A group-invariant filtering state estimation method for passive flight phase of catapulted unmanned aerial vehicle

Abstract

The application discloses a kind of group invariant filtering state estimation methods of passive flight phase of catapult unmanned aerial vehicle, belong to unmanned aerial vehicle flight control and navigation technical field.Method includes: establishing the state matrix of unmanned aerial vehicle on group manifold to define Lie algebra error column vector;According to the state matrix of unmanned aerial vehicle on group manifold and the inertial measurement unit data obtained, obtain the prior estimation state of current time step;Recursive calculation obtains the prior group error covariance matrix of current time step;Dynamic shielding reduced-order strategy of specific force observation channel is executed, and reduced-order comprehensive observation innovation column vector is constructed;Isomorphic output observation matrix is constructed, the Kalman gain matrix of current time step is calculated, and the Lie algebra error column vector on tangent space is solvedGroup posterior update is carried out to prior estimation state and is output.The application effectively avoids the error accumulation of traditional extended Kalman filter when being impacted by catapulting greatly, and restrains the attitude solution divergence of catapult unmanned aerial vehicle in high dynamic environment.

Description

Technical Field

[0001] This invention belongs to the field of UAV flight control and navigation technology, specifically relating to a group-invariant filtering state estimation method for the passive flight phase of a catapult-launched UAV. Background Technology

[0002] With the rapid advancement of drone technology, these flight platforms have demonstrated significant application value in fields such as environmental monitoring, disaster relief, and field reconnaissance. To improve portability, folding-arm drones have gradually become a research hotspot. These drones require a catapult to gain sufficient initial velocity within a very short time during takeoff to ensure smooth arm deployment and allow the aircraft to enter a stable flight state.

[0003] However, after being launched aerodynamically, a folding-arm drone inevitably experiences a brief "passive flight phase" (e.g., 0 to 2 seconds). During this phase, the drone's propulsion system is not yet fully engaged, and the aircraft typically operates in a "bullet-like" shape, relying on initial kinetic energy for purely inertial passive flight. This phase is characterized by significant aerodynamic disturbances, drastic attitude changes, and extremely high non-gravity axial overloads, making it a "high-risk blind spot" where uncontrolled tumbling is highly likely to occur during the entire flight mission.

[0004] To address this extreme physical challenge, traditional navigation estimation algorithms have revealed serious theoretical and engineering flaws in practical applications. First, during the highly dynamic ejection phase, the update frequency of conventional GPS signals is too low (typically 1 to 10 Hz) and there is a data delay of 100 to 500 ms. Coupled with severe multipath effects at high acceleration, traditional algorithms relying on GPS updates are completely unable to meet the high dynamic accuracy requirements within the short ejection timeframe. Second, in pure inertial calculations without GPS, existing extended Kalman filters (such as the PX4 EKF2 algorithm commonly used in open-source flight controllers) linearize their error state in the tangent space. Faced with the enormous impact of ejection, system errors easily exceed the linear range, leading to severe nonlinear error accumulation when repeatedly linearized at time-varying estimation points. Even more critically, traditional attitude estimation algorithms heavily rely on accelerometers to extract gravity vectors for horizontal attitude correction. With initial launch velocities reaching tens of meters per second and immense aerodynamic drag at the moment of exiting the launch tube, traditional algorithms mistakenly interpret the massive forward nongravitational deceleration as a change in the direction of gravity, leading to catastrophic divergence in attitude calculations. This attitude calculation collapse can further trigger a cascading disaster in position and velocity states, ultimately causing the UAV navigation system to completely fail and posing a crash hazard. Summary of the Invention

[0005] To address the technical problem of nonlinear error accumulation in traditional attitude estimation algorithms under high maneuvering and extreme impact, this invention provides a group-invariant filtering state estimation method for the passive flight phase of a catapult-launched UAV. This method is specifically designed for the highly dynamic, short-duration, and GPS-free scenario of UAV catapult takeoff. It utilizes the left-invariant structure of the extended attitude Lie group to fix the linearization point at the unit element, transforming the traditional algorithm's method of "repeated linearization at time-varying estimation points" into "one-time linearization at the fixed origin combined with group motion." This method effectively avoids the error accumulation of traditional extended Kalman filtering under extreme catapult impact and suppresses attitude divergence in catapult-launched UAVs under high-dynamic environments.

[0006] The technical solution of this invention is:

[0007] According to a first aspect of the present invention, a group-invariant filtering state estimation method for the passive flight phase of a catapult-launched unmanned aerial vehicle is provided, comprising:

[0008] Step S1: Obtain initial physical constraint parameters; wherein, the physical constraint parameters include the launch elevation angle of the catapult launcher and the initial launch velocity;

[0009] Step S2: Construct the real attitude rotation matrix from the machine system to the world system and the real velocity column vector in the world system based on the initial physical constraint parameters. Establish the state matrix of the UAV on the group manifold based on the real attitude rotation matrix, the real velocity column vector in the world system and the real position column vector.

[0010] Step S3: Based on the state matrix of the UAV on the group manifold, define the group invariant error form of the UAV state as a left invariant error matrix, and expand the left invariant error matrix on the extended attitude Lie group; through the logarithmic mapping relationship from Lie group to Lie algebra, map the expansion of the left invariant error matrix on the extended attitude Lie group into a 9-dimensional Lie algebra error column vector; based on the 9-dimensional Lie algebra error column vector, establish the derivation relationship between the state matrix of the UAV on the group manifold and the estimated state matrix; then, initialize the group error covariance matrix.

[0011] Step S4: Based on the state matrix of the UAV on the group manifold and the acquired inertial measurement unit data, construct a state transition model to obtain the prior estimated state at the current time step.

[0012] Step S5: Based on the initialized group error covariance matrix and the inertial measurement unit data, recursively calculate the prior group error covariance matrix for the current time step.

[0013] Step S6: Execute the order reduction strategy of dynamically shielding the force observation channel to construct a reduced-order comprehensive observation information column vector based on prior estimated state, air pressure and geomagnetic measurement data;

[0014] Step S7: Construct the uniform output observation matrix; calculate the Kalman gain matrix for the current time step based on the uniform output observation matrix and the prior group error covariance matrix;

[0015] Step S8: Solve for the Lie algebra error column vector in the tangent space based on the Kalman gain matrix and the reduced-order synthetic observation innovation column vector;

[0016] Step S9: Based on the solved Lie algebra error column vector, perform manifold posterior update on the prior estimated state and output it.

[0017] Furthermore, S4 specifically includes:

[0018] At the initial moment of the passive flight phase, the state matrix of the established UAV on the group manifold is extracted. The true attitude rotation matrix, true velocity column vector, and true position column vector in the matrix are assigned values ​​as the initial posterior estimated attitude rotation matrix. Posterior estimation of velocity column vector and posterior estimated position column vector Using this as the starting point for iteration, during the passive flight phase, inertial measurement unit data is acquired, a state transition model is constructed, and the prior estimated state for the current time step is obtained:

[0019] ;

[0020] ;

[0021] ;

[0022] In the formula, , and These represent the time steps from the current time step. The prior estimated attitude rotation matrix, the prior estimated velocity column vector, and the prior estimated position column vector are obtained by integral calculation; , and Representing time steps The posterior estimated attitude rotation matrix, the posterior estimated velocity column vector, and the posterior estimated position column vector; It is an exponential mapping function; Indicates the current time step The column vector of the machine's angular velocity, actually measured by the gyroscope in the inertial measurement unit; Indicates the current time step The mechanical system specific force column vector obtained from actual measurements by the accelerometers in the inertial measurement unit; This represents a column vector of world-system gravitational constants pre-stored in the microprocessor; This indicates the discrete time interval for a single sampling by the inertial measurement unit. This represents an operator that maps a three-dimensional vector to an antisymmetric matrix.

