Vehicle trajectory prediction from an ego-vehicle perspective is a critical yet challenging problem for intelligent driving systems, particularly in complex traffic environments with uncertain interactions and ego-motion disturbances. Existing methods often rely on global observations, historical ground-truth trajectories, or idealized datasets, which limits their real-world applicability. This article proposes an ego-centric trajectory prediction framework that leverages multisource onboard information for deployment-oriented prediction. Instead of using historical ground-truth trajectories, sequences of detected bounding boxes and their variation rates are adopted to represent surrounding vehicle states. To enhance motion modeling, local optical flow is introduced to capture target-vehicle motion tendencies, while raw images provide complementary scene-level context. In addition, neighboring vehicle states are incorporated to model inter-vehicle interactions, and onboard ego-motion signals are explicitly encoded to compensate for apparent motion disturbances caused by the moving observation coordinate system. A cross-attention-based fusion mechanism is designed to align and integrate optical-flow-based motion cues with structured bounding-box features, and a multimodal long short-term memory (LSTM) decoder generates multiple plausible future trajectories with corresponding probabilities. Experiments on the nuScenes dataset and real-vehicle tests on a modified Dongfeng AX7 demonstrate that the proposed method achieves accurate, robust, and real-time trajectory prediction, validating its effectiveness and practicality for real-world ego-centric driving scenarios.

As a crucial component linking the perception and decision-making modules of automated vehicles, vehicle trajectory prediction plays a key role in improving driving safety and efficiency. In complex driving environments, simply extrapolating future trajectories from historical states is insufficient for accurate prediction, as vehicle interactions can induce significant trajectory variations. In this paper, a graph-guided vehicle trajectory prediction method inspired by bidirectional dynamic interaction is proposed. Given the cyclic coupling of interactions, undirected graphs are constructed based on distance and velocity. Multilayer graph convolutional networks (GCNs) are employed to extract bidirectional interaction features among vehicles at each time step. Building upon this, a temporal correlation extractor based on gated recurrent units (GRUs) is established, extending single-step interactions to continuous-time interactions and capturing dynamic interaction representations. Furthermore, to leverage the improved trajectory prediction accuracy enabled by bidirectional dynamic interactions, relevant features are mapped into a latent space via a conditional variational autoencoder (CVAE), enabling the modeling of continuous and complex driving behaviors. The proposed method achieves an 11.2

As a key component of intelligent transportation systems, vehicular platoon cooperative control has increasingly become a focus of research due to its potential in enhancing road safety, alleviating traffic congestion, and reducing energy consumption. The ability to coordinate multiple vehicles in a platoon offers a promising solution to optimize traffic flow and improve overall driving efficiency. This paper proposes an online Receding Horizon Dual Heuristic Programming (RHDHP) control strategy for addressing the path tracking problem of multi-vehicle platoons in complex and dynamic traffic environments. The RHDHP method is employed to control the vehicular platoon, aiming to enhance both stability and accuracy in trajectory tracking. Finally, through joint simulations using Simulink and CarSim, the effectiveness of the proposed method in platoon trajectory tracking control is validated, demonstrating its effectiveness in dynamic traffic conditions.

Traffic flow control is essential for mitigating congestion and improving road efficiency. Traffic density waves characterize spatiotemporal variations in vehicle density and provide a useful tool for analyzing macroscopic traffic dynamics. In this article, boundary control of traffic density waves is investigated using a Markov-switching stochastic parabolic partial differential equation (MSSPDE) model. The model captures the macroscopic evolution of traffic density across multiple traffic conditions, including free flow, mild congestion, and severe congestion. Both partially unknown and fully known transition probabilities are considered. A boundary feedback controller is designed on the basis of the outlet boundary density and the spatial integral of traffic density. Stochastic stability analysis is performed for the closed-loop MSSPDE system. Sufficient conditions are derived for mean-square exponential stability (MSES) and robust MSES under parametric uncertainties. An H-infinity boundary control problem is also addressed to guarantee a prescribed disturbance attenuation level over a finite time horizon. Numerical simulations and quantitative comparisons demonstrate the effectiveness and practical relevance of the proposed control designs.

