This paper studies the distributed consensus protocol for the connected vehicle platoon with heterogeneous time-varying delays and switching topologies. A third-order dynamics model with powertrain inertial lag is proposed to characterize the node longitudinal dynamics of vehicles in platoon. A novel distributed adaptive consensus protocol considering the time-varying delays and the random switched inter-vehicular communication topologies is designed to stabilize the heterogeneous vehicle platoon in the presence of external disturbance. The delay-range-dependent approach is used to deal with the system heterogeneous time-varying delays by considering the characteristics of the heterogeneous platoon. Directed graphs are adopted to describe the accessible information flow among vehicles. The necessary and sufficient conditions for the unified closed-loop vehicle platoon system are derived by using matrix analysis and Lyapunov-Krasovskii approach. Numerical simulations demonstrate the proposed method is effective.

Road traffic accidents, especially those accidents with multiple-vehicle collision usually cause injuries and mortalities. Currently, studies on collision avoidance mainly focus on the control strategies for adjacent two vehicles or multiple vehicles in a single platoon direction. This paper proposes a bi-directional collision avoidance system for multiple vehicles to minimize the collision risk under the model predictive control (MPC) framework through switching the vehicle-following mode based on the inter-vehicular states. A hierarchical structure with an upper layer and a lower layer is designed. A dual-operational mode switching strategy and the vehicle-following model are formulated in the upper layer, together with the development of the acceleration redistribution cooperative strategy for vehicle platoon. While the lower layer is designed to track the desired acceleration received from the upper layer by considering the practical situation of the control commands. To tackle complex transitional operation, a dynamic weighted tuning strategy is proposed and integrated it with the MPC. The numerical results show that the proposed system outperforms the conventional collision avoidance system and is effective to avoid a collision or minimize the total impact of the vehicle platoon when the collision is unavoidable.

Coordinating conflicting movements at unsignalized intersections requires simultaneously achieving collision-free safety and high traffic efficiency under strongly coupled multi-vehicle interactions and vehicle dynamic constraints. Existing multi-agent reinforcement learning (MARL) methods face critical challenges, including unstable value estimation caused by delayed rewards, as well as credit ambiguity and counterfactual infeasibility under shared reward setups. To address these issues, we propose a cooperative MARL framework tailored for collaborative vehicle passage at unsignalized intersections. Built upon the multi-agent deep deterministic policy gradient (MADDPG) algorithm, the proposed approach introduces historical context into global value estimation to enhance the reliability of capturing and evaluating temporal dependencies. Furthermore, it designs a dynamic counterfactual coalition value construction mechanism that adheres to vehicle dynamic constraints. Under a consistent temporal context, individual marginal contributions are computed and used for policy updates, thereby obtaining stable and interpretable explicit credit signals while mitigating the free-rider effect. Experimental results demonstrated that the proposed method achieves superior cooperative performance and better robustness than conventional approaches in terms of collision rate, success rate, and traffic efficiency.

This paper proposes a distributed cooperative control strategy to address the stability problem of connected vehicle platoon subjected to stochastic switching communication topologies, time-varying communication delays, denial-of-service (DoS) attacks, and external disturbances. Based on third order vehicle longitudinal dynamics, a multi-vehicle closed-loop system model is established that integrates Markovian random topology switching, DoS attacks, and uniformly bounded time-varying delays. Meanwhile, a distributed cooperative control protocol relying on relative state information is designed. By employing Lyapunov–Krasovskii functionals, Markovian jump system theory, and matrix inequality techniques, the dynamic evolution of the multi-vehicle closed-loop system is analyzed for both normal communication intervals and DoS attack intervals, and sufficient conditions for mean-square exponential stability of the closed-loop system are derived. When the feedback gain is given, the obtained conditions can be converted into linear matrix inequalities (LMIs); when the feedback gain is to be designed, an alternating-iteration-based solution method for bilinear matrix inequalities (BMIs) is further developed. Finally, numerical simulation results demonstrate that the proposed method exhibits satisfactory overall performance under complex communication conditions and maintains the driving safety of the vehicle platoon.

Reliable cooperative control of connected vehicle platoons remains challenging when stochastic communication-topology variations coexist with heterogeneous time-varying transmission delays. This study investigates a distributed consensus control problem for connected vehicle platoons subject to Markovian switching topologies and heterogeneous time-varying inter-vehicle communication delays. A third-order longitudinal vehicle model with powertrain inertial lag is combined with a continuous-time Markov chain (CTMC) to describe randomly switching communication topologies. A distributed control protocol incorporating the constant time headway policy, full velocity difference car-following behavior, and leader-pinning information is formulated, while link-dependent time-varying delays are embedded in the closed-loop error system. Based on a mode-dependent Lyapunov–Krasovskii functional, Jensen’s integral inequality, and the Schur complement, sufficient linear matrix inequality (LMI) conditions are derived for mean-square asymptotic stability of the nominal delayed system; the LMI is feasible at ${\overline{\tau }}_{\mathsf{\ell }}=0.10$ s and ${\mu }_{\mathsf{\ell }}=0.04$, remains feasible at $0.30$ s, and becomes infeasible at $0.375$ s. An input-to-state stability (ISS) corollary extends the result to bounded time-varying leader acceleration, showing that tracking errors remain bounded during dynamic EUDC phases and converge when the leader acceleration and its derivative vanish. Using the same Lyapunov–Krasovskii certificate, a mean-square $L_{\infty }$ bounded-propagation result is further established; this result characterizes uniform-in-vehicle error bounds and does not assert strict non-amplification. Numerical simulations under the Extra-Urban Driving Cycle (EUDC) show that the follower vehicles maintain cooperative tracking under topology switching and bounded communication delays, with position, velocity, and acceleration errors remaining bounded during the dynamic phases and converging as the leader acceleration vanishes. In the EUDC simulation, the largest leader-relative position error is approximately $-60.75$ m for Vehicle 10 at $t\approx 380$ s, the maximum adjacent local spacing-error peak ratio is $2.32$, and the position error returns below 1 m by $t\approx 400.8$ s. The minimum inter-vehicle gap is $5.87$ m and the minimum safety margin after the initial 10 s is $0.84$, with all safety-margin indices remaining positive; these results indicate that the proposed framework maintains platoon coordination and car-following safety under stochastic communication uncertainties.

Most existing dynamic Simultaneous Localization and Mapping (SLAM) methods, when integrating neural networks, typically rely on high-performance computing support, and their tracking threads often need to wait for semantic results. This paper proposes a SLAM approach based on a lightweight neural network and an asynchronous thread decoupling mechanism, which enables dynamic SLAM without waiting for neural network feedback, making it suitable for resource-constrained platforms. The neural network consists of a Re-parameterized Feature Representation Backbone (RFR-Backbone), a Dynamic Enhanced High-Speed Feature Pyramid Network (DEHS-FPN), and a Cross-Scale Shared Decoupled Detection Head (CSSD-Head). Compared to the YOLO11n model, the proposed network reduces the number of parameters by 48.6% and computational cost by 38%. The decoupling mechanism asynchronously separates the semantic thread from the tracking thread and introduces a dynamic mask prediction mechanism based on optical flow propagation. This allows the tracking thread to update dynamic regions without waiting for detection results, effectively avoiding the latency caused by traditional sequential semantic reasoning. Experimental results on the TUM RGB-D dynamic dataset show that the proposed method significantly outperforms ORB-SLAM3 and several representative dynamic SLAM approaches across various dynamic scenarios. The system achieves stable single-frame processing times of 38–43ms per frame using only a CPU, surpassing many GPU-dependent dynamic SLAM methods.