This article presents an adaptive robust control design based on the diffeomorphism approach to handle uncertainties and inequality constraints in electromagnetic active powertrain mounts (APMs). The uncertainties, including high-frequency engine excitations and system parameter variations, are characterized via fuzzy set theory. The diffeomorphism approach is further employed to systematically incorporate both equality and inequality constraints into a unified constraint-following control framework. Building upon this formulation, an adaptive robust control strategy is established, ensuring both uniform boundedness (UB) and uniform ultimate boundedness (UUB) of the system. Furthermore, a two-player Nash game is formulated to optimize the control parameters, where the existence of a Nash equilibrium is theoretically established, and an efficient solution method is provided. Finally, the proposed control strategy is applied to the APM, and its effectiveness is verified through numerical simulations.
Robots can generate semantically plausible motions yet fail once physical contact begins, because contact geometry, force, friction, slip, deformation, jamming, and damage risk are local, dynamic, and often unavailable from vision alone. Using robot manipulation as the primary lens, this survey presents a contact-state-centered view of tactile intelligence. We distinguish geometric, mechanical, and dynamic physical states from task-conditioned diagnostics such as stability and jamming, and organize tactile intelligence through an action–contact–feedback loop linking contact-state estimation, action-conditioned consequence prediction, and action adaptation. We further examine how this interface supports tactile policies, vision-language-action (VLA) reasoning, world models and planning, and fast reactive feedback across multiple time scales. Representative manipulation domains, scalable data and simulation resources, benchmarks, and evaluation protocols are reviewed under the same state–prediction–intervention perspective. This formulation connects semantic task generation with embodied physical execution by making postcontact interaction observable, predictable, and correctable.
Aggressive flight is crucial for quadrotors to perform penetration and adversarial missions. However, the uncertainties induced by aggressive flight are more severe, and extreme flight states also pose a challenge to the reliability of uncertainty estimation. Therefore, first, to mitigate the effects of anomalous states and measurement noise on the observer estimation process and to achieve smooth estimation of time-varying uncertainties, an improved adaptive nonlinear extended state observer (ANESO) with peak reduction and noise suppression is designed, and its input-to-state stability (ISS) analysis is executed. Second, the ANESO is integrated into the highly dynamic trajectory-tracking control framework of the quadrotor, providing smooth and reliable estimation guarantees for uncertainty compensation, thereby improving the accuracy and stability of tracking aggressive trajectories under complex uncertainties. Finally, extensive comparative simulations and experiments on aggressive trajectory tracking (with velocities exceeding $15~\text {m}/\text {s}$ and accelerations of 2 g) have demonstrated the superior smooth uncertainty estimation capability and tracking performance of the proposed method.
Cyber–physical systems (CPSs) often suffer from internal system faults and external cyberattacks, which makes the study of secure and privacy-preserving fault-tolerant (FT) control increasingly important and practically relevant. In the physical domain, faults of system components disrupt system stability or seriously degrade control performance. In the information domain, the network architecture and intensive data exchange render the system vulnerable to malicious attacks and eavesdropping attacks. This survey paper presents a review of secure control methods for CPSs from an FT perspective. First, the architecture of FT control in the physical domain is introduced. Then, we present the characteristics of existing malicious attacks and the corresponding secure FT control methods in the presence of coexisting physical faults and malicious attacks. Furthermore, some typical privacy-preserving methods and the existing secure FT control approaches under eavesdropping attacks are analyzed. Finally, existing experimental validation studies on secure FT control are summarized and discussed, and challenging issues are envisioned to facilitate potential future research.
Object detection is a core perception task in autonomous driving, surveillance, robotics, and smart cities. Deep-learning-based single-modality detectors, operating on individual RGB, thermal infrared, depth, or LiDAR streams, have achieved strong performance but still struggle with low illumination, adverse weather, long-range targets, and sensor noise. Cross-modality object detection (CMOD) alleviates these issues by fusing complementary cues from heterogeneous sensors. This survey provides a unified view of deep-learning-based object detection from single-modality to cross-modality regimes. Canonical single-modality detectors are reviewed, including convolutional neural network (CNN)-based two-stage and one-stage architectures and Transformer-based set-prediction frameworks, with emphasis on backbone, neck, and head designs that transfer to CMOD. CMOD methods are categorized into CNN-, Transformer-, and state space model (SSM)-based families, and their fusion-timing patterns, alignment mechanisms, and robustness strategies are analyzed. Representative datasets, metrics, and benchmark results on VOC, COCO, LLVIP, M3FD, and KITTI, spanning both dense RGB–infrared (RGB–IR) and sparse–dense RGB–LiDAR settings, together with typical application domains, are summarized to form a consistent empirical context. Finally, key limitations and future trends are outlined, covering scalable multimodal learning, alignment- and efficiency-aware fusion, and the integration of CMOD with multimodal foundation and vision–language models.