Accurate prediction of cyclist trajectories is essential for safe and reliable autonomous driving and intelligent transportation systems (ITSs) in complex traffic scenarios. To address the challenges posed by cyclists’ diverse intentions and non-linear motion patterns, we propose PhysiCycle, a novel multi-task learning framework that jointly predicts future trajectories and turning intentions. This framework integrates interpretable physical modeling with deep learning to enhance both prediction accuracy and behavioral consistency. Our model integrates a dual-path encoder to extract temporal motion cues and behavioral features, an intention classifier module, and a physically consistent decoder with bicycle kinematics consistency constraints. Experimental results on a real-world cyclist action dataset demonstrate that our method significantly outperforms baseline models in both intention classification and trajectory accuracy, achieving strong physical plausibility and generalization performance.

Trajectory prediction in emergencies scenarios is crucial for autonomous driving. Yet vehicle motion signals in these situations are highly nonlinear and nonstationary, with complex temporal and dynamic dependencies. Because mainstream datasets mainly cover regular driving and omit emergencies, existing models can still achieve satisfactory performance under normal conditions even without explicitly modeling exteroceptive cues, kinematic signals, or semantic intentions. However, during highly nonstationary and dynamic processes such as sudden cut-in or emergency braking, these weakly dependent architectures reveal significant shortcomings in generalization and robustness. To address these challenges, this paper proposes RDI-Pred, a multi-source temporal prediction framework that integrates Risk-Dynamics-Intention synergy from the perspectives of time-series signal processing and dynamic system modeling. First, we build a risk-aware exteroceptive encoder that uses prior-enhanced risk attention for risk scoring. Furthermore, a tri-agent interaction micrograph is constructed among the ego vehicle(Ego), target vehicle(TV), and closest in-path vehicle (CIPV) to model localized spatiotemporal dependencies, thereby enabling early-stage perception of exteroceptive risks. Next, we design a multi-scale Dynamics encoder that captures motion dynamics at short, mid, and long horizons. A 1D-CNN with a sliding window extracts short-term transients, BiGRU (Bidirectional Gated Recurrent Unit) states describe mid-term behavior, and a BiLSTM (Bidirectional Long Short-Term Memory) with self-attention models long-term dependencies, yielding a robust dynamic prior for trajectory decoding. Finally, we add cut-in intention recognition auxiliary task to constrain and re-score multi-modal trajectory candidates in decoding, promoting intention-aligned trajectories and suppressing mismatched ones. On the large-scale ESP high-risk dataset, RDI-Pred surpasses MTR with +32.9% mAP, -44.0% minADE, -45.9% minFDE, and-46.6% MR, showing clear performance gains across all key metrics. The results confirm its accuracy and robustness under emergency high-risk conditions, offering a practical path toward zero-tolerance safety in autonomous driving. Our code will be made publicly available at https://github.com/penglo/RDI-Pred-Risk-Dynamics-Intention-Collaborative-Vehicle-Trajectory-Prediction-in-Emergency-Scenarios .

To address the limitations of traditional test scenario construction, including single-dimensional screening metrics and the challenge of balancing efficiency and quality, this paper proposes a method for constructing typical test scenarios for autonomous driving safety evaluation. First, the Safety Impact Coefficient (SIC) is introduced as a screening metric to analyze the coupling relationships between autonomous driving safety characteristics and scenario elements, enabling the identification of key scenario factors. Next, a PICT variable-strength combination strategy is adopted to construct multidimensional test scenario sets, designed to meet diverse requirements for evaluating autonomous driving safety. Finally, a k-prototypes clustering method is utilized to generate representative and targeted typical test cases. Validation results on the SPMD dataset and through simulation tests demonstrate that the proposed method can efficiently and accurately generate critical test cases that cover various safety impact levels. Moreover, the constructed test cases exhibit strong rationality and effectiveness in autonomous driving system testing. This research addresses the lack of specificity in traditional test scenario construction methods, offering robust technical support for enhancing testing efficiency and reducing costs in autonomous driving safety evaluations.

