vector rotor UAV rotates around its X-axis. The angle of rotation around the Y-axis of the aircraft is the pitch angle. The angle of rotation around the Z-axis of the fuselage is the yaw angle. Then the body coordinate system Relative to the reference coordinate system rotation matrix for:

[0025] ,

[0026] This is the standard rotation matrix function for single-axis rotation about the Z-axis of each vector element;

[0027] The lift generated by the propeller is:

[0028] ,

[0029] ,

[0030] in, For propeller thrust in the engine system The expression in For propeller thrust in The expression in This represents the thrust generated by the propeller, and is a non-negative scalar. The aerodynamic lift coefficient, This refers to the propeller rotation angle;

[0031] Define the direction of the propeller's counterclockwise rotation as +1 and the direction of its clockwise rotation as -1. The aerodynamic lift coefficient of the propeller is... The aerodynamic counter-torque coefficient of the propeller is Then the first Each motor generates thrust when it rotates forward. for:

[0032] ,

[0033] No. The counter torque generated when the motor rotates forward for:

[0034] ,

[0035] No. The motor generates thrust when it reverses. for:

[0036] ,

[0037] No. The counter torque generated when the motor reverses for:

[0038] ,

[0039] During the period when the propeller switches from forward to reverse rotation, the first Each motor generates thrust. for:

[0040] ,

[0041] No. The counter torque generated by each motor is:

[0042] ,

[0043] Combining the above equations, we obtain the first equation when the rotor rotates clockwise. The thrust generated by the propeller for:

[0044] ,

[0045] When the rotor rotates clockwise The torque generated by the propeller for:

[0046] ,

[0047] in, For the first When the propeller rotates clockwise, the direction of rotation is as follows: when looking at the propeller along the Z-axis of the vector thrust unit, the propeller rotates clockwise. The propeller rotates clockwise to take , This refers to the moment when the propeller's rotational speed drops to 0 when it switches from forward to reverse rotation. For the current moment, for The first derivative.

[0048] Furthermore, the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter UAV reverses direction in step 1 are expressed as follows:

[0049] When the rotor reverses The thrust generated by each propeller is:

[0050] ,

[0051] When the rotor reverses The torque generated by each propeller is:

[0052] .

[0053] Furthermore, the dynamic model of the vector rotor UAV is expressed as follows:

[0054] ,

[0055] ,

[0056] in, For the quality of drones, for The second derivative, The position of the UAV in the reference frame. For the inertial tensor of the drone, The angular velocity of the drone within the machine system. The interference force experienced by the UAV in the reference frame, The disturbance torque experienced by the UAV's on-board system. Let be the rotation matrix from the body coordinate system to the reference coordinate system. for The first derivative, The total lift force experienced by the drone, For the representation of gravity in three-dimensional force space in a reference coordinate system, , It is the gravitational constant. This refers to the total torque generated by the propeller on the drone.

[0057] Furthermore, the interference observer is represented as:

[0058] ,

[0059] ,

[0060] ,

[0061] in, To interfere with the observer, As an intermediate variable, For the observation matrix, for The observed values, The mass-inertia matrix, It is a diagonal matrix in which all elements are greater than 0.

[0062] Furthermore, the interference force of the wind on the quadcopter drone is expressed as follows:

[0063] ,

[0064] in, For the first Observer output value in each round It is a 3-order identity matrix. For the first Observer output value in each round Sampling time, for The third-order principal form, These are the observed values ​​of aerodynamic forces. To control the number of rounds.

[0065] Furthermore, the compensation correction amount of the rotor at the moment of switching. Represented as:

[0066] ,

[0067] in, for Time and The difference in acceleration at time t.

[0068] Furthermore, the specific steps of step 5 are as follows:

[0069] Acquire images of the environment in front of the drone and images of the environment in front of the drone. Convert to depth map Then, the depth map Pooling is performed, and the minimum value is taken among every 5*5 pixels to obtain a simplified depth map. Then the depth map All depth values ​​and thresholds Compare, when less than the threshold The depth value ratio reached If the value exceeds a certain threshold, it is determined that the target point or target point shelter has been reached, and no obstacle avoidance is performed. The depth value ratio reached If the obstacle is not present, then it is determined to be an obstacle-free situation, and no obstacle avoidance is performed; otherwise, obstacle avoidance is performed.

