A meta-reinforcement learning method based on the universal policy-online system identifier structure is proposed to address the issue of terminal guidance with large environmental uncertainty and diverse task types. The method consists of a reinforcement learning training phase for the UP and a supervised learning phase for the OSI， and ensures reliable convergence of the training in multi-task environments through a phased transfer learning design and a pseudo-Monte Carlo-based small-variance policy gradient estimation， so that the resulting guidance policy can be adapted to varieties of terminal guidance task scenario. Simulation results show that the resulting guidance policy can meet the requirements of terminal position and path angle error in multiple terminal guidance scenarios.

The hyperbolic orbit plane of the planetary gravity assist is determined by the entry speed and escape speed and based on two-body dynamics. The engagement and escape hyperbolic orbit's asymptotic angle are adjusted by the height at the near-planet point. Combined with the accelerating/braking speed increment near the planet， the direction of the escape velocity is determined. Based on the conversion algorithm between the heliocentric elliptical orbit and the planet's hyperbolic orbit， as well as the iterative algorithm for the height near the planet， the transfer orbit is quickly calculated， and the parameters of the approach/escape hyperbolic orbit are determined. By taking the 2015 XF261 asteroid defense as an instance， a Venus-assisted transfer orbit design is implemented， and the transfer orbit， the transfer duration and velocity increment requirements of the probe are presented. The simulation results show that rapid calculation of the analytical solution of the planetary gravity-assisted transfer orbits can significantly reduce the search time for transfer orbit.

Regarding the limitation of existing feedback gain designs based on specific nominal trajectories， a tracking guidance method for Mars powered descent in wide area based on control contraction metric （CCM） is proposed to match the emerging onboard trajectory planning capability. The CCM conditions for the powered descent model are analyzed. The CCM matrix and system dynamics are parameterized and approximated as polynomial functions， and then the CCM matrix is solved offline by using the method of sum of squares programming. During the flight， the control input is obtained through numerical integration based on the CCM matrix. The simulation results show that the nominal trajectories of different initial and final motion states and flight durations can be tracked by using the contraction control method under the conditions of specified range of mass and thrust.

The precision contribution of the missile-borne hybrid inertial navigation system（HINS） in missile-borne platform application is researched in this paper. On the basis of clarifying the principle and characteristics of HINS， its theoretical advantages in the mission chain of missile usage are analyzed. The error models of HINS is provided and its propagation characteristics are both analyzed. The impact on the precision contribution of HINS is focused towards researches regarding the errors of inertial instrument， initial alignment errors， trajectory characteristics， rotation modulation schemes， the integrated navigation scheme and the other factors. The reasons for the different opinions on the precision advantages of HINS in the related professional fields are analyzed， and a much more reasonable evaluation scheme for the precision advantages of missile-borne HINS is further proposed to support and improve the precision evaluation system of missile-borne HINS， which can serve as important reference for the engineering application of missile-borne HINS.

A majorant-based control method is proposed for quadrotor unmanned aerial vehicle trajectory tracking to enhance system robustness and tracking precision. Firstly， an extended state observer is designed for the position loop to compensate for the effects of external disturbances. Secondly， a position loop controller is developed via majorant systems. Furthermore， regarding the inherent cascade-control structure of quadrotor system， the design methodologies for both the observer and controller are concurrently applied to the attitude loop. Simulation results demonstrate that the proposed control method achieves convergence of the steady-state error to specified values， enables high-precision trajectory tracking and presents computational efficiency of practical implementation.

In response to the impact of spaceborne radars installation error on data-fusion accuracy， a multi-radar spatial registration algorithm based on same target observations is proposed. The measurement vectors of the same target obtained by different tracking radars is processed by using this algorithm under a common reference coordinate system through cross-product operations， and the cross-product vectors are projected onto a specified plane to establish the relationship with the relative installation bias angles， and the unbiasedness of the least squares estimation algorithm is verified by analyzing the characteristics of derived observation errors. Through actual on-orbit satellite conditions based simulation experiments， the results demonstrate that the proposed method can reliably and rapidly achieve spatial registration of different radars without relying on satellite attitude measurement.

