A spatial non-cooperative tumbling target parameter estimation method based on adaptive hybrid Kalman filter

By adopting an adaptive hybrid Kalman filtering method that combines extended Kalman filtering and unscented Kalman filtering, the problem of high accuracy and real-time performance in attitude estimation of non-cooperative tumbling targets in space is solved. This method achieves robustness and algorithm convergence in complex space environments, supporting precise capture by space robotic arms.

CN123115618APending Publication Date: 2026-11-17SHENYANG INST OF AUTOMATION - CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611612881.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing filtering methods struggle to simultaneously achieve both high accuracy and real-time performance in attitude estimation for non-cooperative rolling targets in space, and lack robustness to time-varying noise and theoretical guarantees of convergence, thus failing to meet the requirements of real-time on-orbit missions.

Method used

An adaptive hybrid Kalman filtering method is adopted, which combines extended Kalman filtering and unscented Kalman filtering. Through adaptive switching strategy and time-varying noise processing, a 13-dimensional state vector and linear observation equation are constructed to achieve real-time estimation of attitude quaternions, three-axis angular velocities and inertia parameters.

Benefits of technology

It effectively balances attitude estimation accuracy and computational efficiency, improves robustness in complex space environments, ensures algorithm convergence, is suitable for rigid tumbling targets rotating around arbitrary axes, and supports precise capture by space robotic arms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115618A_ABST
    Figure CN123115618A_ABST
Patent Text Reader

Abstract

The application provides a space non-cooperative tumbling target parameter estimation method based on adaptive hybrid Kalman filtering, and relates to the fields of aerospace engineering and space robot technology. First, an assumption is proposed, attitude kinematics differential equations and attitude dynamics equations without external torque are derived, and then a basic filtering algorithm is designed: the prediction-update iteration process of extended Kalman filter (EKF) and unscented Kalman filter (UKF) is realized respectively, a dynamic switching strategy based on the trace of error covariance matrix is proposed, the filtering algorithm is adaptively selected according to the current estimation error, and the reference covariance matrix is updated to accelerate the convergence when EKF is continuously executed; a forgetting factor is introduced to limit the memory length of the algorithm, and corresponding noise statistical characteristic real-time update formulas are designed for EKF and UKF respectively, the mean and covariance matrix of process noise and observation noise are estimated in the online iteration process, and the time-varying noise caused by non-uniform illumination and debris interference in the space environment is inhibited.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of aerospace engineering and space robotics, and in particular to a method for estimating parameters of a space non-cooperative tumbling target based on adaptive hybrid Kalman filtering. Background Technology

[0002] The development of space non-cooperative target acquisition technology has made the estimation of attitude motion parameters of tumbling targets a core technology. Limited by onboard computing resources, high-precision real-time estimation has become a critical prerequisite for engineering applications. The accuracy of attitude estimation directly determines the success rate and safety of space robotic arm acquisition.

[0003] Existing filtering methods struggle to balance estimation accuracy and computational efficiency, exhibiting large errors or poor real-time performance under strong nonlinearity and lacking rigorous convergence theory guarantees. Consequently, they fail to meet the requirements of real-time on-orbit missions. Therefore, there is an urgent need to develop efficient and reliable attitude estimation algorithms to ensure the smooth implementation of non-cooperative target operations and support the engineering applications of aerospace technology. Summary of the Invention

[0004] To address the shortcomings of existing technologies, a parameter estimation method for space non-cooperative tumbling targets based on adaptive hybrid Kalman filtering is proposed. This method overcomes the deficiencies of existing space non-cooperative tumbling target attitude estimation methods, which struggle to balance high accuracy and real-time performance, have poor robustness to time-varying noise, and lack theoretical guarantees of convergence. It effectively solves the problems of large estimation errors, high computation time, and interference from the complex space environment under strong nonlinearity.

[0005] On one hand, this invention provides a method for estimating the parameters of a space non-cooperative tumbling target based on adaptive hybrid Kalman filtering, comprising the following steps:

[0006] Step 1: Mathematical modeling of the tumbling target: Propose hypotheses, derive the differential equations of attitude kinematics and the attitude dynamics equations without external torque, construct a 13-dimensional state vector containing attitude quaternions, three-axis angular velocities and inertia parameters, and establish nonlinear discrete state space equations and linear observation equations.

[0007] The specific assumption is as follows:

[0008] (1) Ignore the influence of gravity and do not consider orbital dynamics;

[0009] (2) The non-cooperative tumbling target is regarded as a cube-shaped decommissioned satellite with two solar panels and one antenna, whose center of mass CM is located at the geometric center and rotates around its arbitrary axis with a certain initial velocity;

[0010] (3) The rolling target is in a free-floating state, and the external force or torque is negligible;

[0011] (4) The rolling target is rigid;

[0012] (5) The system noise caused by the complex space environment is time-varying Gaussian noise.

[0013] Step 1.1: Define the inertial reference coordinate system and the target body coordinate system, use unit quaternions to represent the target attitude, and derive the attitude kinematic differential equations: ;

[0014] in For attitude quaternions, It is the time rate of change of the attitude quaternion; This is the antisymmetric matrix corresponding to the angular velocity. The three-axis angular velocities in the target's body coordinate system;

[0015] Step 1.2: Based on the angular momentum theorem and the assumption that the target is freely floating without external torque, the attitude dynamics equations without external torque are derived: ;

[0016] in, The angular acceleration vector of the rolling target. For the rolling target about inertia;

[0017] Step 1.3: Using the six independent elements of the inertia matrix—namely, the three principal inertia and the three inertia products—as parameters to be estimated, construct a 13-dimensional state vector: ;

[0018] in, It is an inertial variable;

[0019] Step 1.4: Combine the attitude kinematics differential equations, the attitude dynamics equations without external torques, and the invariant inertia parameter to construct continuous state-space equations: ;

[0020] in, For continuous state change rate, These are the state components corresponding to the attitude quaternions. The state component corresponding to the angular velocity. These are the state components corresponding to the inertia. This includes modeling errors that incorporate the motion disturbance of space targets by small space debris and unstable airflow in outer space.