In this study, event-triggered intermittent boundary stabilization of nonlinear traffic density waves governed by diffusion-dispersion PDEs is investigated. The boundary controller is activated intermittently, and its input is updated by a state-dependent triggering rule, thereby reducing communication updates and actuator usage. Sufficient conditions for asymptotic stabilization that relate the triggering threshold, forced triggering interval, duty ratio, and boundary feedback gains are established via Lyapunov functionals. An observer-based extension is presented for limited state measurements, and robustness under bounded parametric uncertainties is analyzed. Simulations show that event-triggered intermittent boundary control (ET-IBC) converges in 1.853 s with 25 updates. Compared with continuous boundary control, the convergence time increases by only 14.4%; compared with periodic intermittent control, the convergence time decreases by 23.3%, and the final state energy is reduced by more than one order of magnitude.

A game-based cooperative steering control (GCSC) approach is introduced to facilitate effective collaboration between human drivers and automation, incorporating the neuromuscular delay inherent in human responses. In this framework, a dynamic coordination between driver and automation goals is achieved through the establishment of a game equilibrium in instances of driver and automation conflict. In response to the challenges posed by frequent modifications in driving weights and their subsequent burden on human drivers, this article proposes a strategy that integrates fixed initial weights with dynamic adjustments to driver-automation driving weights. Moreover, a comprehensive evaluation method including subjective and objective evaluation indexes is proposed. Different drivers are invited to perform virtual driving experiments, and the experimental results are analyzed by the proposed evaluation method. It is concluded that the driver's driving weight should be kept at a high level during cooperative steering control when the driver's intention cannot be perfectly obtained, and the determination of the driving weight should also consider the driver's driving skills.

The performance of the path tracking control system is poor when vehicles navigate under extreme conditions of high speed and large curvature due to the uncertainty of vehicle model parameters, modeling errors and disturbances. To address this issue, this paper introduces a Linear Model Predictive Controller (GP-SOMPC) based on Gaussian Process (GP) and Snake Optimizer (SO) to achieve high-performance trajectory tracking control under large curvatures through learning. This algorithm adopts a two-degree-of-freedom vehicle dynamics model and models disturbances and unmodeled dynamics as Gaussian Processes (GP) using collected experimental data. Additionally, an Adaptive Weighting Coefficient Model Predictive Controller based on the Snake Optimizer (SO) method is presented to improve the poor adaptation of vehicle trajectory tracking accuracy according to reference path curvature under fixed weighting coefficients in traditional Model Predictive Control (MPC), which results in inadequate trajectory tracking precision. The simulation experiments have verified that the proposed GP-SOMPC algorithm can enhance trajectory tracking precision under high-speed, large curvature conditions and exhibits good adaptability to variations in path curvature.

Accurate trajectory prediction is a fundamental prerequisite for autonomous vehicles to understand surrounding traffic participants and to make safe planning decisions. Especially, in high-risk highway cut-in maneuvers, vehicle behavior is strongly constrained by limited available gaps and bidirectional safety requirements from both leading and following vehicles. These constraints are further complicated by heterogeneous driving styles and fine-grained dynamical precursor cues, making the interactions difficult to characterize using generic prediction models. Thus, this paper reformulates the highway cut-in prediction problem from a three-vehicle Gap-constrained interaction perspective involving the target-lane leader, the cut-in vehicle, and the target-lane follower, explicitly and structurally modeling the functional distinctions among leader-induced constraints, the decision-making agent, and follower-induced constraints, and proposes a trajectory prediction framework for highway cut-ins termed GC-StyleTP. Specifically, this paper proposes: (1) a Dynamic-Precursor Aware Encoder, DPAE that performs temporal modeling of the target vehicle’s dynamical time series to highlight critical precursor patterns prior to cut-in formation. (2) an XGBoost-based Online Driving Style Recognition module (X-ODSR) that outputs a style-probability prior to characterize inter-agent differences in gap acceptance and risk preference. (3) a Risk–Gap Causal Fusion Transformer (RGCF-Former) that adopts a three-stream architecture to model the evolution of risk under bidirectional safety constraints from the leading and following vehicles, together with the dynamics of available gaps, and produces physically meaningful risk-aware embeddings via temporally causal self-attention and cross-stream fusion. Experimental results on the large-scale highway cut-in dataset ESP show that, compared with the baseline model MTR, GC-StyleTP improves mAP6 by 32.0%, reduces minADE6 and minFDE6 by 42.86% and 41.67%, respectively, and decreases MR6 by 39.13%. Moreover, the proposed method outperforms representative state-of-the-art baselines across multiple metrics, further validating its effectiveness and reliability in high-risk cut-in scenarios.

