Jiacheng Yin

LiDAR configuration selection in roadside sensing remains constrained by limited controlled evidence across sensors. To address this gap, we present MR-LiDAR, a controlled cross-configuration benchmark comparing four LiDAR configurations from 16 to 128 beams on dynamic vehicle and vulnerable road user (VRU) targets. Its cross-configuration main set covers matched daytime, nighttime, and vehicle-occlusion conditions, while three deployment and operating condition subsets examine mounting height, target speed, and curved-road geometry. By reducing confounding from scene and trajectory variation, MR-LiDAR links sensor configuration to target point acquisition and algorithmic detection outcomes. We characterize distance-dependent point loss, define effective beam count as the number of vertical beams intersecting the target’s angular extent, and introduce a point-based Occlusion Ratio. The results show that 16- and 32-beam configurations degrade earlier with distance, while VRUs exhibit greater point sparsity than motor vehicles. Nominal beam count alone is insufficient to explain roadside scanning capability. In the cross-configuration main set, the 80- and 128-beam configurations achieve comparable results, consistent with their nearly identical downward beam layouts. A uniform 80-beam counterfactual yields only about one third of the effective beam count of the actual non-uniform configuration. Illumination and the tested speed range have little influence on target-point-count patterns, whereas mounting height and near-field road geometry can substantially alter configuration ordering. Finally, an empirical minimal-sufficient configuration matrix combines frame-level geometric sufficiency with matched-box quality and includes threshold sensitivity analysis, providing bounded configuration guidance for the tested conditions. MR-LiDAR Benchmark is publicly available at https://github.com/ShunlaiC/MR-LiDAR-Benchmark .

Taxi traffic safety arises from the complex interplay among road infrastructure, driver behavior, vehicle characteristics, and environmental conditions. However, driver, vehicle, and environmental factors are typically challenging to regulate directly through policy or design, whereas road characteristics can be systematically adjusted through infrastructure planning and traffic management strategies. Therefore, understanding how specific road features contribute to crash risk offers a more practical and sustainable pathway for enhancing taxi travel safety. This study aims at investigating how road characteristics heterogeneously influence the severity of taxi-involved crashes. To accomplish this, a latent class clustering method is first leveraged to classify crashes with similar contextual features. Subsequently, a partial proportional odds model is applied to quantify the class-specific impacts of road characteristics on taxi-involved crash severity. The taxi crash data reported by the London police from January 2019 to June 2024 are utilized to achieve this study. The results demonstrate substantial heterogeneity in how road characteristics affect taxi-involved crash outcomes across different crash classes. Among the examined factors on taxi-involved crashes, speed limits and junction controls exert the most pronounced influences. Higher speed limits ( ≥ 40 mph) substantially elevate the probabilities of serious injuries (+14.76%) and fatalities (+0.93%). In contrast, moderate speed limits (31−40 mph) are associated with notably lower marginal effects on serious injuries (+8.27%) and fatalities (+0.46%). The installation of traffic signals at junctions decreases the likelihood of serious injuries (−11.68%) and fatalities (−0.50%). These findings indicate that uniform safety policies may be insufficient for mitigating class-specific taxi risks. This study contributes empirical evidence and practical insights for road safety management. By identifying the heterogeneous influence patterns of road characteristics, the findings support targeted infrastructure design, speed management, and signal control strategies to effectively reduce both the frequency and severity of taxi-related crashes.

Proactive lane-change risk assessment requires estimating whether an intended maneuver will lead to unsafe interactions before the maneuver is completed. This is challenging in naturalistic driving data because actual crashes are extremely rare, and binary collision labels provide little discriminative information for learning risk. We propose IntentDiff, an intent-conditioned diffusion framework for proactive lane-change risk assessment. The framework uses predicted future trajectories as the basis for risk estimation. A vectorized scene context learning module combines a VectorNet backbone with a Vector Quantized Variational Autoencoder (VQ-VAE) to map agent–map interactions into discrete intent codes. These codes organize complex traffic situations into interpretable intent prototypes and provide semantic guidance for trajectory generation. Conditioned on the learned intent code, a diffusion model generates kinematically consistent multimodal trajectories of the target vehicle. On the forecast trajectories, Monte Carlo rear-end risk is evaluated against the four bounding vehicles and fused into a Lane-Change Risk Index (LCRI). On the highD dataset, the framework attains an average displacement error of 0.42 m over a 5-s horizon. The forecast-based LCRI agrees closely with the index computed from realized future trajectories, indicating that most high-risk lane changes can be identified before the maneuver is completed. Grouping LCRI by intent code further reveals systematic variation in risk across lane-change maneuvers, suggesting that the learned codebook captures risk-relevant interaction patterns in addition to maneuver semantics.