[0021] Step 1.5: Discretize the continuous state equations using the Euler method to obtain the nonlinear discrete state-space equations:

[0022] ;

[0023] in, Let k be the system state vector at the k-th discrete time. , Let Δt be the continuous state change rate at the k-th discrete time, and Δt be the filter iteration step size. The noise in the discrete process at time k;

[0024] Step 1.6: Based on the characteristic that attitude quaternions, angular velocity, and inertia parameters can be directly measured by sensors, construct the linear observation equation:

[0025] ;

[0026] Z tk This is the observation vector of the system over continuous time, containing information on the target's attitude, angular velocity, and inertia measured by sensors. It is a 13th order identity matrix. The measurement noise at the k-th discrete time;

[0027] Step 2: Basic filtering algorithm design: Implement the prediction-update iteration process of the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF) respectively, so as to obtain the posterior estimate of the state and the posterior estimate of the error covariance at the current time.

[0028] The prediction-update iteration process of the Extended Kalman Filter (EKF) includes the following steps:

[0029] Step A2.1: Set the initial state posterior estimate and the initial error covariance matrix Define process noise and observation noise For zero-mean Gaussian white noise, set its covariance matrix respectively. and ;

[0030] Step A2.2: Linearize the discrete-time state transition equation of the nonlinear system as follows: Among them, by calculating the nonlinear state transition function The posterior estimate at the previous time step Jacobian matrix at the location We obtain the state transition matrix from step k to step k+1. Based on the posterior estimate from the previous time step, calculate the prior state prediction for the current time step. Sum of error covariance prior predicted values ;

[0031] Step A2.3: Calculate the Kalman gain matrix: ;

[0032] in The observation matrix;

[0033] Combined with the observations at the current moment The update yields the posterior estimate of the state at the current time step. ; Posterior estimate of error covariance ,in It is the Kalman gain matrix of the k-th iteration, and the discrete observation function. Let be the prior error covariance matrix at step k;

[0034] The prediction-update iteration process of the Unscented Kalman Filter (UKF) includes the following steps:

[0035] Step B2.1: Initialization: Configure unscented transform scaling parameters κ is the second scaling parameter, set to 0. This is a parameter tuning parameter specifically used in the unscented transform to utilize prior statistical information about the state distribution. Setting it to 0 allows for the calculation of the scaling factor. and Sigma point weights:

[0036] ;

[0037] Where L is the dimension of the state vector. Let be the mean weight corresponding to the i-th Sigma point. Let be the covariance weight corresponding to the i-th Sigma point;

[0038] Step B2.2: Unscented Transformation and State Prediction: Based on the current state estimate and error covariance matrix, generate 2L+1 Sigma points:

[0039] ;

[0040] in, Let be the i-th Sigma sampling point generated by the unscented transform. State vector The mean, Let be the error covariance matrix of the state vector. i This indicates retrieving the i-th row of the matrix within the parentheses;

[0041] Substitute each Sigma point into the nonlinear state transition function. Propagate the data to obtain the predicted value of the i-th Sigma point after one state transition; calculate the prior state prediction value. Sum of error covariance prior predicted values ;

[0042] Step B2.3: Observation Prediction: Substitute the propagated Sigma points into the observation function. ; Calculate prior predictions based on observations Observation error covariance matrix and the cross-correlation matrix between state and observation ;

[0043] Step B2.4: Combine the observations at the current moment Update to obtain the posterior estimate of the state at the current time. Posterior estimate of error covariance ;

[0044] Step 3: Hybrid switching rule design: A dynamic switching strategy based on the trace of the error covariance matrix is ​​proposed. The hybrid threshold coefficient and the reference covariance matrix are set. The filtering algorithm is adaptively selected according to the current estimation error. The reference covariance matrix is ​​updated when EKF is executed continuously to speed up convergence.

[0045] Step 3.1: Preset the core parameters of the mixing strategy: Set the mixing threshold coefficient Used to control the execution ratio of EKF and UKF; initializes the reference covariance matrix. This serves as the benchmark threshold for judging the magnitude of the estimation error.

[0046] Step 3.2: Real-time calculation of the overall estimation error index: In each discrete period of filtering iteration, calculate the posterior error covariance matrix at the current time. traces This is used to quantify the overall error level of the current 13-dimensional state estimation;

[0047] Step 3.3: Adaptive algorithm switching based on error threshold:

[0048] if The prediction-update iteration process of the Unscented Kalman Filter (UKF) is executed.

[0049] if The prediction-update iteration process of the Extended Kalman Filter (EKF) is executed.

[0050] Step 3.4: Reference Covariance Update During Continuous EKF Execution: When continuously executing the prediction-update iteration process of Extended Kalman Filter (EKF), according to... The reference covariance matrix is ​​updated in real time, and the mixing threshold is gradually reduced to maintain the continuous execution of EKF.

[0051] Step 4: Adaptive processing of time-varying Gaussian noise: A forgetting factor is introduced to limit the memory length of the algorithm. Corresponding noise statistical characteristics are designed for EKF and UKF respectively to update the formula in real time. The mean and covariance matrix of process noise and observation noise are estimated online to suppress time-varying noise caused by uneven illumination and debris interference in the space environment.

[0052] Step 4.1: Establish a time-varying noise model: Assume that the space noise is time-varying and independent Gaussian noise, defined as:

[0053] ;

[0054] ;

[0055] in, This represents the average process noise. To observe the noise, Process noise covariance matrix Observational noise mean Observation noise covariance matrix ;

[0056] Step 4.2: Introduce the forgetting factor mechanism: Set the forgetting factor The value ranges from 0.960 to 0.99. Calculate the adaptive weighting coefficients at the k-th discrete time step. ;

[0057] Step 4.3: When executing the prediction-update iterative process of the Extended Kalman Filter (EKF), the noise statistics update formula driven by the first-order linearized error is used:

[0058] ;

[0059] ;

[0060] ;

[0061] ;

[0062] When performing the prediction-update iterative process of the Unscented Kalman Filter (UKF), the noise statistical characteristics update formula driven by the unscented transform error is used:

[0063] ;

[0064] ;

[0065] ;

[0066] Step 4.4: After each filtering iteration is completed, the corresponding noise update formula is automatically called according to the currently executed prediction-update iteration process. The statistical characteristics of process noise and observation noise are updated online and used for prediction and update calculation of the next filtering iteration. By letting k=k+1 and executing it in a loop, the process returns to step 2 to realize continuous real-time estimation of target attitude motion parameters.

[0067] On the other hand, the method for estimating the attitude motion parameters of a space non-cooperative tumbling target based on adaptive hybrid Kalman filtering achieves real-time calculation of the target state through a space non-cooperative target on-orbit acquisition system, including a visual measurement unit, a spaceborne computing unit, a space robotic arm control unit, and a target acquisition execution unit.