[0070] When avoiding obstacles, view the environment in front of the drone. Inputting data into a neural network yields the virtual tension output by the neural network. Virtual tension Represented as:

[0071] ,

[0072] in, For virtual force coefficients,

[0073] The neural network specifically performs the following steps:

[0074] Step 51: Image of the environment in front of the drone First, the image passes through three convolutional layers, then a fully connected layer receives the output of the convolutional layers to obtain the visual feature vector at the current time step. ;

[0075] Step 52: Obtain the visual feature vector at the current moment. The hidden layer state, along with the hidden layer state from the previous time step, is input into the gated recurrent unit to obtain the hidden layer state at the current time step. ;

[0076] Step 53: Obtain the visual feature vector at the current moment. and the hidden layer state at the current moment Input to the decision layer, and obtain the output vector. ,in, The direction vector of the virtual force is in Projection of the axis The direction vector of the virtual force is in Projection of the axis;

[0077] The loss function of the neural network is expressed as:

[0078] ,

[0079] ,

[0080] in, The pixel pointing from the image center of the depth map vector, for The depth value, The weighted depth centroid vector of the entire map, the weight function , This is the loss coefficient.

[0081] Furthermore, the system state equation is expressed as:

[0082] ,

[0083] ,

[0084] ,

[0085] ,

[0086] ,

[0087] ,

[0088] in, For feedback gain, Let be the solution matrix of the Riccati equation. For the state weight function, for The first derivative, For the deviation of the system, For system input, For the system matrix, For the input matrix, It is a state variable.

[0089] Furthermore, the desired command from the servo when the rotor reverses. and and the expected rotational speed of the propeller. Represented as:

[0090] ,

[0091] ,

[0092] ,

[0093] ,

[0094] ,

[0095] ,

[0096] in, For vectors antisymmetric matrix, for Origin and The distance between the origins.

[0097] The beneficial effects of this invention are:

[0098] This application employs a wind speed estimation method and a linear quadratic controller to accurately capture real-time wind field changes. This data is then used to compensate for the effects of severe turbulence in the airflow around the drone, compounded by the influence of random environmental wind fields, thereby resolving the issues of drone attitude oscillation and trajectory deviation. This application can significantly improve attitude stability in complex wind environments, greatly enhancing mission accuracy and reliability. Attached Figure Description

[0099] Figure 1 This is a schematic diagram of the body coordinate system and the vector thrust unit coordinate system;

[0100] Figure 2 A schematic diagram of wind speed estimation and wind resistance control for a vector quadcopter UAV;

[0101] Figure 3 Here is a flowchart for wind speed estimation;

[0102] Figure 4 This is a control flowchart for a vector rotor unmanned aerial vehicle. Detailed Implementation

[0103] It should be noted that, where there is no conflict, the various embodiments disclosed in this application can be combined with each other.

[0104] Specific Implementation Method 1: The high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor described in this implementation method includes the following steps:

[0105] Step 1: Obtain the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter is rotating forward, and the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter is rotating in reverse, so as to construct a dynamic model of the vector rotor drone;

[0106] Step 2: Based on the dynamic model of the vector rotor UAV and combined with the interference observer, obtain the interference force of wind on the quadcopter UAV;

[0107] Step 3: Utilize the wind's interference force on the quadcopter drone to obtain the compensation correction amount at the moment of rotor switching. ;

[0108] Step 4: Utilize The observer is optimized, and the optimized observer is used to obtain the force generated by the wind acting on the UAV. and torque ;

[0109] Step 5: Acquire an environmental image and input the image into the neural network to obtain the virtual tension output by the neural network;

[0110] Step 6: Construct the system state equations and base them on the forces generated by the wind acting on the UAV. Torque And virtual thrust, to obtain the desired servo command when the rotor reverses. and and the expected rotational speed of the propeller. This is used to control the propeller. and These represent the desired steering angles for the inner and outer servos, respectively.

[0111] Wind speed estimation

[0112] The core concept of this application is that the state information of a vector rotor UAV changes under the influence of wind. By analyzing the dynamic changes in state information and combining this with an interference observer, the interference force exerted by the wind on the UAV can be estimated.