Regarding circular restricted three-body subject， a prediction method based on deep neural network is proposed for two impulse Earth-Moon transfer orbit. Firstly， three types of two-impulse transfer orbit family are selected， and the state quantities of each transfer orbit are used to establish two-impulse transfer orbit data set in which input state and predicted state of orbit are determined. Secondly， a deep neural network is constructed for orbit state prediction， and， each class of two-impulse transfer orbits in the data set is used as the training set， and the neural network is trained by the track data in the training set. Finally， according to the training results， the state quantities about transfer orbit are predicted by setting the initial state of the low-Earth orbit， and then the optimal Earth-Moon two-impulse transfer orbit in each orbit family is selected according to the prediction results. The simulation results show that the application of deep neural network can quickly predict initial state of orbit， and the prediction results have fairly smaller error intervals and can be applied to selection of optimal two-impulse transfer orbit.

In response to the subjects of long development cycles， requirement verification difficulty， interfaces unclearness and processes amendment complexity in traditional system engineering development， the thrust regulation controller is taken as an example in this paper to conduct research on model-based design methods. MagicDraw known as tool is used for system modeling and behavior simulation， SCADE is used for the tool of software design and code automatic generation， and ModelCenter is used for joint simulation of system behavior model and software design model. It is an exploration and research on the specific implementation of model-based systems engineering in complex aerospace systems engineering. This method can serve as important application value for the development of complex control systems and the new ideas and technical paths are suggested to design the future spacecraft thrust adjustment control.

The issue of high-reliability numerical computation software is addressed in aerospace embedded systems being susceptible to floating-point bugs， and a static symbolic execution-based method is proposed for detecting floating-point bugs in numerical software. Through the method procedure， the floating-point expression is established by symbolization under related constraints， and interval computation and interval constraint propagation is introduced in non-relational numerical abstract domains， and math functions are symbolically modeled in order to improve the accuracy of detecting floating-point bugs. Regarding application to actual floating-point numerical software and aerospace testing， accuracy and efficiency are effectually improved， and multiple genuine floating-point bugs are successfully identified and confirmed by developers， which significantly enhances the reliability of aerospace software.

According to stability under target intelligence recognition under interference， a intelligence system testing method is proposed for uncertainty-metric-based guidance target mutation. Regarding interference scenarios that are typical in remote-sensing images， diverse object-level mutation strategies are combined with proposed method， specifically including object insertion， deletion， and replacement， for perturbing the original test data. Concurrently， it introduces model-predicted uncertainty metrics to guide the generation of efficient test cases， thereby， the original test set is augmented. Experiments are conducted by using the YOLOv5 model on the large-scale MAR20 aircraft remote-sensing dataset. The results demonstrate that the test datasets generated by our method which performs superior error-detection efficiency and are more effective at revealing latent defects in the model. Consequently， this approach offers a practical and effective means for robustness testing of object detection systems.

The selection of stable spin angular velocity for aircraft is focused in this paper. A method for choosing a stable angular velocity based on the spin stability criterion of aircraft is researched and proposed. Through theoretical analysis of the coning motion dynamics of spinning vehicles， the coupled effects of static stability moment， Magnus moment， and damping moment on the directional stability of the projectile body are fundamentally revealed. Based on a coning motion angular dynamics model under the constraints of deviations between the center of mass and the center of pressure， a system characteristic equation is established and the Routh-Hurwitz stability criterion is then employed to formulate a dynamic stability criterion. Furthermore， on the basis of the traditional three-degree-of-freedom trajectory model， the derived dynamic stability criterion for coning motion is used to deduce the feasible region for the spin angular velocity. A stable spin rate scheme for the aircraft is subsequently designed based on this region. Numerical simulation results demonstrate that the coning motion divergence is effectively suppressed by using proposed method that simultaneously reduces control energy consumption in the pitch/yaw channels. The achievement presented hereinabove provide a technical pathway to directional stable flight with low-energy-consumption attitude control by applying the spin characteristics of aircraft， which can offer significant theoretical and practical value.