[0023] Furthermore, S5 specifically includes:

[0024] Using the group error covariance matrix initialized in step S3 The posterior group error covariance matrix at the initial moment of the passive flight phase Combining the inertial measurement unit data obtained in step S4, the prior group error covariance matrix for the current time step is recursively calculated. The expression is as follows:

[0025] ;

[0026] ;

[0027] ;

[0028] In the formula, This represents the calculated current time step. The prior group error covariance matrix; Indicates time step The posterior group error covariance matrix; For the current time step The discretized state transition matrix; Indicates transpose; This represents the predefined system process noise covariance matrix; This indicates the discrete time interval for a single sampling by the inertial measurement unit. for An identity matrix of dimension 1 express An identity matrix of dimension 1 express A zero matrix of dimension; For the current time step The continuous-time systematic error adjoint matrix, This represents an operator that maps a three-dimensional vector to an antisymmetric matrix. Indicates time step The posterior estimate of the attitude rotation matrix; Indicates the current time step The column vector of the machine's angular velocity, actually measured by the gyroscope in the inertial measurement unit; Indicates the current time step The mechanical system specific force column vector obtained by actual measurement from the accelerometer in the inertial measurement unit.

[0029] Furthermore, S6 specifically includes:

[0030] Extract the current time step obtained in step S4 The prior estimated position column vector in the prior estimated state Compared with prior estimated attitude rotation matrix And combined with the current time step Barometer altitude measurements acquired in real time from the inertial measurement unit With the three-axis geomagnetic vector measurement value of the magnetometer A reduced-order comprehensive observation novel column vector is constructed. :

[0031] ;

[0032] In the formula, Indicates the current time step The resulting reduced-order synthetic observation information column vector; This represents the calculated current time step. Barometer height residual scalar; This represents the calculated current time step. Three-axis geomagnetic observation residual column vector; Represents the column vector of prior estimated positions Vertical extracted Axial direction component; Indicates transpose; This represents the pre-stored local world-system reference geomagnetic vector.

[0033] Furthermore, S7 specifically includes:

[0034] Constructing a nonlinear system-level integrated observation model Equivalent output observation matrix Through the comprehensive observation model At the identity element of the Lie group, the analytical partial derivative mapping matrix obtained by linearizing the Lie algebraic perturbation based on the defined Lie algebraic error column vector is as follows:

[0035] ;

[0036] Based on the constructed isotropic output observation matrix The Kalman gain matrix is ​​calculated using the following formula:

[0037] ;

[0038] ;

[0039] In the formula, Indicates the current time step The new information covariance matrix; This represents a pre-stored diagonal matrix of the overall measurement noise covariance, which includes the noise characteristics of barometer and magnetometer measurements. Indicates the current time step The Kalman gain matrix; Indicates the current time step The prior group error covariance matrix; This represents the pre-stored local world-system reference geomagnetic vector; Indicates transpose; express A zero matrix of dimension 1 express The zero matrix; [0,0,1] represents a 1×3 dimensional matrix; This represents an operator that maps a three-dimensional vector to an antisymmetric matrix; express The inverse matrix.

[0040] Furthermore, S8 specifically includes:

[0041] The current time step calculated in step S7 Kalman gain matrix As a linear mapping operator, it is applied to the current time step constructed in step S6. Reduced-order comprehensive observation new information column vector Above, through matrix multiplication, the Lie algebraic error column vector in the current time step tangent space is derived and calculated. :

[0042] ;

[0043] In the formula, , , These are the components of the attitude, velocity, and position dimensions in the Lie algebra error column vector at the current time step.

[0044] Furthermore, S9 specifically includes:

[0045] Using the components of attitude, velocity, and position in the Lie algebra error column vector in the current time step space calculated in step S8 For the current time step obtained in step S4 The prior estimated state is corrected posteriorly on the manifold tangent space; simultaneously, based on the current time step calculated in step S7... Kalman gain matrix and equivalent output observation matrix For the current time step obtained in step S5 Prior group error covariance matrix Perform a closed-loop update; the specific update formula is as follows:

[0046] ;

[0047] ;

[0048] ;

[0049] ;

[0050] In the formula, , , These represent the current time step obtained after posterior correction in the tangent space of the manifold. Posterior estimation of attitude rotation matrix, posterior estimation of velocity column vector, and posterior estimation of position column vector; prior estimation of state... , and These represent the time steps from the current time step. The prior estimated attitude rotation matrix, the prior estimated velocity column vector, and the prior estimated position column vector are obtained by integral calculation; To complete the update and obtain the current time step Posterior group error covariance matrix; for An identity matrix of dimensionality; It is an exponential mapping function; This represents an operator that maps a three-dimensional vector to an antisymmetric matrix.

[0051] Furthermore, the catapult launcher serves as a physical constraint guide rail for launching UAVs at different launch elevation angles. The catapult launcher includes: a horizontal support base, a rotating bracket, a gas cylinder, a launch profile bracket, a launch tube, a planar cross-connecting plate, a launch tube connecting cover, a steering seat, a steering profile base frame, a gas valve, a solenoid valve, and a piston. The horizontal support base is equipped with the steering profile base frame and two sets of parallel-arranged launch profile brackets. The planar cross-connecting plate is installed at the end of the launch profile bracket away from the horizontal support base, and the two spaced-apart steering seats are installed on the steering profile base frame. The first end of the rotating bracket is rotatably engaged with the steering seat, and the second end of the rotating bracket engages with the planar cross-connecting plate at different positions to adjust the angle of the rotating bracket. The third end of the rotating bracket is equipped with a gas cylinder and a launch tube. The gas cylinder's outlet is connected to a tube body via a gas valve and a solenoid valve, and the tube body's output end is connected to the interface of the launch tube connecting cover at the end of the launch tube. The launch tube contains a piston, and gas is filled into the carbon fiber launch tube space between the launch tube connecting cover and the piston through the gas cylinder.

[0052] Furthermore, the rotating bracket includes rotating profile rods and a pneumatic storage and release bracket; one end of the two parallel rotating profile rods serves as the first end of the rotating bracket for hinged connection with the steering seat; the other end of the two parallel rotating profile rods serves as the second end of the rotating bracket; and a pneumatic storage and release bracket perpendicular to the rotating profile rods is provided on one side of the two parallel rotating profile rods as the third end of the rotating bracket.

[0053] According to a second aspect of the present invention, a group-invariant filter state estimation system for the passive flight phase of a catapult-launched unmanned aerial vehicle (UAV) is provided, comprising modules of the group-invariant filter state estimation method for the passive flight phase of a catapult-launched UAV as described in any of the above-mentioned embodiments. For details not described in the above-mentioned modules, please refer to the specific content regarding the estimation method in the content of this invention.

[0054] The beneficial effects of this invention are:

[0055] First, this method overcomes high-dynamic singularity and achieves geometric autonomy in pose updating. Addressing the severe tumbling problem caused by high overload during the catapult-assisted passive flight phase, this invention rigorously maps the UAV's actual state to an extended attitude Lie group in step S2. Manifold space, and utilize three-dimensional special orthogonal groups The attitude rotation matrix is ​​represented. Compared to traditional linear filtering methods based on Euler angles or quaternions, this scheme utilizes the underlying constraints of group geometry to effectively avoid the gimbal lock-up (singularity) problem under high dynamic roll. Furthermore, when dealing with high-frequency integration of extremely high angular velocities during ejection, microprocessors are prone to floating-point truncation errors, leading to the deorthogonalization and degradation of the traditional attitude matrix. This invention utilizes the equivariant mapping mechanism from Lie algebra to Lie groups to mathematically guarantee that the state update result always resides on the rigid body's pure rotational manifold. This effectively prevents the unbounded accumulation of numerical integration errors from inducing attitude divergence, ensuring that the posterior state (superscript "...") output through step S9 is consistent with the previous approach. It has better manifold fidelity and stability.