[0113] The inertial coordinate system is selected as the northeast reference coordinate system. Establish the body coordinate system at the rotation centers of the four vector thrust units. This application further analyzes the kinematic characteristics of each vector drive unit of the UAV. A vector thrust unit with the following structure was established at its rotation center: Figure 1 The local coordinate system shown . The origin of the coordinate system is at the end of the boom and does not rotate with the tilting mechanism. In addition, the radial tilt coordinate system... With tangential tilt coordinate system As sub-coordinate systems, they correspond to the motion planes of the two-degree-of-freedom steering mechanism of the propulsion unit: the radial coordinate system's axis is defined along the extension direction of the arm, while the tangential coordinate system is orthogonal to it. This hierarchical coordinate system architecture provides a mathematical basis for accurately describing the spatial orientation changes of vector thrust, and also facilitates the mapping and calculation of control commands to the servo tilt angle. The spatial distance parameters from the origin of each local coordinate system to the origin of the body coordinate system are defined as key structural parameters of the system. Their values ​​are obtained through 3D modeling software and will be used as important compensation terms in subsequent controller design calculations.

[0114] Definition of the first Coordinate system of vector thrust unit With body coordinate system The difference between them is the rotation angle around the Z-axis of the machine body. ,but Compared to rotation matrix for:

[0115] (1)

[0116] In the formula, . The standard rotation matrix function for rotation about a single Z-axis:

[0117] (2)

[0118] Depend on The result is obtained by rotating twice, and its rotation matrix is ​​calculated using the following method:

[0119] (3)

[0120] , These are the standard rotation matrix functions for rotation around the X and Y axes of each vector element, respectively:

[0121] (4)

[0122] The roll angle is defined as the angle by which a vector rotor UAV rotates around its X-axis. The angle of rotation around the Y-axis of the aircraft is the pitch angle. The angle of rotation around the Z-axis of the fuselage is the yaw angle. Then the body coordinate system Relative to the reference coordinate system rotation matrix Represented as:

[0123] (5)

[0124] In the aircraft system, vector quadcopter UAVs are mainly affected by gravity, interference, lift generated by the propeller, and Coriolis force. Among these, The lift generated by the propeller was calculated using the following method:

[0125] (6)

[0126] (7)

[0127] In the formula, It is the propeller thrust in the engine system The expression in It is the propeller thrust in The expression in It is the thrust generated by the propeller, and it is a non-negative scalar. It is the aerodynamic lift coefficient. It is the propeller rotation angle.

[0128] Define the direction of the propeller's counterclockwise rotation as +1 and the direction of its clockwise rotation as -1. The aerodynamic lift coefficient of the propeller is... The aerodynamic counter-torque coefficient of the propeller is Then the first The formula for calculating the thrust generated when a motor rotates forward is:

[0129] (8)

[0130] No. The counter-torque generated when the motor rotates forward is:

[0131] (9)

[0132] No. The formula for calculating the thrust generated when a motor reverses is:

[0133] (10)

[0134] No. The counter torque generated when the motor reverses is:

[0135] (11)

[0136] During the switching between forward and reverse rotation, the propeller speed gradually decreases. The time increment decreases to 0, after which the propeller begins to reverse, and the rotational speed continuously increases. During the transition from forward to reverse rotation of the propeller, the [time increment]... The formula for calculating the thrust generated by a single motor is:

[0137] (12)

[0138] No. The formula for calculating the counter torque generated by each motor is:

[0139] (13)

[0140] Combining (3), (7), and (8), we can obtain the first... The thrust generated by each propeller:

[0141] (14)

[0142] Similarly, when the rotor rotates clockwise... The formula for the torque generated by a propeller is:

[0143] (15)

[0144] It is the first When the propeller rotates clockwise, the direction of rotation is as follows: when looking at the propeller along the Z-axis of the vector thrust unit, the propeller rotates clockwise. The propeller rotates clockwise to take... .