[0068] The visual measurement unit acquires images and point cloud raw observation data of the tumbling target, completes target recognition and feature extraction, and outputs relative motion measurement information.

[0069] The onboard computing unit receives relative motion measurement information output by the vision measurement unit, calculates the high-precision state of the tumbling target, performs task logic judgment, robotic arm trajectory planning, and base attitude coupling dynamics calculation, generates the desired motion trajectory of the robotic arm, and sends the planned desired motion trajectory of the robotic arm to the space robotic arm control unit.

[0070] The space robotic arm control unit receives the desired motion trajectory of the robotic arm from the onboard computing unit, outputs capture action commands, and controls the robotic arm itself.

[0071] The target capture execution unit is a robotic arm end effector; it receives capture action commands output by the space robotic arm control unit and completes the task.

[0072] The aforementioned on-orbit capture system for non-cooperative targets in space adaptively and adaptively suppresses time-varying measurement noise in space by dynamically switching between the prediction-update iteration process of the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF). It outputs in real time the attitude quaternions, three-axis angular velocities, and inertia parameters of a tumbling target rotating around any axis, providing motion parameter inputs for the capture by the space robotic arm.

[0073] The beneficial effects of adopting the above technical solution are as follows:

[0074] This invention provides a method for estimating the attitude motion parameters of a spatial non-cooperative tumbling target based on adaptive hybrid Kalman filtering, which has the following advantages:

[0075] 1. This invention can effectively balance attitude estimation accuracy and computational efficiency. It combines the speed of extended Kalman filtering with the high accuracy of unscented Kalman filtering. The attitude and angular velocity estimation accuracy is close to that of unscented Kalman filtering, and the average CPU time is only 35.8% of that of unscented Kalman filtering, thus solving the real-time problem under the limited computing resources of spaceborne systems.

[0076] 2. This invention employs a time-varying noise adaptive processing mechanism, introducing a forgetting factor to update the noise statistical characteristics in real time. Compared with traditional fixed noise filtering methods, it significantly improves robustness in complex environments such as uneven illumination in space and interference from space debris, and effectively suppresses the impact of time-varying measurement noise on estimation accuracy.

[0077] 3. The present invention has rigorously proven the sufficient condition for local convergence of the filter, filling the theoretical gap in the convergence of hybrid Kalman filters, mathematically guaranteeing that the algorithm will not diverge, and providing reliable theoretical support for engineering applications.

[0078] 4. The method of the present invention is applicable to rigid tumbling targets rotating around any axis, overcoming the limitation of existing methods that can only handle targets rotating around the principal axis of inertia. The parameters can be flexibly configured according to mission requirements, and it is easy to implement in engineering. It can be extended to various space non-cooperative target operation missions such as space debris cleanup, on-orbit maintenance, and rendezvous and docking. Attached Figure Description

[0079] Figure 1 A schematic diagram of the process for estimating the attitude motion parameters of a spatial non-cooperative tumbling target based on adaptive hybrid Kalman filtering according to an embodiment of the present invention;

[0080] Figure 2 An on-orbit schematic diagram of a space tumbling target according to an embodiment of the present invention;

[0081] Figure 3 Block diagram of the extended Kalman filter according to an embodiment of the present invention;

[0082] Figure 4 Block diagram of the unscented Kalman filter according to an embodiment of the present invention;

[0083] Figure 5 A schematic diagram of the implementation device of this invention. Detailed Implementation

[0084] The specific implementation methods of this application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0085] Example 1:

[0086] Specifically, this invention targets non-cooperative tumbling targets in space.

[0087] On the one hand, this invention provides a method for estimating the parameters of a space non-cooperative tumbling target based on adaptive hybrid Kalman filtering, such as... Figure 1 As shown, it includes the following steps:

[0088] Step 1: Mathematical modeling of the tumbling target: Propose hypotheses, derive the differential equations of attitude kinematics and the attitude dynamics equations without external torque, construct a 13-dimensional state vector containing attitude quaternions, three-axis angular velocities and inertia parameters, and establish nonlinear discrete state space equations and linear observation equations;

[0089] The specific assumption is as follows:

[0090] (1) Ignore the influence of gravity and do not consider orbital dynamics;

[0091] (2) The non-cooperative tumbling target is regarded as a cube-shaped decommissioned satellite with two solar panels and one antenna, whose center of mass CM is located at the geometric center and rotates around its arbitrary axis with a certain initial velocity;

[0092] (3) The rolling target is in a free-floating state, and the external force or torque is negligible;

[0093] (4) The rolling target is rigid;

[0094] (5) The system noise caused by the complex space environment is time-varying Gaussian noise.

[0095] Step 1.1: Define the inertial reference coordinate system and the target body coordinate system. Use unit quaternions to represent the target attitude to avoid Euler angle singularities. In space, a tumbling target is in a state of loss of control, and its attitude changes significantly. Therefore, unit quaternions are chosen to represent the attitude of the tumbling target to avoid singularities. The attitude kinematic differential equations are derived as follows: ;

[0096] in For attitude quaternions, It is the time rate of change of the attitude quaternion; This is the antisymmetric matrix corresponding to the angular velocity. The three-axis angular velocities are defined in the target body coordinate system; in this embodiment... The triaxial angular velocity is ω tx ω ty ω tz These correspond to the magnitudes of the angular velocities along the x, y, and z axes of the target body, respectively. × is the antisymmetric matrix operator for the vector, and T is the transpose of the vector.

[0097] In this embodiment, as shown Figure 2 As shown, and These are the inertial reference coordinate system and the coordinate system fixed on the rolling target, respectively, acting on... The external torque of the tumbling target is ,in These are the three-axis torque components in the body coordinate system.