[0056] Secondly, the impact of nonlinear truncation error is reduced, improving the system's global convergence capability. Addressing the weakness of traditional Extended Kalman Filter (EKF) in easily diverging under large initial errors, this invention constructs a left-invariant error estimation architecture based on the Lie algebra error column vector defined in step S3. During the covariance recursion in step S5, the system's error propagation transfer equation exhibits good autonomy within the Lie group space, i.e., the prior group error covariance matrix (… The evolutionary calculation process does not depend on the prior predicted state (superscript ") calculated in step S4. The valuation mechanism effectively mitigates the nonlinear truncation error introduced by frequent Taylor expansions of the operating point, thereby improving the algorithm's robustness and convergence performance under severe nonlinear flight conditions.

[0057] Furthermore, dynamic channel order reduction suppresses measurement noise and optimizes real-time computation. To address the distortion of the force comparison channel in the inertial measurement unit (IMU) caused by the sudden acceleration during ejection, this invention employs a dynamic shielding strategy for the force comparison observation channel in step S6, extracting only the barometer height and magnetometer three-axis vector measurements to construct a reduced-order integrated observation information column vector. Subsequently, in step S7, Lie algebra perturbation linearization is performed on this vector, deriving an equivariant output observation matrix. It exhibits a relatively sparse block matrix structure. This method effectively isolates the interference of severe vibration noise from the accelerometer on the state estimation system at the physical level, and at the algorithmic computation level, it reduces the matrix inversion load when the microprocessor solves the Kalman gain matrix, which helps to ensure the real-time performance of the system under high-frequency clock ticks. Attached Figure Description

[0058] Figure 1 This is a flowchart of the group-invariant filtering state estimation method for the passive flight phase of a catapult-launched UAV according to the present invention.

[0059] Figure 2 To and Figure 3Comparison of divergence in three-axis attitude tracking using the traditional Extended Kalman Filter (EKF) algorithm under the same initial physical constraints.

[0060] Figure 3 This is a comparison diagram of the three-axis attitude tracking of the LIEKF group-invariant filtering algorithm used in this invention during the passive flight phase of catapult launch.

[0061] Figure 4 This is a comparison image of the three-dimensional position tracking of the LIEKF group-invariant filtering algorithm used in this invention during the passive flight phase of catapult launch.

[0062] Figure 5 This is a schematic diagram of the overall structure of the catapult launcher.

[0063] Figure 6 This is a schematic diagram of the entire catapult launcher from the left side.

[0064] Figure 7 This is a top-view structural diagram of the catapult launcher.

[0065] Figure 8 This is a schematic diagram of the aerodynamic storage and release structure of the catapult launcher.

[0066] Figure 9 This is a schematic diagram of the profile angle adjustment frame structure for the catapult launcher.

[0067] Figure 10 This is a schematic diagram illustrating the assembly of a catapult launcher for use with a catapult-launched aircraft.

[0068] Attached reference numerals: 1-Gas cylinder fixing plate, 2-Gas cylinder, 3-Launch profile bracket, 4-Launch tube, 5-Rotating profile rod, 6-60° sharp angle side connecting plate, 7-Plane cross connecting plate, 8-Launch tube fixing plate, 9-90° angle aluminum, 10-120° angle aluminum, 11-Launch tube connecting cover, 12-Launch tube base, 13-Gas cylinder profile frame, 14-Swivel seat, 15-Swivel profile base frame, 16-60° angle aluminum, 17-Launch frame profile horizontal base frame, 18-60° side connecting plate, 19-Solenoid valve, 20-Stainless steel corrugated pipe, 21-Launch frame profile longitudinal base frame, 22-Gas valve, 23-Gas cylinder fixing plate lower plate, 24-Launch tube profile base frame short rod, 25-Launch tube profile base frame long rod, 26-Piston. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.

[0070] Example 1: The execution code of the method in this example is burned and runs in the flight control microprocessor of the catapult-launched UAV. This microprocessor is connected via a data bus to: a high-frequency inertial measurement unit (IMU, sampling rate set to 400Hz), a barometer (sampling rate set to 50Hz), and a magnetometer (sampling rate set to 50Hz). In conjunction with... Figure 1 The control flowchart shown illustrates that the state estimation method of this invention includes the following specific execution steps:

[0071] Step S1: Obtain initial physical constraint parameters; wherein, the physical constraint parameters include the launch elevation angle of the catapult launcher. and initial velocity of the ejection ;

[0072] Step S2: The microprocessor constructs the true attitude rotation matrix from the machine system to the world system based on the obtained initial physical constraint parameters. The true velocity column vector in the world system Based on the true attitude rotation matrix The true velocity column vector in the world system column vector of actual positions Establish the state matrix of the UAV on the swarm manifold; the specific establishment process is as follows: construct the true attitude rotation matrix from the machine system to the world system based on the launch elevation angle of the catapult launcher. Based on the preset initial launch velocity and launch elevation vector Construct the true velocity column vector in the world system Rotate the actual pose matrix The true velocity column vector in the world system column vector of actual positions Unified mapping and definition in extended pose Lie groups The state matrix of the UAV on the group manifold is represented as follows: Matrix form:

[0073] ;

[0074] In the formula, This indicates that the drone is in an extended attitude Lie group. The state matrix on; Let represent the true attitude rotation matrix from the machine frame to the world frame, constructed from the launch elevation angle, and satisfy . , It represents a three-dimensional special orthogonal group whose internal matrix elements satisfy the strict mathematical constraints that the transpose is equal to the inverse and the determinant is positive one. It is used to characterize the rigid three-dimensional pure rotational attitude of a UAV without singularity in the Lie group manifold space. and Let these represent the true velocity column vector and the true position column vector in the world system (with the true position column vector of the launch point set as the origin), both belonging to three-dimensional real space. ; express A zero matrix of dimension 1 is a matrix whose elements are all zero. A zero matrix with 1 row and 3 columns.

[0075] By applying the above solution, the status of drones can be unified as follows: The matrix form allows microprocessors to use legal matrix multiplication within the manifold space to replace the vector addition and subtraction methods of traditional algorithms for error calculation.

[0076] Step S3: Based on the UAV state matrix established in Step S2 on the group manifold, define the group-invariant error form of the UAV state as a left-invariant error matrix, and expand the left-invariant error matrix on the extended attitude Lie group; through the logarithmic mapping relationship from Lie group to Lie algebra, map the expansion of the left-invariant error matrix on the extended attitude Lie group into a 9-dimensional Lie algebra error column vector. Based on the 9-dimensional Lie algebra error column vector Establish the state matrix of the UAV on the swarm manifold With the estimated state matrix The derivation of the relation is then performed; next, the group error covariance matrix is ​​initialized. .

