Recently, Graph Transformers have emerged as a promising solution to alleviate the limitations of Graph Neural Networks (GNNs) and enhance graph representation performance. Unfortunately, Graph Transformers are computationally expensive due to the quadratic complexity inherent in self-attention when applied over large-scale graphs, especially for node tasks. In contrast, Spiking Neural Networks (SNNs), with event-driven and binary spikes properties, can perform energy-efficient computation. In this work, we propose a novel insight into integrating SNNs with Graph Transformers and design a Spiking Graph Attention (SGA) module. The matrix multiplication is replaced by sparse addition and mask operations. The linear complexity enables all-pair node in teractions on large-scale graphs with limited GPU memory. To our knowledge, our work is the first attempt to introduce SNNs into the attention of Graph Transformers. Furthermore, we design SpikeGraphormer, a dual-branch architecture, combining a sparse GNN branch with our SGAdriven Graph Transformer branch, which can simultaneously perform all-pair node interactions and capture local neighborhoods. SpikeGraphormer consistently outperforms existing state-ofthe-art approaches across various datasets and makes substantial improvements in training time, inference time, and GPU memory cost (10 similar to 20 & times; lower than vanilla self-attention). It also per forms well in cross-domain applications (image and text classification). We release our code at SpikeGraphormer.

To address the control challenges posed by strong nonlinearity, rapid time-varying characteristics, and overactuation of heterogeneous actuators in hypersonic morphing vehicles, this paper proposes a composite control scheme that integrates nonlinear disturbance rejection control with differential geometric allocation theory. The controller is directly designed based on the original six-degree-of-freedom nonlinear model, thereby avoiding mismatches caused by model linearization. An adaptive super-twisting extended state observer is constructed to achieve fast and accurate estimation of morphing-induced perturbations and external disturbances, while an adaptive sliding mode controller is employed to ensure system robustness and effectively suppress chattering. Furthermore, a Pareto front-based differential geometric control allocation theory is proposed, which transforms the conventional weight-dependent constrained optimization problem into an unconstrained multi-objective optimization on a Riemannian manifold, enabling an autonomous balance between control smoothness and energy consumption. Simulation results demonstrate that the proposed method offers significant advantages in precise attitude tracking and synergistic allocation among heterogeneous actuators for hypersonic morphing vehicles.

This paper proposes a novel intelligent anti-saturation attitude control framework for near-space hypersonic vehicles (NSHVs) under multi-constraint conditions, including actuator magnitude/rate limits and aerodynamic uncertainties. The framework integrates an adaptive nonsingular terminal sliding mode control with a Twin Delayed Deep Deterministic Policy Gradient (TD3)-based online gain optimization mechanism. A two-time-scale hierarchical structure is developed, where a nonlinear gain-based anti-saturation mechanism is embedded directly into the inner-loop control law to enforce actuator constraints while preserving finite-time convergence. Adaptive laws are designed to estimate lumped uncertainties online, eliminating the need for prior disturbance bounds. Furthermore, a TD3 agent is employed to dynamically optimize the learning rate matrices of the adaptive laws in real time, using a composite reward function that balances tracking accuracy, overshoot suppression, and control effort. Rigorous Lyapunov-based stability analysis is provided, treating the TD3-updated gains as slowly time-varying parameters, and proving that all closed-loop signals are uniformly ultimately bounded. Extensive simulations under varying angle-of-attack commands, altitude-dependent dynamic pressure conditions, full-speed envelope responses, and ±20% aerodynamic perturbations demonstrate that the proposed scheme achieves high-precision tracking, strict constraint satisfaction, and superior robustness compared to conventional methods.

In this paper, a sliding mode guidance law for impact angle control without violating a seeker's field-of-view limits is proposed against targets with various motions and unknown acceleration, including stationary, constant-velocity moving and manoeuvring targets. To develop the guidance law, the kinematic conditions for engagement geometry are defined, and the sliding mode control is applied to satisfy the homing constraint and impact angle control. Then, the relation between look angle, the desired line-of-sight angle and desired impact angle is established to guarantee the target in the field of the missile's view. The stability of the proposed approach is analysed using Lyapunov theory. Furthermore, the look angle is examined to verify the field-of-view constraint, and a capturability analysis is conducted. To evaluate the performance of the proposed law, numerical simulations demonstrate that the proposed approach achieves satisfactory miss distance and impact angle error while adhering to the field-of-view limit.