[0145] Combining (3), (7), and (10), we can obtain the first step when the rotor reverses. The thrust generated by each propeller:

[0146] (16)

[0147] Similarly, when the rotor rotates clockwise... The formula for the torque generated by a propeller is:

[0148] (17)

[0149] From (14) and (15), it is easy to see that the force and torque generated by the propeller during forward rotation satisfy the following equation:

[0150] (18)

[0151] Similarly, the force and torque generated by the propeller during reverse rotation satisfy the following equation:

[0152] (19)

[0153] The drone will now be considered as a whole and its dynamics will be analyzed. The total lift force acting on the drone... Equal to the force generated by 4 propellers The sum of the torques generated by the propellers on the drone. equal With distance cross product plus The sum is:

[0154] (20)

[0155] (twenty one)

[0156] In the formula, , yes Origin and The distance between the origins.

[0157] definition , It is a vector By combining the antisymmetric matrix (20) and (21), we can obtain the following control allocation equation:

[0158] (twenty two)

[0159] in, It is the control efficiency matrix, and its definition formula can be obtained from... and the 3rd order identity matrix Given:

[0160] (twenty three)

[0161] The dynamic model of a vector rotor UAV is quite complex. Ignoring the dynamic terms of the rotational motion of the vector thrust unit and the propeller acceleration, and their resulting dynamic effects, and assuming that the vector rotor UAV is a system with invariant inertial parameters (i.e., inertial tensor, mass, and center of mass position remain constant), a simplified dynamic model of the vector rotor UAV can be obtained:

[0162] (twenty four)

[0163] (25)

[0164] in, It's about the quality of the drone. It refers to the position of the drone in the reference frame. It is the inertial tensor of the drone. , It is the gravitational constant. It is the angular velocity of the drone within the machine system. It refers to the interference force experienced by the drone in the reference frame. It is the disturbance torque experienced by the on-board system of the drone. It is the rotation matrix from the body coordinate system to the reference coordinate system.

[0165] To facilitate the design of the interference observer, (24) and (25) are integrated together to obtain the following matrix form:

[0166] (26)

[0167] (27)

[0168] The interference observer designed in this application takes the following form:

[0169] (28)

[0170] When the disturbance change value is small, it can be approximated as follows:

[0171] (29)

[0172] It can be demonstrated that even with a large rate of change of disturbance, the observer can still achieve observations very well.

[0173] The observation error is defined as:

[0174] (30)

[0175] The derivation yields:

[0176] (31)

[0177] Design intermediate variables for:

[0178] (32)

[0179] It is a diagonal matrix in which all elements are greater than 0.

[0180] The stability of the observer is proven as follows:

[0181] Choose positive definite Lyapunov equations:

[0182] (33)

[0183] Obviously, It is positive.

[0184] but:

[0185] (34)

[0186] Due to the matrix All elements are inherent properties of the vector rotor UAV and are constants, therefore:

[0187] (35)

[0188] thereby:

[0189] (36)

[0190] because They are all diagonal matrices and all their elements are positive numbers, therefore It is a negative definite function, therefore the observer satisfies Lyapunov stability. Error ,thereby .

[0191] Substituting (26) into (28) yields:

[0192] (37)

[0193] Since wind speed can be calculated using only the aerodynamic forces generated by the wind field, only the observation formulas for aerodynamic forces are shown here. From the part extracting the wind disturbance force in (37), we can obtain:

[0194] (38)

[0195] According to the difference formula:

[0196] (39)

[0197] (38) can be transformed into a discrete form that is easier for simulation platforms and microcontrollers to process:

[0198] (40)

[0199] in, That is the sampling time. yes The third-order principal form, These are observations of aerodynamic forces. It controls the rounds.

[0200] Because the rotor switches from forward to reverse rotation at the instant... The direction of the flip causes a large jump in its value, which in turn makes... and This also causes significant changes, resulting in changes in the observed values. The observation deviated significantly from the observed values ​​and required reconvergence. This entire process severely impacted the control performance of the subsequent wind-resistant controller, necessitating optimization of the observer. This application addresses this by introducing a compensation correction at the moment of switching. This allows the observer to skip the reconvergence process, achieving seamless switching of force vectors.

