Shixi Wen

3D Gaussian Splatting can now train a full scene in minutes, but training is only the first step of deployment. A usable model must also render fast, stay small, transfer cheaply, and keep the quality the short budget bought. Existing acceleration systems stop at fast training, and none answers whether quality headroom remains under a minute-scale budget. This paper completes the delivery chain on Mini-Splatting~2's 18k protocol. Measurement-placed routing ships an importance gate and an optimizer schedule only on the splits where isolated measurement shows they pay. A fused backend renders faster, and a training-free compression stage shrinks each model to $\approx$33~MiB at near-lossless quality in seconds on CPU. Every benchmark scene passes a strict per-scene quality gate on the multi-seed means against the bare official configuration, and the whole chain trains fastest among the reproduced systems on the same machine. A pre-specified audit of fourteen quality-gain configurations, literature mechanisms and working-point probes alike, finds none that passes its gate, which is why the training signal stays frozen. These properties suit radiance-field services that train and deliver on demand, and we will release the code and the multi-seed reference means.

This study investigates a neural network (NN)-based distributed resilient control problem for the uncertain nonlinear vehicular platoon subject to unknown deception attacks and actuator faults. Here, we consider the case that the state information of the preceding vehicle transmitted via VANETS is subject to a more general deception attack, which combines with both additive and multiplicative attack models. To mitigate the effect of the unknown deception attack, a new transformation is introduced for the compromised tracking errors to establish a new change of coordinates. Then, a novel distributed NN resilient controller is proposed for the nonlinear vehicular platoon based on neural network technique and fractional-order command-filter backstepping technique. It has the advantage of removing the assumptions for certain parameters of vehicle dynamics, known actuator faults and attacks, and the positive sign of the attack weight. It can guarantee that the original tracking errors are uniformly ultimately bounded to an adjustable neighborhood of zero within a finite time. Moreover, the robust string stability can also be satisfied. Both numerical simulations and experiments with laboratory-scale minicars are presented to illustrate the effectiveness and practicality of the proposed resilient controller.

This paper presents a joint communication-control optimization framework for connected vehicle platoons subject to LTE resource constraints, dynamic topology switching, and control input constraints. To ensure platoon stability and tracking precision, a novel two-phase optimization framework is proposed that synergistically coordinates Cooperative Awareness Messages (CAMs) broadcast scheduling with distributed optimal control strategies. Confronting the NP-hard nature of this co-design problem, a hierarchical suboptimal approach is developed. First, a dynamic error-aware assignment mechanism is established, where an enhanced greedy algorithm is proposed to regulate CAMs broadcast frequencies (1-10 Hz) based on real-time tracking errors of following vehicles. Subsequently, an optimal tracking controller is designed to minimize spacing errors while considering string stability through a nonconvex optimization technique. Theoretical insights reveal the intrinsic coupling between communication reliability, CAMs broadcast frequency, and platoon control performance. Simulation examples with comparative results are presented to illustrate the effectiveness of the proposed co-design algorithm.

The fixed-time control is investigated for an uncertain two-dimensional (2-D) planar vehicular platoon system with global prescribed performance. First, a novel fixed-time prescribed performance control (FxTPPC) method is designed to guarantee the tracking errors converge to the prescribed region with small overshoot inside a predefined time. Moreover, the restriction of PPC method that relies on the initial condition of tracking errors is also released, which means global PPC can be achieved. Then, a new control scheme, by virtue of the composite fixed-time sliding mode surface, is constructed for the third-order 2-D plane platooning system, such that the tracking errors, including distance errors and heading errors, approach to a predetermined region in a given time, while avoiding the singularity problem and improving the convergence rate of the system. In addition, fixed-time string stability and reachability of fixed-time prescribed performance are also achieved. Finally, by constructing numerical simulations and experiment studies, the validity of the proposed scheme is demonstrated in the end.