[0077] Furthermore, S3 specifically refers to:

[0078] To ensure the geometric autonomy of error dynamics during the passive flight phase, the microprocessor defines the group-invariant error form of the UAV state as a left-invariant error matrix based on the UAV state matrix established in step S2 on the group manifold. The left-invariant error matrix in the extended attitude Lie group Expanded above:

[0079] ;

[0080] To perform linearization operations in the Euclidean tangent space of the Kalman filter, the microprocessor maps the expansion of the left-invariant error matrix over the extended attitude Lie group to a 9-dimensional Lie algebraic error column vector over the tangent space of the manifold using the logarithmic mapping from Lie group to Lie algebra. The definition is as follows:

[0081] ;

[0082] Based on the 9-dimensional Lie algebra error column vector Establish the state matrix of the UAV on the swarm manifold With the estimated state matrix The derivation formula is as follows:

[0083] ;

[0084] In the formula, Represents the left-invariant error matrix; Let be the attitude error rotation matrix on the group manifold; and These are the decoupled velocity and position error column vectors, respectively. , , They represent , and The estimated value, This represents the true attitude rotation matrix from the machine frame to the world frame, constructed from the launch elevation angle. and Let these represent the true velocity column vector and the true position column vector in the world frame, respectively; the attitude error rotation matrix on the group manifold. Through logarithmic mapping Convert to a 3D attitude error column vector ,Right now , , Indicates from arrive Logarithmic mapping; This represents the estimated state matrix recursively derived by the microprocessor, with the superscript " "A unified estimate of the filter;" Represents the estimated state matrix The inverse matrix; This represents a 9-dimensional Lie algebraic error column vector on the tangent space of the manifold. , Represents a nine-dimensional real number space; Represents the exponential mapping function; This represents the operator that maps a three-dimensional vector to an antisymmetric matrix. The above transformation achieves a one-to-one correspondence between the group error and the tangent space vectors, completing a rigorous mapping from the geometric group space to the linear error space.

[0085] Initialize the group error covariance matrix .

[0086] Step S4: Based on the state matrix of the UAV on the group manifold established in step S2 and the acquired inertial measurement unit data (i.e. , ), construct a state transition model, and obtain the prior estimated state at the current time step.

[0087] Furthermore, S4 specifically includes:

[0088] At the initial moment of the passive flight phase (i.e., the discrete time step) (At that time), the microprocessor extracts the state matrix of the UAV on the group manifold established in step S2. The true pose rotation matrix True velocity column vector and the column vector of the actual position They are assigned values ​​as the initial posterior estimated attitude rotation matrices. Posterior estimation of velocity column vector and posterior estimated position column vector Using this as the starting point for iteration, during the passive flight phase, the microprocessor acquires inertial measurement unit data at a frequency of 400Hz, constructs a state transition model, and realizes discretized calculation from the previous time step to the current time step, as shown in the following formula (i.e., from...). Begin calculating the prior estimated state at the current time step: the prior estimated attitude rotation matrix, the prior estimated velocity column vector, and the prior estimated position column vector at the current time step.

[0089] ;

[0090] ;

[0091] ;

[0092] In the formula, the subscript Indicates the current time step defined by the microprocessor clock, subscript Indicates the previous time step; , and These represent the time steps from the current time step. The prior estimated attitude rotation matrix, prior estimated velocity column vector, and prior estimated position column vector obtained by integral calculation, where the superscript " "The unified representation represents the prior prediction state before any observation updates are performed;" , and Representing time steps The posterior estimated attitude rotation matrix, posterior estimated velocity column vector, and posterior estimated position column vector, superscript " "The unified representation has completed the update of the posterior state (when...") At that time, the posterior states respectively represent the initial posterior estimated attitude rotation matrix. Posterior estimation of velocity column vector and posterior estimated position column vector ); This represents the column vector of the system's angular velocities at the current time step, actually measured by the gyroscope in the inertial measurement unit. This represents the system-specific force column vector obtained by the accelerometer in the inertial measurement unit at the current time step; This represents a column vector of world-system gravitational constants pre-stored in the microprocessor; This indicates the discrete time interval for a single sampling by the inertial measurement unit. This represents an operator that maps a three-dimensional vector to an antisymmetric matrix.

[0093] Step S5: Based on the group error covariance matrix initialized in step S3 and the inertial measurement unit data obtained in step S4, recursively calculate the prior group error covariance matrix for the current time step.

[0094] Furthermore, S5 specifically includes:

[0095] The microprocessor uses the group error covariance matrix initialized in step S3. The posterior group error covariance matrix at the initial moment of the passive flight phase Combining the inertial measurement unit data obtained in step S4, the prior group error covariance matrix for the current time step is recursively calculated. The expression is as follows:

[0096] ;

[0097] ;

[0098] ;

[0099] In the formula, This represents the calculated prior group error covariance matrix for the current time step. dimension); Let the posterior group error covariance matrix of the previous time step be (when hour, Equivalent to the initial covariance matrix initialized in step S3 That is, the initial covariance matrix As the Lie algebra error column vector at the initial moment of the passive flight phase The posterior group error covariance matrix ); The discretized state transition matrix for the current time step ( (Dimensional), the discretized state matrix is ​​the column vector of angular velocity obtained from the gyroscope measurement in step S4. The specific force column vector obtained from the accelerometer measurement Calculated in real time; Representation matrix The transpose of the matrix; This indicates that it is preset by the microprocessor. The system process noise covariance matrix; This indicates the discrete time interval for a single sampling by the inertial measurement unit. for An identity matrix of dimension 1 express An identity matrix of dimension 1 express A zero matrix of dimension; For the current time step The continuous-time systematic error adjoint matrix, This represents an operator that maps a three-dimensional vector to an antisymmetric matrix. Indicates time step The posterior estimated attitude rotation matrix can transform the mechanical system acceleration ratio to the ground coordinate system. Based on the above, this invention, by introducing a Lie group structure with left-invariant error, can construct a continuous-time system error adjoint matrix based on the posterior estimated attitude rotation matrix and inertial measurement unit data. This leads to the final result With the true pose rotation matrix Velocity column vector Position column vector It's irrelevant; it's been achieved. The matrix is ​​fixed at the unit element of the manifold, thus ensuring that the system error dynamics remain autonomous even under large maneuvering shocks, effectively avoiding error accumulation and system divergence caused by linearization point drift in traditional extended Kalman filtering.

[0100] Step S6: Execute the order reduction strategy of the dynamic shielding of the force observation channel based on the prior estimated state obtained in step S4 and the acquired air pressure and geomagnetic measurement data. , ), construct a new column vector of reduced-order comprehensive observation.

[0101] Furthermore, S6 specifically includes:

[0102] Considering that during the passive flight phase of the UAV, the mechanical system specific force column vector actually measured by the accelerometer in the inertial measurement unit described in step S4 fluctuates due to the impact of the ejection, the microprocessor executes a reduction strategy to dynamically shield the specific force observation channel, that is, to eliminate the dependence on the mechanical system specific force column vector when constructing the observation model. The microprocessor extracts the prior estimated position column vector of the current time step obtained from the integral prediction in step S4. Compared with prior estimated attitude rotation matrix Combined with the barometer altitude measurement value obtained in real time from the external interface of the inertial measurement unit (IMU) at the current time step. With the three-axis geomagnetic vector measurement value of the magnetometer A reduced-order comprehensive observation novel column vector is constructed. :

[0103] ;

[0104] In the formula, Indicates the current time step The resulting reduced-order synthetic observation information column vector; This represents the calculated barometer height residual scalar; This represents the calculated column vector of three-axis geomagnetic observation residuals; This represents the column vector of prior estimated positions obtained from step S4. Vertical extracted Axial direction component (based on the definition of the North-East-Earth coordinate system, the predicted height is usually taken as a negative value, so it is corrected with a negative sign); This represents the prior estimated attitude rotation matrix obtained in step S4. The matrix obtained by performing a transpose operation; This represents the local world-system reference geomagnetic vector pre-stored inside the microprocessor.

[0105] Step S7: Construct the uniform output observation matrix; calculate the Kalman gain matrix for the current time step based on the uniform output observation matrix and the prior group error covariance matrix obtained in step S5.