[0201] Assuming the wind speed observer has already converged before the rotor switching, at this time Substituting (40) yields the result before switching. The expression is:

[0202] (41)

[0203] in yes The acceleration of the drone in the reference frame at that moment. yes The total lift force experienced by the drone's onboard system at any given moment.

[0204] Assuming the drone is in The rotor reversal is completed at the moment. Substituting (41) into (40) yields the formula for calculating the wind speed observation value when the UAV rotor has just switched to reversal:

[0205] (42)

[0206] Combining (6), (14) and (20), it can be seen that when the rotor rotates forward... The expression is:

[0207] (43)

[0208] Similarly, when the rotor reverses... The expression is:

[0209] (44)

[0210] Due to the angle , , and angular velocity There will be no jumps in a very short period of time, therefore... Time and The time interval can be considered as a constant value, and the two time intervals are... The difference is:

[0211] (45)

[0212] Substituting (45) into equation (42), we can obtain the amount that needs to be compensated for by the wind speed observer after the rotor switches to reverse:

[0213] (46)

[0214] in yes Time and The difference in acceleration at two different moments. The acceleration at two different moments can be read from the accelerometer.

[0215] From the aerodynamic model:

[0216] (47)

[0217] The relationship between wind speed and the aerodynamic force it generates is obtained. Force generated by wind The drag coefficient of the drone. The effective windward area of ​​the drone, It is air density. It is the speed of the wind relative to the drone.

[0218] In the reference coordinate system, the force observed by the observer is decomposed into components along the X, Y, and Z axes. , , Substituting the three values ​​into (47) yields the relative velocities in the three directions. , , :

[0219] (48)

[0220] (49)

[0221] (50)

[0222] Combining the three factors yields the vector form of the relative velocity:

[0223] (51)

[0224] Finally, the difference between the drone's velocity in the reference frame and the relative wind speed can be used to obtain the actual wind speed in the reference frame:

[0225] (52)

[0226] Obstacle perception and position calculation

[0227] An airborne binocular camera acquires images of the foreground environment and converts them into pixel-level depth maps. Based on the camera's intrinsic parameter matrix, the depth maps are converted into 3D point cloud data in the aircraft's coordinate system. Combining the UAV's current attitude angles (obtained from the IMU), a coordinate transformation matrix is ​​used to map the point cloud to the inertial reference coordinate system. A safe distance threshold is set to filter out potential obstacle point sets, and the coordinates of the obstacle centers and their boundary ranges are calculated, providing geometric constraints for subsequent trajectory planning.

[0228] Let the image captured by the camera be The resulting depth map is Since there are a large number of pixels, and adjacent pixels often contain the same information, it is necessary to first pool the depth map, taking the minimum value among every 5*5 pixels to obtain a simplified depth map. For depth maps All depth values ​​and thresholds Compare, when less than the threshold The depth value ratio reached If the value exceeds a certain threshold, it is determined that the target point or target point shelter has been reached, and obstacle avoidance is not performed; similarly, if the value exceeds a certain threshold... The depth value ratio reached If the obstacle is not present, it is determined that there is no obstacle and no obstacle avoidance is required.

[0229] This application uses a combination of Convolutional Neural Networks (CNNs) and Gated Recurrent Units (GRUs) to identify the position of obstacles relative to the drone and output the virtual pulling force required for the drone to perform obstacle avoidance. This combined model includes convolutional layers, fully connected layers, memory layers, and decision layers. The CNN has three convolutional layers with kernel sizes of [sizes to be filled in]. , , The activation function used is LeakyReLU, which is defined as follows:

[0230] (53)

[0231] The fully connected layer first merges and flattens the outputs of the convolutional layers into a one-dimensional vector, and then compresses it into a 128-dimensional visual feature vector. .

[0232] Because single-frame images may become blurry due to vibration during high-speed flight, or obstacles may temporarily move out of the field of view, a recurrent neural network (RNN) structure is introduced to give the network "short-term memory." and Inputting a single-layer GRU yields a hidden layer containing obstacle information from previous time steps. .

[0233] Finally and By substituting the two output values ​​into the decision layer and then into the activation function Tanh, we can obtain the output vector. The final virtual tension output by the neural network is:

[0234] (54)

[0235] in, These are virtual force coefficients. The subscript x represents the x-coordinate of the image captured by the camera, y represents the y-coordinate, η represents the direction vector of the virtual force, ηx represents the projection of the virtual force direction vector onto the x-axis, and ηy represents the projection onto the y-axis.

