This paper presents a framework for developing data-driven, control-oriented models of networked systems, i.e., systems that involve many interacting dynamic components. First, a general formulation named the weak latent dynamics model (wLDM) is developed for learning generic nonlinear dynamics with control. Leveraging the weak form, the wLDM enables more numerically stable and computationally efficient training, as well as more accurate prediction when compared to conventional methods such as neural ordinary differential equations. Building upon the wLDM, we propose the weak graph Koopman bilinear form model, which integrates geometric deep learning and Koopman theory to learn latent-space dynamics for networked systems, especially for challenging cases that have multiple timescales. The proposed methods are demonstrated on three examples of increasing complexity, from academic problems to an application of an electrified aircraft energy system, showing that they achieve superior predictive accuracy and training efficiency when compared to baseline models. Parametric studies provide insights into the effects of hyperparameters in the weak form. The proposed framework shows the capability to efficiently capture control-dependent dynamics in these systems, including stiff dynamics and multiphysics interactions, offering a promising direction for learning control-oriented models of complex networked systems.
Visual localization is vital for autonomous systems but remains challenging under dynamic conditions. Transformers offer strong temporal modeling at quadratic cost, while CNNs are efficient yet limited in long-range dependencies. Existing methods also lack robustness to illumination, weather, and seasonal changes, constraining real-world applicability. To address this, this paper proposes AdapseqNet, a dual-branch architecture that integrates stabilized state-space modeling with differential temporal enhancement. First, a stabilized state-space formulation featuring Lyapunov-constrained parameterization and adaptive discretization is proposed, ensuring asymptotic stability and linear computational complexity for reliable processing of extended sequences. Second, a selective Mamba architecture is developed to combine temporal-state modeling with content-aware gating, enabling adaptive feature selection that emphasizes discriminative cues while suppressing redundancy. Third, a differential enhancement module is designed to extract motion-invariant representations through symmetric temporal differencing and LSTM-based refinement, enhancing resilience to appearance variations caused by lighting, weather, and seasonal changes. Beyond architectural design, multi-scale feature fusion and output distribution control are incorporated to optimize representation quality and ensure consistency for similarity-based retrieval. Extensive experiments on multiple benchmarks demonstrate that AdapseqNet achieves a better localization accuracy across diverse and challenging conditions. Note to Practitioners—Visual localization is crucial for autonomous robots but often fails under varying lighting, weather, or seasonal conditions. We propose a dual-path approach: one path captures long-term patterns using control-inspired stable modeling, while the other extracts motion cues that remain consistent despite appearance changes. This combination enables accurate place recognition even in extreme environments. Our system operates efficiently on standard hardware and was tested on an indoor robot, achieving centimeter-level accuracy. This approach can enhance existing navigation systems without requiring additional sensors. Future work will focus on real-time optimization for outdoor deployment.
Controlling dynamical systems in real-time across multiple scenarios is critical to enabling adaptive control strategies, ensuring stability and efficiency. However, to tailor control actions in response to varying scenarios, traditional optimal control problems typically require several system simulations, which are often computationally demanding due to the high-dimensionality of the underlying spatio-temporal dynamics. In this work, we exploit SHallow REcurrent Decoder networks-based Reduced Order Modeling (SHRED-ROM) to synthesize a real-time closed-loop controller for high-dimensional and parametric dynamics, relying solely on limited state sensor readings. After training the model on a few optimal examples given by an expert demonstrator, SHRED-ROM mimics the expert behavior with effective distributed control actions in new scenarios, alleviating the curse of dimensionality. Moreover, a sensor forecaster is synthesized and used to close the loop at the latent level, thus efficiently mitigating possible sensor failures or delays. The performance of the proposed optimal control strategy is finally assessed on three challenging high-dimensional cases dealing with either parametric density control or fluid flow control.
Modeling tethered space systems is critical for advanced orbital operations. Flexible components such as tethers and space nets are integral to these systems but present significant control challenges due to their high dimensional, strongly coupled, and nonlinear dynamics. While data driven methods offer alternative modeling approaches, they frequently struggle with long term predictive stability and spatial generalization. To address this, we propose the Koopman Graph Dynamics (KGD) framework to learn the structural dynamics by integrating the global linear evolution of the Koopman operator with the local topological priors of Graph Neural Networks. Building upon this representation, we develop a KGD based Model Predictive Control strategy for tethered space systems. Subsequently, the ground experiments on flexible tether and space net demonstrate the high precision modeling capabilities of the proposed method. Crucially, the framework exhibits exceptional capacity for spatial transfer without retraining. Models trained exclusively on small configurations successfully predict and control significantly larger, unseen physical scales. Furthermore, the orbit simulations within a physics engine verify the effectiveness of the proposed approach for tethered space systems.
Accurate vehicle trajectory prediction is essential for autonomous driving systems. However, in dynamic traffic environments, existing methods often fail to fully capture complex spatiotemporal interactions, resulting in error accumulation and degraded stability over extended prediction horizons. To address these limitations, we propose a hybrid CNN-MTF-LSTM architecture with an integrated attention mechanism. The model comprises three core components: 1) a dual-branch encoder that separately processes temporal dependencies through LSTM and spatial interactions through CNN; 2) a Bahdanau attention module that dynamically selects the most relevant historical context at each decoding step; and 3) a multi-task fusion decoder, which is trained using a hybrid teacher-forcing strategy, to output kinematically consistent predictions of future positions, velocities, and accelerations. Comprehensive experiments on the NGSIM dataset validate the proposed model, demonstrating robust performance across structured highway driving scenarios. Comparative results against several baseline methods confirm both the accuracy and computational efficiency of the proposed architecture.
Accurate prediction of future vehicle trajectories is essential for ensuring safety and reliable decision-making in autonomous driving systems. However, existing deep learning-based approaches exhibit several limitations. Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) struggle to effectively model long-term temporal dependencies and complex agent interactions, while Transformer-based architectures often suffer from high computational complexity and limited efficiency. To overcome these challenges, this paper proposes an efficient Mamba-based feature extraction framework for jointly encoding vehicle trajectories and map information. By leveraging state-space modeling and a selective scanning mechanism, the proposed approach effectively captures longrange dependencies and enhances the representation of complex traffic behaviors. Specifically, raw scene data are first normalized and embedded into a unified feature space. A Mamba Encoder is then employed to extract high-level features from historical vehicle trajectories and map elements. Subsequently, Vehicle-Vehicle and Vehicle-Map interaction modules are introduced to explicitly model dynamic interactions among traffic participants and between vehicles and the surrounding map. The resulting high-dimensional features are further fused using an additional Mamba Encoder, while a Global Interaction Module is designed to capture scenelevel dependencies. Finally, a Gated Recurrent Unit (GRU) decoder generates multi-modal future trajectory predictions. Experimental results on the Argoverse 1 dataset demonstrate that the proposed method achieves superior performance in terms of minADE, minFDE, and minMR, while maintaining high computational efficiency.OPEN ACCESS Received: 28/01/2026 Accepted: 16/04/2026