[0106] Furthermore, S7 specifically includes:

[0107] To characterize the sensitivity of system state errors to the measurement space, the microprocessor first constructs a corresponding nonlinear system-level integrated observation model within the manifold space. Its expression is ,in and These are the preset air pressure altitude and geomagnetic vector observation functions, respectively. This is essentially a nonlinear system-level integrated observation model. An analytical mapping from the extended attitude Lie group state space to the sensor measurement space was established. The preset pressure-altitude observation function is characterized as follows: Used to extract from the system state matrix Directly extract vertical The position component along the axial direction is used to construct the predicted pressure altitude; the preset geomagnetic vector observation function is characterized as follows: Used to utilize the system state matrix The true pose rotation matrix The local world system reference geomagnetic vector pre-stored in the microprocessor Projecting these values ​​onto the current body coordinate system constructs a three-dimensional geomagnetic vector prediction. The microprocessor then quantifies the system's observational information by subtracting the predicted values ​​from the actual sensor measurements obtained in step S6.

[0108] Equivalent output observation matrix Through the comprehensive observation model At the identity element of the Lie group, the Lie algebra error column vector is defined according to step S3. The mathematical essence of the analytic partial derivative mapping matrix obtained by linearizing through Lie algebraic perturbations is defined as follows:

[0109] ;

[0110] The microprocessor directly stores the equivalent output observation matrix in memory. The parsing expands and instantiates the data into the following block matrix form:

[0111] ;

[0112] In the above matrix, This indicates that the local world system reference geomagnetic vector pre-stored in step S6 will be used. The mapping is transformed into an antisymmetric matrix to extract the contribution of attitude error to geomagnetic observations; This represents a 1×3 dimensional matrix used to extract the vertical observation components from the position error.

[0113] Based on the constructed isotropic output observation matrix The microprocessor executes the Kalman gain matrix. The calculation formula is as follows:

[0114] ;

[0115] ;

[0116] In the formula, This represents the current time step calculated by the microprocessor. The new information covariance matrix; Indicates the current time step Equivalent output observation matrix The matrix obtained by performing a transpose operation; This represents the diagonal matrix of the overall measurement noise covariance, which includes the noise characteristics of barometer and magnetometer measurements and is pre-stored by the microprocessor. This represents the current time step obtained from the final derivation and calculation. The Kalman gain matrix; Represents the covariance matrix of the new information The inverse matrix obtained by performing matrix inversion. This is the gain matrix. The optimal proportional relationship between the prior error covariance and the observation uncertainty was established, which is the core gain operator for the subsequent steps of solving the Lie algebra error column vector.

[0117] Step S8: Based on the Kalman gain matrix calculated in step S7 and the reduced-order integrated observation innovation column vector constructed in step S6, solve for the Lie algebra error column vector in the tangent space.

[0118] Furthermore, S8 specifically includes:

[0119] The microprocessor will calculate the current time step obtained in step S7. Kalman gain matrix As a linear mapping operator, it is applied to the current time step constructed in step S6. Reduced-order comprehensive observation new information column vector Above, through matrix multiplication, the Lie algebraic error column vector in the current time step tangent space is derived and calculated. :

[0120] ;

[0121] In the formula, This represents the current time step obtained from the solution. The Lie algebra error column vector in the tangent space represents the amount of correction by the observation space information to the prior estimated state; , , These represent the attitude, velocity, and position components of the Lie algebra error column vector in the current time step cut space. Through this step, the observation innovation column vector is... The information was successfully mapped and transformed into a vector increment that corrects the current filtering state.

[0122] Step S9: Based on the Lie algebra error column vector calculated in step S8, perform manifold posterior update on the prior estimated state in step S4 and output it.

[0123] Furthermore, S9 specifically includes:

[0124] The microprocessor utilizes the attitude, velocity, and position components of the Lie algebra error column vector in the current time step space, calculated in step S8. For the current time step obtained in step S4 The prior estimated state is corrected posteriorly on the manifold tangent space; simultaneously, based on the current time step calculated in step S7... Kalman gain matrix and equivalent output observation matrix For the current time step obtained in step S5 Prior group error covariance matrix Perform a closed-loop update; the specific update formula is as follows:

[0125] ;

[0126] ;

[0127] ;

[0128] ;

[0129] In the formula, the superscript " "The unified representation indicates that the updated post-hoc state has been completed;" , , These represent the current time step obtained after posterior correction in the tangent space of the manifold. Posterior estimated attitude rotation matrix, posterior estimated velocity column vector, and posterior estimated position column vector; superscript " "The unified representation represents the prior prediction state before any observation updates are performed;" , , These are the prior estimated states output in step S4, determined by the current time step. The prior estimated attitude rotation matrix, the prior estimated velocity column vector, and the prior estimated position column vector are obtained by integral calculation; To complete the update and obtain the current time step The posterior group error covariance matrix; The current time step in step S5 The prior group error covariance matrix; for 3D identity matrix The current time step calculated in step S7 The Kalman gain matrix; The current time step constructed in step S7 The equivariant output observation matrix; It is an exponential mapping function; This represents an operator that maps a three-dimensional vector to an antisymmetric matrix.

[0130] Finally, the microprocessor calculates the aforementioned posterior estimated pose rotation matrix. and posterior estimated position column vector As the optimal estimated data, it is continuously output to the cascaded loop of the underlying flight control microprocessor to ensure the stability of the catapult-launched UAV's attitude when transitioning from passive flight to controlled flight.

[0131] Example 2: Simulation Comparison and Verification of Passive Flight Phase Filtering Algorithm

[0132] To verify the effectiveness of the algorithm of this invention in the passive flight phase of UAV ejection (the superposition of pure inertial blind thrust and extreme non-gravity overload), a software-in-the-loop (SITL) system-level simulation comparison was conducted on the MATLAB / Simulink platform.

[0133] 1. Experimental conditions:

[0134] The tested UAV was set to launch from a physically constrained guide rail with an initial velocity of 25 m / s and an elevation angle of 60°. To accurately reproduce the physical sensor characteristics of the microelectromechanical system (MEMS), Gaussian white noise and random walk were injected into the model: the gyroscope noise density was set to... The accelerometer noise density is set to And set a fixed zero bias for the three-axis startup (e.g., the zero bias for the accelerometer startup is 1). The control group used the industry-standard built-in traditional extended Kalman filter attitude and heading reference system (AHRS / EKF).

[0135] 2. Comparative Analysis of Results:

[0136] To ensure the rigor of the algorithm comparison, this simulation is configured with... The pre-launch preparation phase is defined as starting at time 2.0, representing the initial moment of the passive flight phase. During this phase, the UAV remains stationary and constrained on the launch pad. Figure 2 The data from the first two seconds shows that the traditional EKF algorithm in the control group achieved high-precision initial alignment in a quiet environment (the attitude error approached 0.5%). This eliminates the interference of initial pose differences on subsequent impact tests, ensuring that the comparative experiments are conducted on a "same starting line." Combined with... Figure 3It can be seen that the attitude estimation curve of LIEKF proposed in this invention perfectly coincides with the real static attitude reference. Although the physical noise of the real low-cost MEMS gyroscope is added, the extremely small random walk drift is almost invisible under macroscopic comparison, ensuring extremely high initial alignment accuracy.

[0137] Combination Figure 2 It can be seen that, in When the initial strong impact triggers the passive flight phase of the ejection, the horizontal observation model of the traditional EKF algorithm becomes severely distorted. Because the intense nonlinear impact exceeds the linearization assumptions of the traditional first-order Taylor expansion, it incorrectly couples a large non-gravitational deceleration into the gravity vector calculation, leading to singularity flips and severe zigzag divergence in the calculated three-axis attitude matrix (e.g., ...). Figure 2 The estimated pitch and yaw angles fluctuate violently and deviate from the true trajectory. In actual engineering, this false attitude will directly cause the flight control system to output full control commands, resulting in the UAV rolling uncontrollably and crashing.