Weak-stiffness grinding systems are highly susceptible to force disturbances and prone to dynamic instability, which severely constrains machining accuracy, efficiency, and reliability. To overcome the limitations of conventional approaches that mainly depend on energy-field assistance or external damping compensation and fail to regulate the system intrinsically from the force source, a dynamic stability enhancement method based on microstructure-induced spatial force reconstruction is innovatively proposed. A dynamic stability comprehensive model of grinding system was established, which considered the multi-random characteristics of microstructured grinding wheel, macro-micro multiple regeneration, spatial grinding force, instantaneous grinding vibration, grinding stability boundary and dynamic reliability, thereby elucidating the intrinsic mechanism by which microstructures stabilize weak-stiffness grinding systems through spatial grinding force reconstruction. Comprehensive experiments, including static deflection and modal tests, sub-nanosecond laser fabrication, micro-tooth internal thread grinding, and profile measurements, were conducted to validate the theoretical models. The results demonstrate that the model accurately predicts the dynamic stability of the grinding process, achieving peak simulation accuracies of 94.8 % for spatial grinding force and 97.3 % for vibration. The microstructure angle modifies key cutting conditions such as edge angle, number, radius, and contact length, thereby redistributing grinding forces toward the high-rigidity axial direction. At a microstructure angle of 45 degrees, the axial component proportion reaches the maximum value of 38.1 %, expanding the critical stability width by 47.3 % along the high rotation speed. Furthermore, the microstructured grinding wheel significantly improves the machining quality of large aspect-ratio internal threads, reducing the bottom fillet radius by 90.4 %, while the system reliability reaches 98.87 %.

This paper proposes an MPC stability control strategy based on phase-plane stability domain boundaries to address the issue of poor accuracy of control sequences computed by model predictive control (MPC)-based stability controllers due to constraint accuracy limitations. The stability domain boundary function is added to the state quantity constraints of the MPC stability controller by dividing the phase-plane linear stability domain of the center-of-mass lateral deflection and the swing angle velocity of the moving vehicle. This generates MPC constraints in real-time along with the current driving state of the vehicle, resulting in more accurate stability control of the vehicle. The simulation results show that compared with the traditional MPC stability control strategy, the MPC controller that introduces the boundary function of the phase plane stability domain plays with the limits of vehicle dynamics and has a better stability control effect.

Vehicle collision risk assessment is important in enabling intelligent vehicles to make safe and reliable decisions. However, performing this task from a first-person perspective poses significant challenges. This paper proposes a first-person view collision risk assessment framework based on surrounding vehicle intent prediction and spatial constraints. We use a graph attention network to capture the interaction between the historical trajectories of the target vehicle and its neighboring vehicles, thereby modeling the motion intentions of surrounding vehicles. Then a temporal encoder based on LSTM is used to model the temporal dynamics of the target vehicle's motion, and a decoder guided by a weighted multi-objective loss function recursively generates future trajectories. Finally, based on the predicted trajectories, a collision risk assessment model is constructed by incorporating spatial constraints and comprehensively considering influencing factors. Experimental results proved that our unsupervised risk assessment model demonstrates its temporal responsiveness and potential for early warning.