With the continuous expansion of the expressway network and the rapid growth of traffic demand, traditional single-source traffic detection data is limited in spatial–temporal coverage and accuracy, which can hardly support the refined operation and management of intelligent expressways. Existing data preprocessing methods often fail to fully capture global spatiotemporal features, and traditional PSO-BP neural networks are prone to local optima. To address these issues, this study investigates multi-source traffic data fusion using ETC-DSRC and RTMS microwave data from the Jiangsu section of the G50 Shanghai-Chongqing Expressway. The HaLRTC tensor completion algorithm is adopted to repair missing and abnormal data, fully mining the spatial–temporal correlation characteristics of traffic flow. The beetle antennae search (BAS) mechanism is introduced into the particle swarm optimization (PSO) process to improve particle search behavior and population diversity. On this basis, a BAS-optimized PSO-BP neural network, referred to as BSO-BP in this study, is constructed for multi-source traffic data fusion. In this model, the improved PSO algorithm is used to optimize the initial weights and thresholds of the backpropagation (BP) neural network, thereby improving the global search capability and convergence stability of the fusion model. Taking the average road speed as the fusion target, MAE, RMSE and MAPE are used for accuracy verification. The results show that the proposed model has significantly higher accuracy than single-source data methods and BP, PSO-BP, and GA-PSO-BP models, and can reflect the real traffic state of road sections more accurately.

Accurate bus Origin-Destination (OD) identification based on public transit data is essential for route planning and service optimization. However, current research lacks evaluation metrics specifically designed for the bus OD identification task, often relying on conventional indicators that fail to fully capture the plausibility of identification results. This study proposes an integrated framework encompassing label generation, modeling, and validation. Specifically, we developed a two-layer model that employs a Random Forest for feature selection and a CatBoost classifier for alighting stop identification. To enable validation without ground-truth labels, a novel Trend Consistency Index (TCI) was introduced. A case study using public transit data from Chengdu demonstrates that the proposed method outperforms mainstream algorithms such as XGBoost, achieving the highest scores across all key metrics. Multi-dimensional validation further confirmed its effectiveness, showing high fidelity in reproducing temporal travel patterns, accurately identifying major destination clusters through spatial heatmaps, and yielding an excellent fit for trip distribution curves (R² up to 0.8382). This research provides a robust and practical solution for the bus OD identification problem, offering a reliable quantitative foundation to advance smart public transit management.

This paper investigates a movable antenna (MA)-enabled integrated sensing and communication (ISAC) system under the influence of antenna crosstalk. First, it generalizes the antenna crosstalk model from the conventional fixed-position antenna (FPA) system to MA. Then, a Cramer-Rao bound (CRB) minimization problem driven by joint beamforming and antenna position design is presented. To address the problem, a deep reinforcement learning (DRL) approach, specifically the twin delayed deep deterministic policy gradient (TD3) algorithm is employed to train a flexible beamforming agent. Numerical results show that the proposed crosstalk-resilient (CR) beamforming outperforms other benchmark schemes under antenna crosstalk.

This paper investigates a movable antenna (MA) enabled integrated sensing and communication (ISAC) system under the influence of antenna crosstalk. First, it generalizes the antenna crosstalk model from the conventional fixed-position antenna (FPA) system to the MA scenario. Then, a Cramer-Rao bound (CRB) minimization problem driven by joint beamforming and antenna position design is presented. Specifically, to address this highly non-convex flexible beamforming problem, we deploy a deep reinforcement learning (DRL) approach to train a flexible beamforming agent. To ensure stability during training, a Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm is adopted to balance exploration with reward maximization for efficient and reliable learning. Numerical results demonstrate that the proposed crosstalk-resilient (CR) algorithm enhances the overall ISAC performance compared to other benchmark schemes.