[0138] Combination Figure 3 As can be seen, under identical inputs of strong noise and extremely high overload, the solution method (LIEKF) of this invention successfully isolates the negative updates of the accelerometer caused by overload contamination by utilizing the isovariant architecture and high-frequency autonomous integrals on the group manifold. When faced with extreme physical impacts, its three-axis attitude tracking errors (roll, pitch, and yaw) only exhibit small transient fluctuations and converge rapidly and smoothly after the impact, without any manifold rupture or overshoot divergence, demonstrating extremely strong geometric robustness against impacts.

[0139] Combination Figure 4 It can be seen that, thanks to the aforementioned highly robust attitude handover and transient disturbance rejection mechanisms, in the passive flight phase of catapult launch (2.0s-5.0s) where no external GPS assistance is provided and calculations rely solely on pure inertial estimation, the three-dimensional position calculated by the method of this invention (…) Figure 4 The solid line in the middle closely matches the actual physical trajectory of a gliding ballistic missile. Figure 4 (middle dashed line).

[0140] It is particularly important to note that, due to the physical characteristics of low-cost MEMS inertial sensors for micro-UAVs, prolonged pure inertial calculations inevitably result in integral divergence. However, the passive flight phase of a real catapult-launched UAV is an extremely short transient process. This invention specifically captures the 3-second golden window after catapult launch (i.e., the critical stage where the UAV glides to the highest point of the parabola using initial kinetic energy, preparing to unlock and deploy its arms) for simulation, achieving accurate and convergent 3D position and attitude state estimation. From a kinematic calculation perspective, it effectively avoids the cascading propagation of gravity projection errors and perfectly overcomes the "high overload blind zone" where traditional algorithms are prone to failure. This provides continuous and extremely reliable initial state data support for the UAV's subsequent successful arm deployment, flight control power system intervention, and smooth entry into controlled flight.

[0141] Furthermore, this invention can be used in field experiments, where a catapult launcher is used to construct physical constraint rails with different launch elevation angles (elevation angles range from [30°, 90°]), such as... Figures 5-10 As shown, the catapult launcher includes: a horizontal support base, a rotating bracket, a gas cylinder 2, a launch profile bracket 3, a launch tube 4, a planar cross-connecting plate 7, a launch tube connecting cover 11, a steering seat 14, a steering profile base frame 15, a gas valve 22, a solenoid valve 19, and a piston 26. The vertically aligned launch profile base frame 21 and the horizontally aligned launch profile base frame 17 are connected by a 90° angle aluminum 9 to form the horizontal support base. The steering profile base frame 15 is mounted on the horizontal support base via the 90° angle aluminum 9, and two sets of parallel launch profile brackets 3 are mounted via connectors. The planar cross-connecting plate 7 is installed at the end of the launch profile bracket 3 away from the horizontal support base. Two spaced-apart steering seats 14 are mounted on the steering profile base frame 15. The steering seats 14 serve as the core fulcrum, supporting the rotating bracket and the aerodynamic storage assembly on the rotating bracket for pitching motion (aerodynamic force). The storage assembly includes a gas cylinder 2, a launch tube 4, a launch tube connecting cover 11, a gas valve 22, a solenoid valve 19, a stainless steel bellows 20, a launch tube base 12, and a piston 26. The first end of the rotating bracket is rotatably engaged with the steering seat 14, and the second end of the rotating bracket is engaged with the planar cross-connecting plate 7 at different positions to adjust the angle of the rotating bracket. The third end of the rotating bracket is equipped with the gas cylinder 2 and the launch tube 4 (the launch tube 4 can be a carbon fiber launch tube). The outlet of the gas cylinder 2 is connected to the tube body (such as using a stainless steel bellows 20 as the tube body) through the gas valve 22 and the solenoid valve 19. The output end of the stainless steel bellows 20 is connected to the threaded interface of the launch tube connecting cover 11 at the end of the launch tube 4. The launch tube 4 contains a piston 26, and gas is filled into the carbon fiber launch tube space between the launch tube connecting cover 11 and the piston 26 through the gas cylinder 2 via the launch tube connecting cover 11.

[0142] Furthermore, the rotating bracket includes a rotating profile rod 5 and a pneumatic storage and release bracket; one end of the two parallel rotating profile rods 5 serves as the first end of the rotating bracket for hinged connection with the steering seat 14; the other end of the two parallel rotating profile rods 5 serves as the second end of the rotating bracket; and a pneumatic storage and release bracket perpendicular to the rotating profile rods 5 is provided on one side of the two parallel rotating profile rods 5 as the third end of the rotating bracket.

[0143] Furthermore, the pneumatic storage and release support includes a fixed upper plate, a gas cylinder profile frame 13, a lower gas cylinder fixed plate 23, a short rod 24 of the launch tube profile base frame, and a long rod 25 of the launch tube profile base frame. One end of each of the two long rods 25 of the launch tube profile base frame is fixed to a rotating profile rod 5, and the other end of each long rod 25 of the launch tube profile base frame is fixed to the end of a short rod 24 of the launch tube profile base frame. A gas cylinder profile frame 13 is installed perpendicular to the long rod 25 of the launch tube profile base frame on the side of the long rod 25 of the launch tube profile base frame away from the horizontal support base. A fixed upper plate is installed on the end of the gas cylinder profile frame 13 away from the long rod 25 of the launch tube profile base frame (the fixed upper plate includes a lower gas cylinder fixed plate 1, a lower gas cylinder fixed plate 23, and a launch tube profile base frame 24). The cylinder fixing plate 8) and the lower plate 23 of the cylinder fixing plate are installed at the end of the long rod 25 of the launch tube profile base near the horizontal support base and away from the rotating profile rod 5. The gas cylinder 2 passes through the upper plate 1 and the lower plate 23 of the cylinder fixing plate and is limited by the through hole on the lower plate 23. The launch tube 4 passes through the launch tube fixing plate 8 and the launch tube connecting cover 11 is glued to the side of the launch tube 4 near the horizontal support base. The launch tube connecting cover 11 is supported by the launch tube base 12 fixed on the long rod 25 of the launch tube profile base. The side of the launch tube 4 away from the horizontal support base is designed to be open. The launch tube base 12 adopts a cylindrical structure and is used to embed the launch tube connecting cover 11.

[0144] Furthermore, each set of launching profile brackets 3 has the same structure, including two launching profile rods and a 60° sharp-angle side connecting plate 6. One end of the two launching profile rods arranged at an angle is connected by the 60° sharp-angle side connecting plate 6, and the other end of the two launching profile rods is connected to the launching frame profile longitudinal base 21 in the horizontal support base through a connector to form an angle. The connector includes a 120° angle aluminum 10, a 60° angle aluminum 16, and a 60° side connecting plate 18.

[0145] The aforementioned connection nodes are reinforced with 90° angled aluminum 9, 120° angled aluminum 10, and 60° angled aluminum 16, which can withstand the strong vibrations generated by the instantaneous release of high-pressure gas and ensure that the frame does not twist or deform during launch. The 60° sharp-angled side connecting plate 6, the planar cross connecting plate 7, and the 60° side connecting plate 18 are bolted to the profile, primarily serving to support and stabilize the launch angle. The catapult launcher has fixed angle positions of 30°, 45°, and 90°. By pre-setting locking holes at corresponding positions on the rotating profile rod 5 and bolting them into the holes on the planar cross connecting plate 7, the launch mechanism can be quickly locked at commonly used angles, meeting the rapid use requirements in different operating environments. The launch tube 4 is equipped with a vertically sliding piston 26, which divides the interior of the tube into a lower gas filling area and an upper UAV placement area. This piston-driven method effectively prevents high-pressure gas flow from directly impacting the bottom of the UAV, ensuring a uniform distribution of launch force.