This study investigates the online optimization control problem for a vehicular platoon in the presence of unmodeled vehicle dynamics, unknown external disturbances, and uncertain inter-vehicle communication topology. A novel practical performance-prescribed reinforcement learning-based distributed sliding mode (PRLDSM) control framework is constructed for the platoon to strengthen the robustness and possess the online self-learning capacity for optimizing and control. Specifically, a composite controller is proposed for the platoon which consists of an online optimal controller and a sliding mode controller. By the strong learning capacity provided by the PRLDSM controller, the optimal policy and cost function can be recursively approximated by online simultaneous tuning of both actor and critic neural networks. Moreover, it is proved that all the signals in the closed-loop platoon control system are uniformly ultimately bounded and approach zero by selecting appropriate parameters. Theoretical analysis eventually guarantees that the tracking errors can be stabilized to satisfy both the individual vehicle stability and disturbance string stability with the prescribed transient response and steady-state accuracy. The obvious feature of the proposed PRLDSM controller is that the requirement of the exact vehicle dynamics and topological matrix of the inter-vehicle communication topology can be avoided. Both numerical examples and experimental studies illustrate the effectiveness of the proposed control method.

This paper investigates the platoon control of heterogeneous vehicular cyber-physical systems (VCPSs) subject to external disturbances by using neural network and uniformly quantized communication data. To reduce the adverse effects of quantization errors on system performance, a coupling sliding mode surface is established for each following vehicle. The radial basis function (RBF) neural networks are employed to approximate the unknown external disturbances. Then, a novel platoon control law is proposed for cooperative tracking in which each following vehicle only uses the uniformly quantized data of the neighboring vehicles. And the designed controllers in this paper are fully distributed due to the fact that the selection of each vehicle's controller parameters is independent of the entire communication topology. The string stability of VCPSs in the entire control process is ensured rather than only ensuring the string stability after the sliding mode surface converges to zero. Compared with the existing controller design methods and quantization mechanisms, the neural adaptive sliding-mode platoon controller proposed in this paper is superior in performances including tracking errors, driving comfort and fuel economy. Numerical simulations illustrate the effectiveness and superiority of the designed control strategy.

Sensor and actuator attacks can alter the truth values of vehicle states and control input through false data injection, leading to performance degradation. In this paper, secure control of vehicular platoon control systems (VPCS) subject to joint sensor-actuator attacks, and the unknown disturbances is investigated. First, a novel adaptive sensor fusion algorithm is designed to estimate the actual position with smaller fusion errors. Subsequently, a predefined-time extended state observer (PTESO) is further constructed based on the estimated position to recover the continuous-time states (i.e., velocity and acceleration) and disturbances accurately within a user-defined settling time. Then, a variable exponent predefined-time sliding mode controller (VPTSMC) with few design parameters is developed to guarantee the predefined-time stability of the platoon system, namely string stability and individual vehicle stability. Meanwhile, the singularity phenomenon is also avoided. Finally, a serious of simulations are carried out to show the effectiveness of the suggested control scheme.

Sensor and actuator deception attacks often manipulate states and control commands, which lead to system performance deterioration or even instability. This paper focus on the platoon control problem for a group of connected and automated vehicles (CAVs) subject to both deception attacks and external disturbances. Firstly, modeling both attacks and external disturbances as lumped disturbances, and a novel adaptive fixed-time disturbance observer (AFxTDO) without requiring prior knowledge of the disturbance boundary is further constructed to estimate the lumped disturbances within given time with zero estimation errors. Then, a new global fixed-time stability result with faster convergence rate is developed, together with the given AFxTDO, a variable exponent fixed-time sliding mode control (VFxTSMC) scheme is established, such that the platoon tracking errors can converge to a predetermined region within the given time while avoiding the singularity phenomenon, improving the convergence speed and reducing the number of controller parameters. Meanwhile, individual vehicle stability and string stability also can be guaranteed within fixed-time by the suggested control scheme. Finally, the simulation results reveal the effectiveness and superiority of the proposed algorithm.

This study proposes the reinforcement learning adaptive optimal (RLAO) control scheme for vehicle platoons subject to unmodeled vehicle dynamics, unknown external disturbances and uncertain inter-vehicle communication topology. By combining the actor-critic RL algorithm with adaptive backstepping technique, a novel RLAO control framework is established for each follower to stabilize the tracking errors and optimize the control performance. The proposed RLAO controller does not depend on the exact topological matrix and vehicle dynamics, as only the input and output data from the tracking error dynamics are involved in updating the control input. Based on the optimal control and dynamic programming, a useful algorithm is proposed to implement the RLAO controller such that all the error signals of the platoon control system are ultimately uniformly bounded. Particularly, the tracking errors can converge to a small neighborhood of zero by setting appropriate parameters. Furthermore, the platoon disturbance string stability is guaranteed. A comparative analysis based on simulation examples is discussed to illustrate the effectiveness of the proposed RLAO controller.