Hypersonic morphing vehicles enhance aerodynamic performance through dynamic external configuration adjustments. By integrating morphing commands as dynamic control inputs rather than predefined ones, the vehicle achieves improved adaptability to varying flight conditions. However, this introduces significant coupling between morphing and attitude control loops. This paper proposes the Morphing-Aerodynamic Collaborative Control Algorithm (MACC), which incorporates morphing parameters into the control input vector, enabling autonomous adaptation to environmental changes.The MACC framework leverages dynamics and aerodynamic models for hypersonic vehicles with variable wingspan and sweep angle. The control layer uses a nonsingular fast terminal sliding mode controller to generate robust virtual commands. A third-order tracking differentiator with fixed-time convergence ensures precise guidance command tracking, offering rapid responses and maintaining fixed-time convergence within predefined bounds. Additionally, a fixed-time convergent extended state observer (ESO) estimates system disturbances, providing real-time compensation for unmodeled dynamics and external perturbations. The observer parameters are analytically derived to simplify tuning and reduce implementation complexity.Control allocation is optimized using a modified sequential quadratic programming algorithm. Numerical simulations validate MACC’s effectiveness in attitude control, and Monte Carlo analysis confirms its robustness under severe aerodynamic disturbances. The fixed-time convergence properties of the tracking differentiator and ESO improve transient performance, ensuring rapid disturbance attenuation even with abrupt configuration changes.

To enhance the real-time performance and accuracy of guidance command generation, we propose an online reentry guidance algorithm based on analytical solutions of the hypersonic glide trajectory (HGT). Initially, an altitude-velocity profile is designed in the longitudinal plane to satisfy both path and terminal constraints. Based on this profile, we derive analytical solutions for the flight path angle (FPA) and bank angle. Subsequently, by employing the Newton-Raphson method to linearize the reentry motion equations, analytical solutions for the latitude and heading angle are obtained. Furthermore, we introduce an improved particle swarm optimization (IPSO) algorithm to optimize the profile parameters. This approach significantly enhances the algorithm's global convergence by narrowing the parameter optimization range and adaptively adjusting the inertia weight and cognitive factors. Finally, we present an online guidance algorithm that combines the HGT analytical solutions with the IPSO algorithm. This algorithm effectively achieves lon-gitudinal and lateral guidance by continuously updating the altitude-velocity profile and bank angle symbol in real time. Simulation results demonstrate that the proposed algorithm is fast, efficient, accurate, and holds significant potential for broader application.

A novel fixed-time cooperative guidance law is developed to enable multiple flight vehicles to simultaneously intercept various target motions with desired impact angle, including stationary, constant-velocity moving and manoeuvring targets, with a unified guidance structure that requires no mode switching or target motion classification. First, a fixed-time distributed cooperative guidance law is developed using the theory of multi-intelligence cooperative control, which is formulated along the line of sight (LOS) direction. This approach ensures that the impact time is regulated, enabling multiple flight vehicles to intercept the target simultaneously within a fixed time. In the second phase, employing a sliding manifold and a fixed-time reaching law, the normal acceleration of each flight vehicle is designed to ensure that the LOS angle converges to its target value within a fixed duration. Unknown components of the target's acceleration are estimated via fixed-time observers and incorporated into the guidance commands, enhancing precision and robustness in the guidance process. In conclusion, the proposed approach proves the effectiveness and superiority through numerical simulations, including various target motions, switching communication topology, robustness against uncertainties based on Monte Carlo experiment and comparative studies.

This paper proposes a two-phase game guidance strategy for the three-body confrontation scenario according to the linear quadratic differential game method, which includes an Attacker, an Interceptor, and a Target. The interception probabilities between the Attacker and the Interceptor-Target team are estimated using the probability density function. The desired zero-effort miss distances associated with the interception probabilities for the three-body conflict are acquired by virtue of the gradient descent method. The game combat is divided into two phases by introducing the switching time. In Phase 1, a differential game strategy is developed to guide the zero-control miss distance between the Attacker and the Interceptor-Target into the desired position, which guarantees that the Attacker has the maximizing probability of intercepting the Target and the minimizing probability of being captured by the Interceptor. In Phase 2, a differential game guidance strategy is proposed to ensure that the Attacker evades the pursuit of the Interceptor and intercepts the Target at the preset impact angle. Finally, numerical simulation verifies the effectiveness of the two-stage game guidance strategy.