[0098] from arrive rotation transformation matrix Using Euler axis-angle parameter (e, ) is represented as:

[0099] ;

[0100] in, It is the unit vector of the rotation axis. It is the rotation angle about the axis. E3 is a third-order identity matrix, e x e y e z These are the xyz axis components of the rotation axis;

[0101] Step 1.2: Based on the angular momentum theorem and the assumption that the target is freely floating without external torque, the attitude dynamics equations without external torque are derived: ;

[0102] in, The angular acceleration vector of the rolling target. For the rolling target about inertia;

[0103] In this embodiment, according to the angular momentum theorem, the tumbling target relative to its center of mass... The angular momentum is:

[0104] ;

[0105] in, It is the total mass of the tumbling target. Let be the angular momentum of each part of the target with respect to the center of mass. It is the centroid vector. Let be the centroid vectors of each part of the tumbling target. Let i be the mass of each part of the tumbling target, and let i be the i-th infinitesimal element. It is the translational velocity vector of the center of mass, ω tz ω ty ω tx These are the angular velocity components of the tumbling target about the Zt, Yt, and Xt axes of the body coordinate system, such as... Figure 2 As shown;

[0106] ;

[0107] in , , These are the moments of inertia of the target about its three principal axes x, y, and z. , , These are the inertia products of the xy plane, xz plane, and yz plane, respectively.

[0108] According to the theorem of angular momentum, the external torque acting on the center of mass express:

[0109] ;

[0110] t represents the parameter relative to the body coordinate system of the rolling target; it represents the angular momentum vector of the rolling target with respect to its center of mass. Substituting and based on assumption (3), the external torque The target attitude dynamic equations are obtained as follows:

[0111] ;

[0112] in, It is an inertial variable.

[0113] Step 1.3: Using the six independent elements of the inertia matrix—namely, the three principal inertia and the three inertia products—as parameters to be estimated, construct a 13-dimensional state vector: ;

[0114] in, It is an inertial variable;

[0115] Step 1.4: Combine the attitude kinematics differential equations, the attitude dynamics equations without external torques, and the invariant inertia parameter to construct continuous state-space equations: ;

[0116] in, For continuous state change rate, These are the state components corresponding to the attitude quaternions. The state component corresponding to the angular velocity. These are the state components corresponding to the inertia. This includes modeling errors that incorporate the motion disturbance of space targets by small space debris and unstable airflow in outer space.

[0117] Step 1.5: Discretize the continuous state equations using the Euler method to obtain the nonlinear discrete state-space equations:

[0118] ;

[0119] in, Let k be the system state vector at the k-th discrete time. , Let Δt be the continuous state change rate at the k-th discrete time, and Δt be the filter iteration step size. The noise in the discrete process at time k;

[0120] Step 1.6: Based on the characteristic that attitude quaternions, angular velocity, and inertia parameters can be directly measured by sensors, construct the linear observation equation:

[0121] ;

[0122] Z tk This is the observation vector of the system over continuous time, containing information on the target's attitude, angular velocity, and inertia measured by sensors. It is a 13th order identity matrix. The measurement noise at the k-th discrete time;

[0123] In this embodiment, the attitude quaternion, angular velocity, and inertia can be measured by sensors. Therefore, the linear observation equation is:

[0124] ;

[0125] in, For measuring noise.

[0126] The nonlinear discrete state-space equation for a non-cooperative tumbling target is:

[0127] ;

[0128] in, , ;

[0129] Step 2: Basic filtering algorithm design: Implement the prediction-update iteration process of the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF) respectively, so as to obtain the posterior estimate of the state and the posterior estimate of the error covariance at the current time.

[0130] The prediction-update iteration process of the Extended Kalman Filter (EKF) includes the following steps:

[0131] Step A2.1: Set the initial state posterior estimate and the initial error covariance matrix Define process noise and observation noise For zero-mean Gaussian white noise, set its covariance matrix respectively. and ;

[0132] Step A2.2: Linearize the discrete-time state transition equation of the nonlinear system as follows: Among them, by calculating the nonlinear state transition function The posterior estimate at the previous time step Jacobian matrix at the location We obtain the state transition matrix from step k to step k+1. Based on the posterior estimate from the previous time step, calculate the prior state prediction for the current time step. Sum of error covariance prior predicted values ;

[0133] Step A2.3: Calculate the Kalman gain matrix: ;

[0134] in The observation matrix;

[0135] Combined with the observations at the current moment The update yields the posterior estimate of the state at the current time step. ; Posterior estimate of error covariance ,in It is the Kalman gain matrix of the k-th iteration, and the discrete observation function. Let be the prior error covariance matrix at step k;

[0136] The prediction-update iteration process of the Unscented Kalman Filter (UKF) includes the following steps:

[0137] Step B2.1: Initialization: Configure unscented transform scaling parameters (Usually taken between 0.001 and 0.01), κ is the second-order scaling parameter, set to 0. This is a parameter tuning parameter specifically used in the unscented transform to utilize prior statistical information about the state distribution. Setting it to 0 allows for the calculation of the scaling factor. and Sigma point weights:

[0138] ;

[0139] Where L is the dimension of the state vector, and in this embodiment L=13; Let be the mean weight corresponding to the i-th Sigma point. Let be the covariance weight corresponding to the i-th Sigma point;

[0140] Step B2.2: Unscented Transformation and State Prediction: Based on the current state estimate and error covariance matrix, generate 2L+1 Sigma points:

[0141] ;

[0142] in, Let be the i-th Sigma sampling point generated by the unscented transform. State vector The mean, Let be the error covariance matrix of the state vector. iThis indicates retrieving the i-th row of the matrix within the parentheses;

[0143] Substitute each Sigma point into the nonlinear state transition function. Propagation is performed to obtain the predicted value of the i-th Sigma point after one state transition, which means that the quantity at time k is predicted based on the information at time k-1; the prior state prediction value is calculated. Sum of error covariance prior predicted values ;

[0144] Step B2.3: Observation Prediction: Substitute the propagated Sigma points into the observation function. ; Calculate prior predictions based on observations Observation error covariance matrix and the cross-correlation matrix between state and observation ;

[0145] Step B2.4: Combine the observations at the current moment Update to obtain the posterior estimate of the state at the current time. Posterior estimate of error covariance ;

[0146] In this embodiment, an unscented Kalman filter is performed in the initial stage to quickly suppress large initial errors and achieve preliminary convergence of state estimation.

[0147] Generally, the initial error is large, so we use the UKF algorithm to quickly suppress the error.

[0148] The flowchart for UKF parameter estimation is as follows: Figure 3 As shown.

[0149] State variables of unscented transformation Corresponding sigma point and weight They are respectively:

[0150] ;

[0151] in, for dimensionality Its mean, For its variance, It is a scaling parameter. Determine the sigma point at The diffusion velocity in the vicinity is a small positive number. This is a secondary scaling parameter, usually set to 0. The value of and It relates to prior knowledge of the distribution; for a Gaussian distribution, To be the optimal value, It is the i-th row of the square root of the upper triangular matrix.