[0236] The loss function for neural networks is designed as follows:

[0237] The depth map used for training Each pixel in the vector is considered as a vector. (Pointing to this pixel from the center of the image), its depth value is Then calculate the weighted depth centroid vector of the entire map. :

[0238] (55)

[0239] Where the weight function This means that the farther away from the obstacle, the greater the weight. This is how it's calculated. It will automatically point to the deepest, emptiest hole.

[0240] Define loss function for:

[0241] (56)

[0242] in, It is the loss coefficient. and The more consistent the directions, the smaller the loss.

[0243] Wind control

[0244] The control process of a vector rotor UAV is quite complex. First, the UAV needs to acquire external environmental information, including its own position, attitude, and obstacle positions, using sensors such as an inertial measurement unit (IMU), global positioning system (GPS), camera, or lidar. This information is then transmitted as input data to the controller. The controller, based on the control algorithm and in conjunction with the trajectory planning target, flight status, and obstacle information, performs real-time calculations to generate the total control torque that meets the requirements for attitude stability and trajectory tracking. To adapt to the special configuration of the vector propulsion system, the control quantities must be converted into executable physical commands through a thrust distribution matrix. In this process, a geometrically constrained optimization algorithm is introduced to solve for the optimal action combination of each actuator while satisfying the tilt angle limits of each servo and the motor thrust saturation boundary conditions. Finally, the servo tilt angle command and the motor speed adjustment signal are converted into physical quantities through a high-precision servo drive module, thus constructing a complete closed-loop control link from environmental perception to action execution.

[0245] The defined linear system and its performance metrics are as follows:

[0246] (57)

[0247] Define the desired position of a vector rotor UAV in a reference frame as follows: The desired attitude in the machine system is Then, the position and attitude trajectory tracking deviations can be expressed as follows:

[0248] (58)

[0249] (59)

[0250] Therefore, the velocity and angular velocity deviations, and the acceleration and angular acceleration are expressed as:

[0251] (60)

[0252] (61)

[0253] (62)

[0254] (63)

[0255] in It is the system input. This is the deviation of the system. Combining (58) and (63) yields the system's state equation:

[0256] (64)

[0257] in:

[0258] (65)

[0259] (66)

[0260] After repeated debugging, the state weight function was selected. for:

[0261] (67)

[0262] Selecting the state weight function for:

[0263] (68)

[0264] Through the algebraic Riccati equation:

[0265] (69)

[0266] The matrix can be solved. Then by Calculate the feedback gain :

[0267] (70)

[0268] LQR's optimal control input In the form of status feedback:

[0269] (71)

[0270] The LQR controller is used to solve for the system input. , Substituting (62) and (63) yields:

[0271] (72)

[0272] (73)

[0273] The virtual force mentioned above The observed wind forces and torques acting on the drone , and , Substituting both into (24) and (25) together, we can solve for... and :

[0274] (74)

[0275] (75)

[0276] Due to the control efficiency matrix in (23) It's not a square array, so it cannot be passed directly. Solve This application uses the pseudo-inverse method for control allocation.

[0277] The basic formula for the pseudo-inverse method is:

[0278] (76)

[0279] In the formula, It is a matrix The pseudo-inverse matrix can be used to solve (22):

[0280] (77)

[0281] Based on formula (77), and combined with the rotation matrix ,get:

[0282]

[0283] Furthermore, based on equations (6), (7), and (8), the expected command of the servo motor when the rotor is rotating forward is... and and the expected rotational speed of the propeller The calculation is as follows:

[0284] (78)

[0285] Similarly, based on equations (6), (7), and (10), the desired command of the servo motor when the rotor reverses can be obtained. and and the expected rotational speed of the propeller Each arm contains two servos, which rotate in perpendicular directions, thus theoretically allowing the thrust generated by each propeller to reach any direction. and These are the expected steering angles corresponding to the inner and outer servos, respectively.

[0286] (79)

[0287] in It is a three-dimensional vector composed of three scalars. (1) / (2) / (3) means taking the 1st / 2nd / 3rd scalar in the vector.