In autonomous driving scenarios, the high dynamic relative motion between vehicles and pedestrians introduces significant challenges for trajectory prediction. Rapidly changing speeds and directions often lead to unevenly sampled trajectory points, which make it difficult to capture fine-grained motion details and increasing the complexity of prediction. To address these challenges, we propose a novel context-aware hierarchical trajectory prediction framework that leverages heterogeneous fusion, bidirectional residual cross-attention (BRCA), and an adaptive stepwise decoding mechanism. The hierarchical fusion effectively extracts temporal dependencies from sequence information and spatial-dynamic features from visual data. Meanwhile, the BRCA mechanism enhances intermodal interactions by capturing hidden relationships and complementary patterns, while the adaptive decoder refines predictions dynamically, incorporating global motion trends and local dynamic details to improve accuracy. Extensive experiments on the JAAD and PIE datasets demonstrate the effectiveness and superiority of our approach.

This study proposes a personalized motion planning framework for autonomous driving, driven by driver voice commands. The method begins with the recognition and parsing of voice commands to identify personalized driving preferences. An artificial potential field(APF) model is then constructed and dynamically adjusted according to the parsed requirements. A vehicle motion planning model is subsequently developed, integrating the personalized artificial potential field with Model Predictive Control (MPC) to achieve individualized driving behavior. The effectiveness of the method is validated through MATLAB/SCANeR co-simulations, confirming its capability to adapt motion planning to driver-specific requirements.

To address the stochasticity and uncertainty in driver behavior, an event-triggered shared control strategy is proposed, enabling real-time adaptation while ensuring system stability. The vehicle control system is modeled as a nonlinear time-varying (NTV) system, incorporating human-machine interaction, driver behavior, variable speeds, and uncertain tire dynamics. Sparse Gaussian process regression (GPR) is employed for online system identification, eliminating the need for an exact system model. An adaptive feedback linearization (AFL) controller is designed, with asymptotic stability proven using a common Lyapunov function (CLF). The event-triggered mechanism updates the model and controller only when GPR uncertainty exceeds an adaptive threshold. Simulations and driver-in-the-loop experiments assess the proposed algorithm, demonstrating its robustness and adaptability to different driver behaviors, while reducing communication overhead and ensuring precise steering control.

In intelligent transportation systems (ITS), pedestrian trajectory prediction is one key task for ensuring the safety and efficient operation of autonomous vehicles (AVs). In pedestrian‒vehicle mixed environments, especially in shared spaces without traffic signals, pedestrians may abruptly alter their path by deciding to cross the road or stop waiting. These abrupt changes in trajectory are not fully reflected in their historical movement patterns, making the prediction of future trajectories solely based on past data challenging. To address this issue, a hierarchical optimization-based (HOB) pedestrian trajectory prediction model is proposed, which is designed to generate more accurate trajectory prediction from the ego-centric perspective by leveraging multisource information. The model consists of two main stages: strategic trajectory prediction and tactical trajectory prediction. In the strategic trajectory prediction phase, historical trajectories and ego vehicle speed are combined to extract features through a multiscale convolutional temporal attention mechanism and generate multimodal strategic trajectories via a conditional variational autoencoder (CVAE). In the tactical trajectory prediction phase, a dynamic window-based aggregation mechanism is employed, which organically integrates strategic features and further fuses scene-level and local-level optical flow information, progressively decoding to generate improved tactical trajectories. Extensive experimental results on the JAAD and PIE datasets reveal that the proposed method outperforms advanced approaches in terms of overall prediction performance.

The embedding of high-level traffic semantics has elevated the precision of vehicle trajectory prediction tasks to a new level. However, owing to the absence of feature-level integration, the information from high-definition maps is underutilized. To this end, a map search-based vehicle trajectory prediction method conditioned on lane segments is proposed in this article. The map is discretized into a graph, where nodes represent lane centerline segments. On this basis, the agent-to-agent, agent-to-map, and map-to-map modules are designed to depict heterogeneous interaction patterns involving vehicles and pedestrians. In addition, a goal node querying mechanism is introduced, which integrates vehicle motion, interaction, and traffic flow states and serves as prior information for trajectory prediction. Finally, a feasible path selection strategy is proposed, generating traffic rule-related prediction trajectories point by point, fully utilizing map information. The experimental results on the nuScenes dataset indicate that the proposed method achieves state-of-the-art prediction accuracy compared with advanced methods.