[0152] Calculate the predicted mean:

[0153] ;

[0154] ;

[0155] The predicted covariance is:

[0156] ;

[0157] The predicted observations are calculated as follows:

[0158] ;

[0159] ;

[0160] The corresponding covariance and cross-correlation matrices are:

[0161] ;

[0162] ;

[0163] in, yes The prior prediction value;

[0164] State estimate Sum of mean squared error matrix for:

[0165] ;

[0166] ;

[0167] Step 3: Hybrid switching rule design: A dynamic switching strategy based on the trace of the error covariance matrix is ​​proposed. The hybrid threshold coefficient and the reference covariance matrix are set. The filtering algorithm is adaptively selected according to the current estimation error. The reference covariance matrix is ​​updated when EKF is executed continuously to speed up convergence.

[0168] Step 3.1: Preset the core parameters of the mixing strategy: Set the mixing threshold coefficient Used to control the execution ratio of EKF and UKF; initializes the reference covariance matrix. This serves as the benchmark threshold for judging the magnitude of the estimation error.

[0169] Step 3.2: Real-time calculation of the overall estimation error index: In each discrete period of filtering iteration, calculate the posterior error covariance matrix at the current time. traces This is used to quantify the overall error level of the current 13-dimensional state estimation;

[0170] Step 3.3: Adaptive algorithm switching based on error threshold:

[0171] if The prediction-update iteration process of the Unscented Kalman Filter (UKF) is executed.

[0172] if The prediction-update iteration process of the Extended Kalman Filter (EKF) is executed.

[0173] Step 3.4: Reference Covariance Update During Continuous EKF Execution: When continuously executing the prediction-update iteration process of Extended Kalman Filter (EKF), according to... The reference covariance matrix is ​​updated in real time, the mixing threshold is gradually reduced, the EKF is maintained in a continuous execution state, and the overall convergence speed of the algorithm is further accelerated.

[0174] In this embodiment, a hybrid Kalman filter is designed to balance the accuracy and efficiency of the filter. To avoid jumps in the mean square error function that could lead to instability in the parameter estimation system, the following hybrid strategy is established: If Execute the UKF algorithm; if Execute the EKF algorithm;

[0175] in, The trace of the matrix, the mixing threshold coefficient , covariance matrix The reference covariance matrix is ​​used. To improve the convergence speed of the algorithm, the parameter estimation continuously executes the EKF iterative loop. ;

[0176] Once the switching conditions are met, the extended Kalman filter algorithm is used to reduce computation time.

[0177] We linearize the discrete-time state transition equation of the nonlinear system as follows:

[0178] ,

[0179] in, It is the state transition matrix from step k to step k+1. , This is the process noise of EKF, and it is the mean value. and variance White noise, that is, .

[0180] According to the observation function formula, the observation transfer equation can be written as:

[0181] ;

[0182] in, , This is the observation noise of EKF, which is the mean. ,variance White noise, that is, .

[0183] Combining the above formulas, the linear time-varying state transition equation of EKF can be obtained as follows:

[0184] ;

[0185] in, ;

[0186] like Figure 4 As shown, the iterative calculation process of EKF is as follows:

[0187] initialization:

[0188] , , ;

[0189] in, For the initial state variables, It is the state variable at step 0. The estimated value.

[0190] Model prediction:

[0191] ;

[0192] in, For k-1 step state variables The estimated value, yes The prior prediction value, It is the error covariance matrix for k-1 steps. for The prior prediction value.

[0193] Model correction:

[0194] ;

[0195] in, It is Kalman gain. yes The observed values.

[0196] Step 4: Adaptive processing of time-varying Gaussian noise: A forgetting factor is introduced to limit the memory length of the algorithm. Corresponding noise statistical characteristics are designed for EKF and UKF respectively to update the formula in real time. The mean and covariance matrix of process noise and observation noise are estimated online to suppress time-varying noise caused by uneven illumination and debris interference in the space environment.

[0197] Step 4.1: Establish a time-varying noise model: Assume that the space noise is time-varying and independent Gaussian noise, defined as:

[0198] ;

[0199] ;

[0200] in, This represents the average process noise. To observe the noise, Process noise covariance matrix Observational noise mean Observation noise covariance matrix ;

[0201] Step 4.2: Introduce the forgetting factor mechanism: Set the forgetting factor The value ranges from 0.960 to 0.99. Calculate the adaptive weighting coefficients at the k-th discrete time step. This is used to limit the length of historical data that the algorithm remembers, thereby enhancing its ability to track time-varying noise.

[0202] Step 4.3: When executing the prediction-update iterative process of the Extended Kalman Filter (EKF), the noise statistics update formula driven by the first-order linearized error is used:

[0203] ;

[0204] ;

[0205] ;

[0206] ;

[0207] When performing the prediction-update iterative process of the Unscented Kalman Filter (UKF), the noise statistical characteristics update formula driven by the unscented transform error is used:

[0208] ;

[0209] ;

[0210] ;

[0211] Step 4.4: After each filtering iteration is completed, the corresponding noise update formula is automatically called according to the currently executed prediction-update iteration process. The statistical characteristics of process noise and observation noise are updated online and used for prediction and update calculation of the next filtering iteration. By letting k=k+1 and executing it in a loop, the process returns to step 2 to realize continuous real-time estimation of target attitude motion parameters.

[0212] In this embodiment, the iterative loop is entered, the error threshold is compared in real time, and the attitude motion parameters are continuously output.

[0213] After executing the prediction-update process of the corresponding filtering algorithm, the posterior state estimate at the current time is obtained. and posterior error covariance matrix ;

[0214] Depending on the algorithm type executed in this iteration, different noise update formulas are used to calculate the updated process noise and observation noise statistical characteristics. The specific formulas are as follows:

[0215] Due to the uncertainty of the space environment, the measurement system is susceptible to interference from external time-varying noise, thus an adaptive technique is introduced. Space noise is assumed to be time-varying, independent Gaussian noise, defined as:

[0216] , ,

[0217] in, , , , .