[0146] Furthermore, at least one set of O-rings is fitted around the outer periphery of the piston 26 to enhance the airtightness between the piston and the inner wall of the launch tube 4; combined with the smooth inner wall of the tube, it can generate greater thrust at lower air pressure, ensuring high efficiency of energy conversion.

[0147] Furthermore, a wireless remote control switch control box and a remote control switch are provided to control the on / off state of the solenoid valve 19, enabling remote activation of the transmitter to protect the operator's safety.

[0148] Furthermore, the launch tube 4 can be replaced with different inner diameter sizes. Simply replace the launch tube connecting cover 11 and the launch tube base 12 accordingly. This allows for the provision of more replacement parts to adapt to different ejection requirements.

[0149] Based on experiments and relevant data, the installation and implementation process of this invention is described below:

[0150] I. Installation and Debugging Process

[0151] 1. Main framework construction: Based on Figures 5 to 7 As shown, a horizontal support base is first constructed using industrial aluminum profiles. The longitudinal base frame 21 and the transverse base frame 17 of the launcher profile are placed at a 90° angle and secured with 90° angle aluminum 9 and bolts. A steering seat 14 is installed on the steering profile base frame 15 fixed to the horizontal support base, ensuring flexible and unhindered pitch and rotation. The launcher profile bracket is then installed via connectors and a 60° angled side connecting plate 6; a planar cross connecting plate 7 is fixed to the launcher profile bracket.

[0152] 2. Assembly of the pneumatic storage assembly: Place the gas cylinder 2 in the upper plate 1 of the gas cylinder fixing plate, and use the upper plate 1 and the lower plate 23 of the gas cylinder fixing plate to form a clamping structure and lock it. Screw the manual air valve 22 and the solenoid valve 19 sequentially into the gas cylinder outlet. The solenoid valve 19 is reliably connected to the air inlet of the launch tube connecting cover 11 through the rigid stainless steel bellows 20 to ensure that the pipeline does not shift under the impact of high-pressure airflow. After applying an appropriate amount of lubricating silicone grease to the piston 26, place it inside the launch tube 4 and check whether the O-ring on the piston 26 is tightly fitted to the tube wall. Fix the launch tube 4 to the third end of the rotating bracket through the launch tube connecting cover 11 and the launch tube base 12, and ensure that the rotating profile rod 5 and the short rod 24 of the launch tube profile base frame are firmly connected to the long rod 25 of the launch tube profile base frame in the rotating bracket to provide sufficient structural rigidity; hinge the second end of the rotating bracket to the steering seat 14.

[0153] 4. Rotation bracket angle preset and locking: The rotating profile rod 5 has standard positioning holes for forming 30°, 45° and 90° launch elevation angles as locking holes. During debugging, the bolt can be inserted into the corresponding locking hole through the planar cross connection plate 7 and locked quickly by the T-nut; or it can be freely rotated to the required angle within the range of 30° to 90° and then locked by fasteners.

[0154] II. Implementation and Launch Process

[0155] 1. Preparation Phase: Check the inner wall of launch tube 4 for foreign objects. Position the catapult-launched drone with its arms retracted into a "bullet" shape inside the carbon fiber launch tube, ensuring its tail is stably supported on the flat platform of piston 26. At this point, piston 26 divides the tube into a lower gas filling area and an upper drone placement area.

[0156] 2. Filling Stage: Open the manual air valve 22 to fill the gas cylinder 2 with compressed gas using an external air source or a pre-stored air source. The pressure is monitored in real time by a pressure gauge integrated into the air circuit to prevent exceeding the device's limits. Simultaneously, a toggle switch on the air valve allows for tightening during filling to ensure gas is supplied to the cylinder; the valve is loosened only when launch is required.

[0157] 3. Remote Launch: The operator first loosens the gas valve and checks for leaks. If there are no problems, the operator retreats to a safe area and issues the opening / closing command for the solenoid valve 19 via the wireless remote control switch. After the coil of the solenoid valve 19 is energized, the valve core opens instantly, and high-pressure gas is injected into the gas filling area through the stainless steel bellows 20. The thrust generated by the gas expansion acts evenly on the bottom of the piston 26, driving the piston 26 to propel the drone at high speed along the internal track.

[0158] 4. Takeoff transition: The UAV gains initial momentum the moment it leaves the nozzle, and then its arms quickly extend and lock under the action of aerodynamics, realizing the transition from the passive flight phase to the stable and controlled flight phase.

[0159] In practical use, the catapult-launched drone is loaded into the preset launch elevation angle of the catapult-launched drone's pneumatic launch device. In the lower launch tube 4, the system enters the preparation and inflation phase. After inflation is complete, it enters the remote launch phase. When the solenoid valve 19 is energized, the valve core opens instantaneously. Simultaneously, the method of this invention running in the flight control microprocessor estimates the state of the catapult-launched UAV during its passive flight phase after leaving the launch chamber. The catapult-launched UAV can be any existing catapult-launched aircraft capable of folding and unfolding.

[0160] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A group-invariant filtering state estimation method for the passive flight phase of a catapult-launched unmanned aerial vehicle, characterized in that, include: Step S1: Obtain initial physical constraint parameters; wherein, the physical constraint parameters include the launch elevation angle of the catapult launcher and the initial launch velocity; Step S2: Construct the real attitude rotation matrix from the machine system to the world system and the real velocity column vector in the world system based on the initial physical constraint parameters. Establish the state matrix of the UAV on the group manifold based on the real attitude rotation matrix, the real velocity column vector in the world system and the real position column vector. Step S3: Based on the state matrix of the UAV on the group manifold, define the group invariant error form of the UAV state as a left invariant error matrix, and expand the left invariant error matrix on the extended attitude Lie group; through the logarithmic mapping relationship from Lie group to Lie algebra, map the expansion of the left invariant error matrix on the extended attitude Lie group into a 9-dimensional Lie algebra error column vector; based on the 9-dimensional Lie algebra error column vector, establish the derivation relationship between the state matrix of the UAV on the group manifold and the estimated state matrix; then, initialize the group error covariance matrix. Step S4: Based on the state matrix of the UAV on the group manifold and the acquired inertial measurement unit data, construct a state transition model to obtain the prior estimated state at the current time step. Step S5: Based on the initialized group error covariance matrix and the inertial measurement unit data, recursively calculate the prior group error covariance matrix for the current time step. Step S6: Execute the order reduction strategy of dynamically shielding the force observation channel to construct a reduced-order comprehensive observation information column vector based on prior estimated state, air pressure and geomagnetic measurement data; Step S7: Construct the uniform output observation matrix; calculate the Kalman gain matrix for the current time step based on the uniform output observation matrix and the prior group error covariance matrix; Step S8: Solve for the Lie algebra error column vector in the tangent space based on the Kalman gain matrix and the reduced-order synthetic observation innovation column vector; Step S9: Based on the solved Lie algebra error column vector, perform manifold posterior update on the prior estimated state and output it.

2. The group-invariant filtering state estimation method for the passive flight phase of a catapult-launched UAV according to claim 1, characterized in that, Specifically, S4 is: At the initial moment of the passive flight phase, the state matrix of the established UAV on the group manifold is extracted. The true attitude rotation matrix, true velocity column vector, and true position column vector in the matrix are assigned values ​​as the initial posterior estimated attitude rotation matrix. Posterior estimation of velocity column vector and posterior estimated position column vector Using this as the starting point for iteration, during the passive flight phase, inertial measurement unit data is acquired, a state transition model is constructed, and the prior estimate of the state at the current time step is obtained: ； ； ； In the formula, , and These represent the time steps from the current time step. The prior estimated attitude rotation matrix, the prior estimated velocity column vector, and the prior estimated position column vector are obtained through integral calculation; , and Representing time steps The posterior estimated attitude rotation matrix, the posterior estimated velocity column vector, and the posterior estimated position column vector; It is an exponential mapping function; Indicates the current time step The column vector of the machine's angular velocity, actually measured by the gyroscope in the inertial measurement unit; Indicates the current time step The mechanical system specific force column vector obtained from actual measurements by the accelerometers in the inertial measurement unit; This represents a column vector of world-system gravitational constants pre-stored in the microprocessor; This indicates the discrete time interval for a single sampling by the inertial measurement unit. This represents an operator that maps a three-dimensional vector to an antisymmetric matrix.