[0288] It should be noted that the specific embodiments are merely explanations and illustrations of the technical solution of the present invention and should not be used to limit the scope of protection. Any modifications made in accordance with the claims and specification of the present invention that are only partial should still fall within the protection scope of the present invention.

Claims

1. A high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor, characterized in that... Includes the following steps: Step 1: Obtain the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter is rotating forward, and the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter is rotating in reverse, so as to construct a dynamic model of the vector rotor drone; Step 2: Based on the dynamic model of the vector rotor UAV and combined with the interference observer, obtain the interference force of wind on the quadcopter UAV; Step 3: Utilize the wind's interference force on the quadcopter drone to obtain the compensation correction amount at the moment of rotor switching. ; Step 4: Utilize The observer is optimized, and the optimized observer is used to obtain the force generated by the wind acting on the UAV. and torque ; Step 5: Acquire an environmental image and input the image into the neural network to obtain the virtual tension output by the neural network; Step 6: Construct the system state equations and base them on the forces generated by the wind acting on the drone. Torque And virtual thrust, to obtain the desired servo command when the rotor reverses. and and the expected rotational speed of the propeller. This is used to control the propeller. and These represent the desired steering angles for the inner and outer servos, respectively.

2. The high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor according to claim 1, characterized in that... The specific process for obtaining the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter UAV rotates forward in step 1 is as follows: Step 11: Select the northeast reference coordinate system The inertial coordinate system is used as the initial coordinate system, and then the body coordinate system is established at the rotation centers of the four vector thrust units of the quadcopter UAV. ; Step 12: In Establish a local coordinate system at the rotation center of the vector thrust unit. , The origin of the coordinate system is at the end of the boom and does not rotate with the tilting mechanism; it is a radial tilting coordinate system. With tangential tilt coordinate system As sub-coordinate systems, they correspond to the motion planes of the two-degree-of-freedom steering mechanism of the thrust unit: Step 13: Define the first Coordinate system of vector thrust unit With body coordinate system The difference in rotation angle around the Z-axis of the aircraft is ,but Compared to rotation matrix for: ， in, , The standard rotation matrix function for rotation about a single Z-axis is expressed as: ， Will Rotate twice to obtain Rotation matrix for: ， , These are the standard rotation matrix functions for single-axis rotation about the X and Y axes of each vector element, respectively. for The internal servo angle corresponding to the number of vector thrust units for The external servo angle corresponding to the thrust vectoring unit: ， The roll angle is defined as the angle by which a vector rotor UAV rotates around its X-axis. The angle of rotation around the Y-axis of the aircraft is the pitch angle. The angle of rotation around the Z-axis of the fuselage is the yaw angle. Then the body coordinate system Relative to the reference coordinate system rotation matrix for: ， This is the standard rotation matrix function for single-axis rotation about the Z-axis of each vector element; The lift generated by the propeller is: ， ， in, For propeller thrust in the engine system The expression in For propeller thrust in The expression in Let be the thrust generated by the propeller, a non-negative scalar. The aerodynamic lift coefficient, This refers to the propeller rotation angle; Define the direction of the propeller's counterclockwise rotation as +1 and the direction of its clockwise rotation as -1. The aerodynamic lift coefficient of the propeller is... The aerodynamic counter-torque coefficient of the propeller is Then the first Each motor generates thrust when it rotates forward. for: ， No. The counter torque generated when the motor rotates forward for: ， No. The motor generates thrust when it reverses. for: ， No. The counter torque generated when the motor reverses for: ， During the period when the propeller switches from forward to reverse rotation, the first Each motor generates thrust. for: ， No. The counter torque generated by each motor is: ， Combining the above equations, we obtain the first equation when the rotor rotates clockwise. The thrust generated by the propeller for: ， When the rotor rotates clockwise The torque generated by the propeller for: ， in, For the first When the propeller rotates clockwise, the direction of rotation is as follows: when looking at the propeller along the Z-axis of the vector thrust unit, the propeller rotates clockwise. The propeller rotates clockwise to take , This refers to the moment when the propeller's rotational speed drops to 0 when it switches from forward to reverse rotation. For the current moment, for The first derivative.