[0218] The real-time iterative formulas for the statistical variables of process noise and observation noise in the adaptive extended Kalman filter based on the first-order linearization error are as follows:

[0219] ,

[0220] ,

[0221] ,

[0222] ;

[0223] The real-time iterative formulas for the statistical variables of process noise and observation noise of the adaptive unscented Kalman filter based on unscented transform error are as follows:

[0224] ,

[0225] ,

[0226] ,

[0227] ;

[0228] in, The forgetting factor, typically ranging from 0.960 to 0.999, is used to limit the memory length of the iterative process and update the optimization target in a timely manner based on the state.

[0229] Let k = k + 1, and select either the EKF or UKF algorithm for the real-time comparison error threshold, then proceed to the next iteration.

[0230] from The attitude quaternions, three-axis angular velocities, and inertia parameters are extracted and output to the robotic arm for real-time trajectory planning and motion following control at the end effector, enabling the end effector to accurately match the tumbling motion of non-cooperative targets in space. The inertia parameters are also used to predict the target's motion trend, plan the capture path in advance, and calculate the attitude stabilization control torque after capture.

[0231] On the other hand, the method for estimating the attitude motion parameters of a space non-cooperative tumbling target based on adaptive hybrid Kalman filtering achieves real-time calculation of the target state through a space non-cooperative target on-orbit acquisition system, including a visual measurement unit, a spaceborne computing unit, a space robotic arm control unit, and a target acquisition execution unit.

[0232] The visual measurement unit is the perception front end of the entire capture system. It acquires images and point cloud raw observation data of the tumbling target, completes target recognition and feature extraction, and outputs relative motion measurement information, such as the position, attitude, and angular velocity of the non-cooperative target relative to the servant spacecraft.

[0233] The onboard computing unit is the core data processing hub; it receives relative motion measurement information output by the visual measurement unit, calculates the high-precision state of the tumbling target by running the attitude / motion parameter estimation algorithm, executes task logic judgment, robotic arm trajectory planning, base attitude coupling dynamics calculation, generates the desired motion trajectory of the robotic arm, and sends the planned desired motion trajectory of the robotic arm to the space robotic arm control unit.

[0234] The space robotic arm control unit is a motion execution controller that receives the desired motion trajectory of the robotic arm from the on-board computing unit, outputs capture motion commands, and controls the robotic arm itself.

[0235] The target capture execution unit is a hardware execution mechanism that directly contacts the target, namely the end effector of the robotic arm (gripper, locking mechanism, flexible capture mechanism); it receives capture action commands output by the space robotic arm control unit and completes the task.

[0236] The aforementioned on-orbit capture system for non-cooperative targets in space adaptively and adaptively suppresses time-varying measurement noise in space by dynamically switching between the prediction-update iteration process of the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF). It outputs in real time the attitude quaternions, three-axis angular velocities, and inertia parameters of a tumbling target rotating around any axis, providing motion parameter inputs for the precise capture by the space robotic arm.

[0237] The aforementioned method for estimating the attitude motion parameters of a space non-cooperative tumbling target based on adaptive hybrid Kalman filtering is applicable to real-time, high-precision attitude parameter estimation of space non-cooperative tumbling targets in space debris cleanup and on-orbit repair missions of failed spacecraft.

[0238] This method is applicable to the estimation of the motion state of rigid tumbling targets rotating around arbitrary axes, and can be extended to various non-cooperative space target operation tasks such as rendezvous and docking, on-orbit refueling, and space debris removal. It can also be applied to the simulation and verification of target motion state in ground hardware-in-the-loop simulation systems.

[0239] like Figure 5 The diagram shown is a structural diagram of the device for implementing the method of the present invention. This embodiment describes a method for estimating the attitude motion parameters of a non-cooperative rolling target in space based on adaptive hybrid Kalman filtering, which includes a method for on-orbit capture of the ground in space.

[0240] Space robot: It consists of a satellite body, dual solar panels and a multi-degree-of-freedom space robotic arm. The end of the robotic arm integrates a hand-eye camera vision measurement unit. The onboard computing unit runs the attitude estimation algorithm of this invention, which can realize six-degree-of-freedom orbit and attitude motion, and complete the identification, tracking and precise capture of the target.

[0241] Non-cooperative tumbling target: Simulating the structure of a spacecraft that fails in orbit, consisting of a cubic body, dual solar panels and a preset capture interface, capable of performing complex tumbling motions around any axis, serving as the measurement object and final capture target of the attitude parameter estimation method of this invention.

[0242] The space robot's hand-eye camera collects real-time image information of the non-cooperative tumbling target, transmits it to the onboard computing unit to run the adaptive hybrid Kalman filter algorithm of this invention, calculates the target's real-time attitude quaternion, three-axis angular velocity and inertia parameters, and drives the space robotic arm to complete the stable capture operation of the target.

[0243] Example 2: Working principle:

[0244] 1. Mathematical modeling and filter initialization stage:

[0245] (1) Target modeling: Based on the assumption of a free-floating rigid target, ignoring the influence of orbital dynamics and external torque, we define an inertial coordinate system and a target body coordinate system, use unit quaternions to represent the attitude to avoid singularities, derive the attitude kinematics and dynamic equations, incorporate the 6 independent elements of the inertia matrix into the state to be estimated, and construct a 13-dimensional nonlinear discrete state space equation and a linear observation equation that includes attitude quaternions, three-axis angular velocity and inertia parameters.

[0246] (2) Parameter initialization: Set the filter iteration step size to match the visual measurement sampling frequency, initialize the initial posterior estimate of the 13-dimensional state vector, the initial error covariance matrix, and the initial statistical characteristics of process noise and observation noise; configure the mixing threshold coefficient β∈[0.7,0.9], the adaptive forgetting factor b∈[0.96,0.99], and the UKF dedicated unscented transform scaling parameter to complete the filter startup preparation.

[0247] 2. Initial large error suppression and rapid convergence stage:

[0248] (1) Algorithm determination: The initial error covariance matrix is ​​large during initialization, which satisfies The conditions are automatically determined to execute the UKF algorithm;

[0249] (2) UKF iteration: Generate 2L+1 Sigma points and propagate them through the nonlinear state transition function. Calculate the state prior prediction value and the error covariance prior prediction value. Combine the current visual observation value to complete the state posterior update. Utilize the high-precision estimation capability of UKF for strongly nonlinear systems to quickly suppress large initial errors.

[0250] (3) Convergence determination: After each iteration, the trace of the posterior error covariance matrix is ​​calculated in real time. When the error drops below the mixing threshold, the preliminary convergence is determined.