This paper addresses the inclination adjustment needs of high lift-to-drag ratio spacecrafts in low Earth orbit by proposing an aeroassisted orbital maneuver method based on multiple passes through the dense atmosphere. With fuel savings as its core objective, the method systematically designs reentry velocity and reentry angle profiles that satisfy strict path constraints, while determining the spatial position and motion state parameters for each reentry based on the initial orbit. During each atmospheric pass, the vehicle maintains a maximum constant bank angle. Upon exiting the atmosphere, the established three-impulse optimization model is applied to perform orbital corrections in the spatial phase. By iterating through the reentry-exit cycle, significant orbital inclination adjustments can be achieved. Simulation validation results demonstrate that, compared to traditional purely propulsive orbital transfer methods, the proposed approach significantly reduces fuel consumption.

With the continuous development of the times, hypersonic technology has become one of the focuses of current aviation research. The process of multi-body separation/load release is ubiquitous. In the process, the lumped uncertainties, including parameter uncertainty, unmodeled dynamics, and unknown disturbances, makes the design of the controller difficult. In order to maintain the stability of the lateral motion of the aircraft during the separation process, a backstepping control method based on command filter and neural network is proposed. A dual command filter to obtain the derivative of the differential signal between the virtual input variable and the virtual control law; Traditional neural network-based backstepping controller requires multiple neural network, and this kind of controller will lead to a more complex and fragile system. Our method requires only a single neural network to approximate matched uncertainties, effectively addressing the complexity and fragility issues inherent in traditional multi-neural-network-based backstepping controllers. Finally, the effectiveness of the designed separation control method is confirmed by simulation.

Conventional analytical approaches for reentry glide trajectories often assume fixed flight profiles, limiting maneuverability. To overcome this, this paper presents an analytical method for three-dimensional glide-phase trajectory generation based on an altitude-velocity profile. The trajectory is decomposed into longitudinal and lateral components. The longitudinal motion is derived directly from the altitude-velocity relationship, while the lateral motion is modeled using a generalized latitude-longitude framework. A perturbation method breaks the lateral model into a sequence of lower-order sub-models, each solved recursively. Lagrange interpolation is used to approximate the integral terms, yielding an efficient analytical solution. Simulations confirm the method’s accuracy, speed, and flexibility for diverse lateral maneuvers, highlighting its practical potential.

Currently, protein–protein interaction (PPI) networks have become an essential data source for protein function prediction. However, methods utilizing graph neural networks (GNNs) face significant challenges in modeling PPI networks. A primary issue is over-smoothing, which occurs when multiple GNN layers are stacked to capture global information. This architectural limitation inherently impairs the integration of local and global information within PPI networks, thereby limiting the accuracy of protein function prediction. To effectively utilize information within PPI networks, we propose GTPLM-GO, a protein function prediction method based on a dual-branch Graph Transformer and protein language model. The dual-branch Graph Transformer achieves the collaborative modeling of local and global information in PPI networks through two branches: a graph neural network and a linear attention-based Transformer encoder. GTPLM-GO integrates local–global PPI information with the functional semantic encoding constructed by the protein language model, overcoming the issue of inadequate information extraction in existing methods. Experimental results demonstrate that GTPLM-GO outperforms advanced network-based and sequence-based methods on PPI network datasets of varying scales.

A two-stage game guidance strategy is designed in this article using the optimal control method in a two-on-two engagement scenario, which consists of a defender (low-value aircraft), a target (high-value aircraft), and two attackers. The probability of the attackers intercepting the target is assumed to devise an impact point in the engagement. The greater the probability, the closer the impact point is to the attacker. Therefore, the engagement is divided by the impact point into two stages. The first stage is a three-player confrontation between the defender, the target, and the impact point, where the defender remains on the line-of-sight between the target and the impact point. The defender is closer to the attacker by pursuing the impact point to increase the success rate of combat. For the second stage, the defender intercepts the attacker posing the greatest threat to the target at the desired terminal angle while the target adopts an optimal avoidance strategy. Simulations are performed to verify the effectiveness and feasibility of the guidance strategy.

In this paper, an orbit design method is presented to design a circle orbit for a single spacecraft without maneuvering to visit multiple targets with the same altitude. Based on the necessary condition of the flight time in circle orbit, one of the visit times can be solved when another visit time is determined. Then, an orbit design to visit three targets is translated into the minimum required velocity difference in the same position, and the Newton iteration method is utilized to find a high-precision result. Simulation results illustrate the effectiveness of the orbit design methods.