3. The high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor according to claim 2, characterized in that... In step 1, the thrust and torque generated by the propeller when each vector thrust unit of the quadcopter UAV reverses its direction are expressed as follows: When the rotor reverses The thrust generated by each propeller is: ， When the rotor reverses The torque generated by each propeller is: 。 4. The high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor according to claim 3, characterized in that... The dynamic model of the vector rotor UAV is represented as follows: ， ， in, For the quality of drones, for The second derivative, The position of the UAV in the reference frame. For the inertial tensor of the drone, The angular velocity of the drone within the machine system. The interference force experienced by the UAV in the reference frame, The disturbance torque experienced by the UAV's on-board system. Let be the rotation matrix from the body coordinate system to the reference coordinate system. for The first derivative, The total lift force experienced by the drone, For the representation of gravity in three-dimensional force space in a reference coordinate system, , It is the gravitational constant. This refers to the total torque generated by the propeller on the drone.

5. The high-speed collision avoidance and landing control method for a vector rotor UAV based on a bidirectional motor according to claim 4, characterized in that... The interference observer is represented as: ， ， ， in, To interfere with the observer, As an intermediate variable, For the observation matrix, for The observed values, For the mass-inertia matrix, It is a diagonal matrix in which all elements are greater than 0.

6. The high-speed collision avoidance and landing control method for a vector rotor UAV based on a bidirectional motor according to claim 5, characterized in that... The interference force of the wind on the quadcopter drone is expressed as: ， in, For the first Observer output value in each round It is a 3-order identity matrix. For the first Observer output value in each round Sampling time, for The third-order principal form, These are the observed values ​​of aerodynamic forces. To control the number of rounds.

7. The high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor according to claim 6, characterized in that... The compensation correction amount of the rotor at the moment of switching Represented as: ， in, for Time and The difference in acceleration at time t.

8. The high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor according to claim 7, characterized in that... The specific steps of step 5 are as follows: Acquire images of the environment in front of the drone and images of the environment in front of the drone. Convert to depth map Then, the depth map Pooling is performed, and the minimum value is taken among every 5*5 pixels to obtain a simplified depth map. Then the depth map All depth values ​​and thresholds Compare, when less than the threshold The depth value ratio reached If the value exceeds a certain threshold, it is determined that the target point or target point shelter has been reached, and no obstacle avoidance is performed. The depth value ratio reached If the obstacle is not present, then it is determined to be an obstacle-free situation, and no obstacle avoidance is performed; otherwise, obstacle avoidance is performed. When avoiding obstacles, view the environment in front of the drone. Inputting data into a neural network yields the virtual tension output by the neural network. Virtual tension Represented as: ， in, For virtual force coefficients, The neural network specifically performs the following steps: Step 51: Image of the environment in front of the drone First, the image passes through three convolutional layers, then a fully connected layer receives the output of the convolutional layers to obtain the visual feature vector at the current time step. ; Step 52: Obtain the visual feature vector at the current moment. The hidden layer state from the previous time step is input into the gated recurrent unit to obtain the hidden layer state at the current time step. ; Step 53: Obtain the visual feature vector at the current moment. and the hidden layer state at the current moment Input to the decision layer, and obtain the output vector. ,in, The direction vector of the virtual force is in Projection of the axis The direction vector of the virtual force is in Projection of the axis; The loss function of the neural network is expressed as: ， ， in, The pixel pointing from the image center of the depth map vector, for The depth value, The weighted depth centroid vector of the entire map, the weight function , This is the loss coefficient.

9. The high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor according to claim 8, characterized in that... The system state equation is expressed as: ， ， ， ， ， ， in, For feedback gain, Let be the solution matrix of the Riccati equation. For the state weight function, for The first derivative, For the deviation of the system, For system input, For the system matrix, For the input matrix, It is a state variable.

10. The high-speed collision avoidance and descent control method for a vector rotor UAV based on a bidirectional motor according to claim 9, characterized in that... The expected command of the servo when the rotor reverses and and the expected rotational speed of the propeller. Represented as: ， ， ， ， ， ， in, For vectors antisymmetric matrix, for Origin and The distance between the origins.

Images