[0251] 3. Dynamic Algorithm Switching and High-Efficiency Operation Phase:

[0252] (1) Algorithm switching: when When the time comes, it automatically switches from the UKF algorithm to the EKF algorithm, and achieves state prediction and update through linearization of the first-order Jacobian matrix, which greatly reduces the amount of computation.

[0253] (2) Threshold dynamic adjustment: If EKF is executed in consecutive iterations, according to The rule update references the covariance matrix, gradually reducing the mixing threshold to maintain the continuous execution state of EKF;

[0254] (3) Abnormal back-cut: When interference causes a sudden increase in error. When the error decreases, the algorithm automatically switches back to UKF to reconverge, and then switches back to EKF after the error decreases.

[0255] 4. Adaptive noise update and real-time parameter output stage:

[0256] (1) Noise online update: After each filtering iteration, the corresponding noise statistical characteristic update formula is automatically called according to the type of algorithm currently being executed. Based on the exponential weighted moving average mechanism, the mean and covariance matrix of the process noise and the observed noise are calculated online for prediction and update calculation in the next iteration.

[0257] (2) Parameter extraction and output: Extract the attitude quaternion, three-axis angular velocity and inertia parameters from the current state posterior estimate and output them to the space robot control unit in real time for robot end trajectory planning and motion following control;

[0258] (3) Iterative loop: Let k=k+1, return to the algorithm decision step, and realize the continuous real-time estimation of the attitude motion parameters of the non-cooperative rolling target in space.

[0259] Example 3:

[0260] This embodiment proposes a computer-readable storage medium that stores executable instructions. When these instructions are executed, if they are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0261] The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the adaptive hybrid Kalman filter spatial non-cooperative tumbling target attitude motion parameter estimation method described in the various embodiments of this application.

[0262] The aforementioned storage media include: flash memory, hard disk, multimedia card, card-type memory (e.g., SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR) memory, random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, APP (Application) application store, and other media capable of storing program verification codes. These media store computer programs, and when executed by a processor, they can implement the various steps of the adaptive hybrid Kalman filter spatial non-cooperative tumbling target attitude motion parameter estimation method described above.

[0263] Example 4:

[0264] This embodiment proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the adaptive hybrid Kalman filter method for estimating the attitude motion parameters of a spatial non-cooperative tumbling target.

[0265] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a computer program product.

[0266] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0267] The scope of protection of this application is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from the scope and spirit of this disclosure. If such modifications and variations fall within the scope of the methods disclosed herein and their equivalents, then the intent of this disclosure also includes such modifications and variations.

Claims

1. A method for estimating the parameters of a space non-cooperative tumbling target based on adaptive hybrid Kalman filtering, characterized in that, Includes the following steps: Step 1: Mathematical modeling of the tumbling target: Propose hypotheses, derive the differential equations of attitude kinematics and the attitude dynamics equations without external torque, construct a 13-dimensional state vector containing attitude quaternions, three-axis angular velocities and inertia parameters, and establish nonlinear discrete state space equations and linear observation equations. Step 2: Basic filtering algorithm design: Implement the prediction-update iteration process of the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF) respectively, so as to obtain the posterior estimate of the state and the posterior estimate of the error covariance at the current time. Step 3: Hybrid switching rule design: A dynamic switching strategy based on the trace of the error covariance matrix is ​​proposed. The hybrid threshold coefficient and the reference covariance matrix are set. The filtering algorithm is adaptively selected according to the current estimation error. The reference covariance matrix is ​​updated when EKF is executed continuously to speed up convergence. Step 4: Adaptive processing of time-varying Gaussian noise: A forgetting factor is introduced to limit the algorithm's memory length. Corresponding noise statistical characteristics are designed for EKF and UKF respectively, and real-time update formulas are designed. The mean and covariance matrices of process noise and observation noise are estimated online to suppress time-varying noise caused by uneven illumination and debris interference in the space environment.

2. The method for estimating the parameters of a space non-cooperative roll target based on adaptive hybrid Kalman filtering according to claim 1, characterized in that, The specific assumptions made in step 1 are as follows: (1) Ignore the influence of gravity and do not consider orbital dynamics; (2) The non-cooperative tumbling target is regarded as a cube-shaped decommissioned satellite with two solar panels and one antenna, whose center of mass CM is located at the geometric center and rotates around its arbitrary axis with a certain initial velocity; (3) The rolling target is in a free-floating state, and the external force or torque is negligible; (4) The rolling target is rigid; (5) The system noise caused by the complex space environment is time-varying Gaussian noise.

3. The method for estimating the parameters of a space non-cooperative roll target based on adaptive hybrid Kalman filtering according to claim 2, characterized in that, Step 1 specifically includes the following steps: Step 1.1: Define the inertial reference coordinate system and the target body coordinate system, use unit quaternions to represent the target attitude, and derive the attitude kinematic differential equations: ; in For attitude quaternions, It is the time rate of change of the attitude quaternion; This is the antisymmetric matrix corresponding to the angular velocity. The three-axis angular velocities in the target's body coordinate system; Step 1.2: Based on the angular momentum theorem and the assumption that the target is freely floating without external torque, the attitude dynamics equations without external torque are derived: ; in, The angular acceleration vector of the rolling target. For the rolling target about inertia; Step 1.3: Using the six independent elements of the inertia matrix—namely, the three principal inertia and the three inertia products—as parameters to be estimated, construct a 13-dimensional state vector: ; in, It is an inertial variable; Step 1.4: Combine the attitude kinematics differential equations, the attitude dynamics equations without external torques, and the invariant inertia parameter to construct continuous state-space equations: ; in, For continuous state change rate, These are the state components corresponding to the attitude quaternions. The state component corresponding to the angular velocity. These are the state components corresponding to the inertia. This includes modeling errors that incorporate the motion disturbance of small space debris on space targets and unstable airflow in outer space; Step 1.5: Discretize the continuous state equations using the Euler method to obtain the nonlinear discrete state-space equations: ; in, Let k be the system state vector at the k-th discrete time. , Let Δt be the continuous state change rate at the k-th discrete time, and Δt be the filter iteration step size. The noise in the discrete process at time k; Step 1.6: Based on the characteristic that attitude quaternions, angular velocity, and inertia parameters can be directly measured by sensors, construct the linear observation equation: ; Z tk This is the observation vector of the system over continuous time, which includes target attitude, angular velocity, and inertia information measured by sensors. It is a 13th order identity matrix. Let be the measurement noise at the k-th discrete time.