3. The group-invariant filtering state estimation method for the passive flight phase of a catapult-launched UAV according to claim 1, characterized in that, Specifically, S5 is: Using the group error covariance matrix initialized in step S3 The posterior group error covariance matrix at the initial moment of the passive flight phase Combining the inertial measurement unit data obtained in step S4, the prior group error covariance matrix for the current time step is recursively calculated. The expression is as follows: ； ； ； In the formula, This represents the calculated current time step. The prior group error covariance matrix; Indicates time step The posterior group error covariance matrix; For the current time step The discretized state transition matrix; Indicates transpose; This represents the predefined system process noise covariance matrix; This indicates the discrete time interval for a single sampling by the inertial measurement unit. for An identity matrix of dimension 1 express An identity matrix of dimension 1 express A zero matrix of dimension; For the current time step The continuous-time systematic error adjoint matrix, This represents an operator that maps a three-dimensional vector to an antisymmetric matrix. Indicates time step The posterior estimate of the attitude rotation matrix; Indicates the current time step The column vector of the machine's angular velocity, actually measured by the gyroscope in the inertial measurement unit; Indicates the current time step The mechanical system specific force column vector obtained by actual measurement from the accelerometer in the inertial measurement unit.

4. The group-invariant filtering state estimation method for the passive flight phase of a catapult-launched UAV according to claim 1, characterized in that, Specifically, S6 is: Extract the current time step obtained in step S4 The prior estimated position column vector in the prior estimated state Compared with prior estimated attitude rotation matrix And combined with the current time step Barometer altitude measurements acquired in real time from the inertial measurement unit With the three-axis geomagnetic vector measurement value of the magnetometer A reduced-order comprehensive observation novel column vector is constructed. : ； In the formula, Indicates the current time step The resulting reduced-order synthetic observation information column vector; This represents the calculated current time step. Barometer height residual scalar; This represents the calculated current time step. Three-axis geomagnetic observation residual column vector; Represents the column vector of prior estimated locations Vertical extracted Axial direction component; Indicates transpose; This represents the pre-stored local world-system reference geomagnetic vector.

5. The group-invariant filtering state estimation method for the passive flight phase of a catapult-launched unmanned aerial vehicle according to claim 1, characterized in that, Specifically, S7 is: Constructing a nonlinear system-level integrated observation model Equivalent output observation matrix Through the comprehensive observation model At the identity element of the Lie group, the analytical partial derivative mapping matrix obtained by linearizing the Lie algebraic perturbation based on the defined Lie algebraic error column vector is as follows: ； Based on the constructed isotropic output observation matrix The Kalman gain matrix is ​​calculated using the following formula: ； ； In the formula, Indicates the current time step The new information covariance matrix; This represents a pre-stored diagonal matrix of the overall measurement noise covariance, which includes the noise characteristics of barometer and magnetometer measurements. Indicates the current time step The Kalman gain matrix; Indicates the current time step The prior group error covariance matrix; This represents the pre-stored local world-system reference geomagnetic vector; Indicates transpose; express A zero matrix of dimension 1 express The zero matrix; [0,0,1] represents a 1×3 dimensional matrix; This represents an operator that maps a three-dimensional vector to an antisymmetric matrix; express The inverse matrix.

6. The group-invariant filtering state estimation method for the passive flight phase of a catapult-launched unmanned aerial vehicle according to claim 1, characterized in that, Specifically, S8 is: The current time step calculated in step S7 Kalman gain matrix As a linear mapping operator, it is applied to the current time step constructed in step S6. Reduced-order comprehensive observation new information column vector Above, through matrix multiplication, the Lie algebraic error column vector in the current time step tangent space is derived and calculated. : ； In the formula, , , These are the components of the attitude, velocity, and position dimensions in the Lie algebra error column vector at the current time step.

7. The group-invariant filtering state estimation method for the passive flight phase of a catapult-launched unmanned aerial vehicle according to claim 1, characterized in that, Specifically, S9 is: Using the components of attitude, velocity, and position in the Lie algebra error column vector in the current time step space calculated in step S8 For the current time step obtained in step S4 The prior estimated state is corrected posteriorly on the manifold tangent space; simultaneously, based on the current time step calculated in step S7... Kalman gain matrix and equivalent output observation matrix For the current time step obtained in step S5 Prior group error covariance matrix Perform a closed-loop update; the specific update formula is as follows: ； ； ； ； In the formula, , , These represent the current time step obtained after posterior correction in the tangent space of the manifold. Posterior estimation of attitude rotation matrix, posterior estimation of velocity column vector, and posterior estimation of position column vector; prior estimation of state... , and These represent the time steps from the current time step. The prior estimated attitude rotation matrix, the prior estimated velocity column vector, and the prior estimated position column vector are obtained through integral calculation; To complete the update and obtain the current time step Posterior group error covariance matrix; for An identity matrix of dimensionality; It is an exponential mapping function; This represents an operator that maps a three-dimensional vector to an antisymmetric matrix.

8. A group-invariant filtering state estimation system for the passive flight phase of a catapult-launched unmanned aerial vehicle, characterized in that, The module includes the group-invariant filtering state estimation method for the passive flight phase of a catapult-launched unmanned aerial vehicle as described in any one of claims 1-7.

9. The group-invariant filtering state estimation system for the passive flight phase of a catapult-launched unmanned aerial vehicle according to claim 8, characterized in that, The catapult launcher serves as a physical constraint guide for launching UAVs at different launch elevation angles. The catapult launcher includes: a horizontal support base, a rotating bracket, a gas cylinder, a launch profile bracket, a launch tube, a planar cross-connecting plate, a launch tube connecting cover, a steering seat, a steering profile base frame, a gas valve, a solenoid valve, and a piston. The horizontal support base is equipped with the steering profile base frame and two sets of parallel-arranged launch profile brackets. The planar cross-connecting plate is installed at the end of the launch profile bracket away from the horizontal support base. Two spaced-apart steering seats are installed on the steering profile base frame. The first end of the rotating bracket rotatably engages with the steering seat, and the second end of the rotating bracket engages with the planar cross-connecting plate at different positions to adjust the angle of the rotating bracket. The third end of the rotating bracket is equipped with a gas cylinder and a launch tube. The gas cylinder's outlet is connected to a tube body via a gas valve and a solenoid valve. The output end of the tube body is connected to the interface of the launch tube connecting cover at the end of the launch tube. A piston is located inside the launch tube, and gas is filled into the carbon fiber launch tube space between the launch tube connecting cover and the piston via the gas cylinder.

10. The group-invariant filtering state estimation system for the passive flight phase of a catapult-launched unmanned aerial vehicle according to claim 8, characterized in that, The rotating support includes rotating profile rods and a pneumatic storage and release support; one end of the two parallel rotating profile rods serves as the first end of the rotating support for hinge connection with the steering seat; the other end of the two parallel rotating profile rods serves as the second end of the rotating support; and a pneumatic storage and release support perpendicular to the rotating profile rods is provided on one side of the two parallel rotating profile rods as the third end of the rotating support.

Images