To delineate a planar engagement scenario through mathematical expressions, the paper first establishes the connection between the desired line-ofsight and flight-path angle. Unlike conventional guidance laws that assume zero target maneuvering, which often results in significant impact angle errors, this approach employs an extended state observer to enhance tracking performance. Utilizing the error of angle, a sliding mode is devised to design the guidance law, with the stability of the proposed method validated through Lyapunov theory. The effectiveness and robustness of the method are demonstrated through numerical scenarios. This method offers precise estimation of target maneuvering and facilitates interception.

In the context of the rapid evolution of autonomous driving technology, the deployment of autonomous vehicle systems has witnessed significant expansion, underscoring the paramount importance of their path planning capabilities. This study proposes an improved Q-learning, specifically designed for the path planning of autonomous vehicles. Initially, the algorithm is enhanced with an advanced artificial potential field method for initializing the Q-table, thereby integrating prior knowledge and increasing the efficiency of the initial iterations. Further, the integration of eligibility traces as a memory mechanism within the algorithm ensures the prolonged influence of state-action pairs throughout the iterative process, thus boosting learning efficiency. Additionally, this paper introduces a dynamic λ adjustment strategy, tailored to meet the exploratory demands of autonomous vehicles and based on the simulated annealing algorithm. This strategy dynamically modulates the decay factor λ in accordance with the frequency of successful arrivals at the destinations, promoting more effective adaptation to varying environmental conditions. The empirical findings confirm the efficacy of the proposed Q-learning, particularly highlighting its utility in the path planning for autonomous vehicles.

In this paper, for the two/three targets in the circle orbit with the same altitude, a numerical method is presented to design an orbit for a spacecraft to visit multiple targets without maneuvering. First, for the two targets' case, an equation is established by the relationship of the two angles that both are the function of the time, one of the angles is the angle between two targets' geocentric position vectors at two times, and another one is the spacecraft flight angle during the two times. The visit time can be solved by the assumption that one of the visit times is known in the equation, and the orbit element can be transferred from the position and velocity vector determined by the solved visit time. Then, for the three targets' case, the problem is decoupled into two cases of two targets, and there is a public target in each case. Based on the method for two targets, two different orbits can be obtained by determining the visit time of the public target. The difference between two orbits can be expressed by the difference in velocity vectors of the public target at different times. A sampling method is utilized to select two velocity vectors with relatively small differences as initial values, the Newton iteration method is used to obtain visit times with higher accuracy. Simulation results illustrate the effectiveness of the orbit design methods.

With the rapid advancement of autonomous vehicle technology, the development of efficient and stable trajectory tracking algorithms is crucial for achieving autonomous navigation. This paper addresses the issue of insufficient stability in traditional MPC for vehicle trajectory tracking and proposes a Q-learning-based multi-step model predictive control (QM-MPC) algorithm. This algorithm integrates the principles of model predictive control and reinforcement learning to accurately predict the future motion trajectory of autonomous vehicles using MPC. Furthermore, it utilizes Q-learning to perform multi-step deep predictions on the MPC-predicted motion trajectory results and propagates these deep predictions back to the current state. Through this approach, the objective cost function of MPC for autonomous vehicles is optimized, thereby further enhancing the efficiency and stability of trajectory tracking. Experimental results demonstrate that the proposed QM-MPC algorithm exhibits robust stability of Unmanned Ground Vehicle (UGV) in trajectory tracking. Specifically, it significantly enhances stability by 4.303

A massive amount of protein sequences have been obtained, but their functions remain challenging to discern. In recent research on protein function prediction, Protein-Protein Interaction (PPI) Networks have played a crucial role. Uncovering potential function relationships between distant proteins within PPI networks is essential for improving the accuracy of protein function prediction. Most current studies attempt to capture these distant relationships by stacking graph network layers, but performance gains diminish as the number of layers increases. To further explore the potential functional relationships between multi-hop proteins in PPI networks, this paper proposes SEGT-GO, a Graph Transformer method based on PPI multi-hop neighborhood Serialization and Explainable artificial intelligence for large-scale multispecies protein function prediction. The multi-hop neighborhood serialization maps multi-hop information in the PPI Network into serialized feature embeddings, enabling the Graph Transformer to learn deeper functional features within the PPI Network. Based on game theory, the SHAP eXplainable Artificial Intelligence (XAI) framework optimizes model input and filters out feature noise, enhancing model performance. Compared to the advanced network method DeepGraphGO, SEGT-GO achieves more competitive results in standard large-scale datasets and superior results on small ones, validating its ability to extract functional information from deep proteins. Furthermore, SEGT-GO achieves superior results in cross-species learning and prediction of the functions of unseen proteins, further proving the method’s strong generalization.