4. The method for estimating the parameters of a space non-cooperative roll target based on adaptive hybrid Kalman filtering according to claim 3, characterized in that, The prediction-update iteration process of the Extended Kalman Filter (EKF) described in step 2 includes the following steps: Step A2.1: Set the initial state posterior estimate and the initial error covariance matrix Define process noise and observation noise For zero-mean Gaussian white noise, set its covariance matrix respectively. and ; Step A2.2: Linearize the discrete-time state transition equation of the nonlinear system as follows: Among them, by calculating the nonlinear state transition function The posterior estimate at the previous time step Jacobian matrix at the location We obtain the state transition matrix from step k to step k+1. Based on the posterior estimate from the previous time step, calculate the prior state prediction for the current time step. Sum of error covariance prior predicted values ; Step A2.3: Calculate the Kalman gain matrix: ; in The observation matrix; Combined with the observations at the current moment The update yields the posterior estimate of the state at the current time step. ; Posterior estimate of error covariance ,in It is the Kalman gain matrix of the k-th iteration, and the discrete observation function. Let be the prior error covariance matrix at step k.

5. The method for estimating the parameters of a space non-cooperative roll target based on adaptive hybrid Kalman filtering according to claim 4, characterized in that, The prediction-update iteration process of the Unscented Kalman Filter (UKF) described in step 2 includes the following steps: Step B2.1: Initialization: Configure unscented transform scaling parameters κ is the second scaling parameter, set to 0. This is a parameter tuning parameter specifically used in the unscented transform to utilize prior statistical information about the state distribution. Setting it to 0 allows for the calculation of the scaling factor. and Sigma point weights: ; Where L is the dimension of the state vector. Let be the mean weight corresponding to the i-th Sigma point. Let be the covariance weight corresponding to the i-th Sigma point; Step B2.2: Unscented Transformation and State Prediction: Based on the current state estimate and error covariance matrix, generate 2L+1 Sigma points: ; in, Let be the i-th Sigma sampling point generated by the unscented transform. State vector The mean, Let be the error covariance matrix of the state vector. i This indicates retrieving the i-th row of the matrix within the parentheses; Substitute each Sigma point into the nonlinear state transition function. Propagate the data to obtain the predicted value of the i-th Sigma point after one state transition; calculate the prior state prediction value. Sum of error covariance prior predicted values ; Step B2.3: Observation Prediction: Substitute the propagated Sigma points into the observation function. ; Calculate prior predictions based on observations Observation error covariance matrix and the cross-correlation matrix between state and observation ; Step B2.4: Combine the observations at the current moment Update to obtain the posterior estimate of the state at the current time. Posterior estimate of error covariance .

6. The method for estimating the parameters of a space non-cooperative tumbling target based on adaptive hybrid Kalman filtering according to claim 5, characterized in that, Step 3 specifically includes the following steps: Step 3.1: Preset the core parameters of the mixing strategy: Set the mixing threshold coefficient Used to control the execution ratio of EKF and UKF; initializes the reference covariance matrix. This serves as the benchmark threshold for judging the magnitude of the estimation error. Step 3.2: Real-time calculation of the overall estimation error index: In each discrete period of filtering iteration, calculate the posterior error covariance matrix at the current time. traces This is used to quantify the overall error level of the current 13-dimensional state estimation; Step 3.3: Adaptive algorithm switching based on error threshold: if The prediction-update iteration process of the Unscented Kalman Filter (UKF) is executed. if The prediction-update iteration process of the Extended Kalman Filter (EKF) is executed. Step 3.4: Reference Covariance Update During Continuous EKF Execution: When continuously executing the prediction-update iteration process of Extended Kalman Filter (EKF), according to... The reference covariance matrix is ​​updated in real time, and the mixing threshold is gradually reduced to maintain the continuous execution of EKF.

7. The method for estimating the parameters of a space non-cooperative roll target based on adaptive hybrid Kalman filtering according to claim 6, characterized in that, Step 4 specifically includes the following steps: Step 4.1: Establish a time-varying noise model: Assume that the space noise is time-varying and independent Gaussian noise, defined as: ; ; in, This represents the average process noise. To observe the noise, Process noise covariance matrix Observational noise mean Observation noise covariance matrix ; Step 4.2: Introduce the forgetting factor mechanism: Set the forgetting factor The value ranges from 0.960 to 0.

99. Calculate the adaptive weighting coefficients at the k-th discrete time step. ; Step 4.3: When executing the prediction-update iterative process of the Extended Kalman Filter (EKF), the noise statistics update formula driven by the first-order linearized error is used: ; ; ; ; When performing the prediction-update iterative process of the Unscented Kalman Filter (UKF), the noise statistical characteristics update formula driven by the unscented transform error is used: ; ; ; Step 4.4: After each filtering iteration is completed, the corresponding noise update formula is automatically called according to the currently executed prediction-update iteration process. The statistical characteristics of process noise and observation noise are updated online and used for prediction and update calculation of the next filtering iteration. By letting k=k+1 and executing it in a loop, the process returns to step 2 to realize continuous real-time estimation of target attitude motion parameters.

8. The spatial non-cooperative roll target parameter estimation method based on adaptive hybrid Kalman filtering as described in claim 1, wherein the target state is calculated in real time through the following system, characterized in that, It includes a vision measurement unit, a spaceborne computing unit, a space robotic arm control unit, and a target acquisition execution unit; The visual measurement unit acquires images and point cloud raw observation data of the tumbling target to complete target recognition and feature extraction. Output relative motion measurement information; The onboard computing unit receives relative motion measurement information output by the vision measurement unit, calculates the high-precision state of the tumbling target, performs task logic judgment, robotic arm trajectory planning, and base attitude coupling dynamics calculation, generates the desired motion trajectory of the robotic arm, and sends the planned desired motion trajectory of the robotic arm to the space robotic arm control unit. The space robotic arm control unit receives the desired motion trajectory of the robotic arm from the onboard computing unit, outputs capture action commands, and controls the robotic arm itself. The target capture execution unit is a robotic arm end effector; it receives capture action commands output by the space robotic arm control unit and completes the task. The system adaptively and dynamically switches the prediction-update iteration process of the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF) to suppress space time-varying measurement noise online. It outputs the attitude quaternion, three-axis angular velocity, and inertia parameters of a tumbling target rotating around any axis in real time, providing motion parameter input for the grasping of the space robotic arm.