This paper proposes a topology-driven approach for consensus of linear heterogeneous multi-agent systems under distributed denial-of-service attacks launched by multiple adversaries. As more sophisticated cyber-attacks have emerged, considerable research interest has recently focused on consensus control under distributed DoS attacks. To address this issue, a novel framework is developed that represents different attack modes using a directed graph, thereby incorporating the directional information of attacked communication channels. Moreover, the topology-driven approach enables dynamic design of the gain matrix according to each attack mode. Based on this framework, a distributed state-feedback controller is constructed, where the standard assumption on system matrices is relaxed. An iterative mode-triggered algorithm is developed, and its performance is verified using numerical and practical case studies.
Reliable state estimation is essential for legged robots operating in complex environments. Although exteroceptive sensors, such as cameras and LiDAR, provide valuable environmental constraints, their reliability can degrade under visual occlusion, motion blur, and geometric degeneracy. Therefore, proprioceptive odometry remains important as a continuous estimation backbone for legged locomotion. However, existing proprioceptive methods often rely on ideal point-contact assumptions or fixed contact confidence, causing the resulting kinematic constraints to become inconsistent or overconfident under rolling motion, foot slippage, and time-varying foot–ground interactions. To address these challenges, this article proposes a hierarchical leg-wise interacting multiple model (HL-IMM) proprioceptive odometry framework. At the local layer, a pair of rolling-consistency residuals is constructed for each stance leg to infer leg-wise contact reliability. At the global layer, a two-model IMM estimator maintains a nominal rolling-consistent ESKF branch and a high-uncertainty ESKF branch. Unlike single-model adaptive-noise schemes, HL-IMM preserves distinct contact-conditioned state hypotheses. The legwise posteriors modulate the contact uncertainty of individual legs, thereby preserving leg-level contact selectivity. The local residual likelihoods are then aggregated to update the global mode probabilities, enabling robot-level contact-mode inference. Extensive simulation and real-world experiments demonstrate improved pose estimation accuracy and robustness across diverse scenarios with moderate computational overhead.
This paper investigates the gradient-free distributed generalized Nash equilibrium (GNE) seeking problem over unreliable communication networks, where each player can only exchange information with its neighbors over some disjoint time intervals. A constrained N-player noncooperative game, where the feasible action set of each player is subject to a coupled equality constraint and a local inequality constraint, is first formulated. To solve this game in the presence of the network unreliability, a resilient distributed GNE seeking algorithm is developed by leveraging the extremum seeking (ES) technique and an average dwell time (ADT) method incorporated by frequency and duration properties of disconnected topologies. Unlike the existing gradient-based protocols that rely on the explicit expressions of local payoff functions, the proposed algorithm is gradient-free by only utilizing the real-time measured payoff values. Moreover, Lagrangian multipliers and penalty terms are designed to address the coexistence of coupled and local constraints. It is proved that the developed algorithm is capable of steering the actions of all players to a small neighborhood of the GNE exponentially. Numerical examples on a price-driven smart charging game are given to verify the effectiveness of our proposed method.
In the field of marine exploration, the demand for high-precision underwater detection based on autonomous underwater vehicles (AUVs) has been steadily increasing. This paper proposes an automated method for the three-dimensional reconstruction of underwater targets using sequences of two-dimensional forward-looking sonar (FLS) images. First, a target domain segmentation method is constructed to reduce the impact of environmental noise on the reconstruction results. Subsequently, a Cartesian coordinate reconstruction model is established to build the accurate correspondence between the sonar image sequences and the target entities. Compared with existing methods, the key advantage of the proposed method lies in its consideration of the complex motion scenarios of AUVs during actual operations, which enhances the environmental adaptability and application universality of the method. Through numerical simulations and tank experiments, the method is comprehensively verified from multiple dimensions, including mechanism analysis, performance evaluation, parameter sensitivity, motion robustness, and position error sensitivity. Moreover, field experiments are further utilized to thoroughly test the stability of the method in real environments. The results demonstrate that the proposed method is highly resilient to practical challenges such as limited target visibility, non-vertical aperture motions, and position estimation errors.
Off-road autonomous driving perception technology faces numerous challenges, including irregular obstacles, significant terrain variations, and extreme weather conditions. These factors complicate the direct application of established on-road perception technologies to off-road environments. An increasing number of researchers are focusing on off-road perception technology, achieving notable progress in this field. This paper provides a comprehensive survey of recent advancements in off-road perception technologies, concentrating on terrain classification, traversable area detection, obstacle detection, and high-precision localization. Particular attention is given to the transferability of on-road perception methods to off-road scenarios, terrain-specific traversability analysis, GNSS-denied localization, adverse-weather robustness, and the practical limitations of representative methods. It also introduces essential datasets, simulation environments and hardware platforms for off-road autonomous driving, serving as a valuable reference for researchers. A detailed analysis of the representative literature is conducted, assessing research hotspots and trends in off-road perception technology using scientometric methods. While onroad perception technology is relatively mature, this paper briefly outlines its key aspects, emphasizing the adaptability required for off-road scenarios. Finally, future directions are proposed to address the challenges facing off-road perception technologies.
