Reward models play a pivotal role in reinforcement learning (RL) and multi-modal trajectory selection for autonomous driving. However, acquiring such rewards typically relies on hand-crafted rule-based objectives or perception ground truth, which hinders generalization for data-scaling. While Vision-Language Models (VLMs) have demonstrated feasibility as reward models in other domains, their effectiveness in driving tasks remains underexplored. In this work, we bridge this gap by (1) introducing DriveReward, a reasoning trajectory evaluation dataset rigorously labeled via temporally-grounded visual guidance, and augmented with counterfactual driving behaviors., (2) alongside a specialized Vision-Language Reward Model. To address the scarcity of failure cases in conventional datasets, we propose a counterfactual data annotation scheme to construct cases encompassing diverse driving styles and erroneous behaviors. Evaluations on our proposed benchmark reveal that even leading open-source and proprietary VLMs fail to excel across all tasks, highlighting significant room for improvement in existing models. Building on these findings, we subsequently tailor a specialized 1B reward model that outperforms larger VLMs on task-specific reward alignment. Finally, we validate our reward model's effectiveness by integrating it into RL finetuning and multi-modal trajectory scoring across multiple baselines, achieving performance comparable to rule-based reward calculations in both open-loop and closed-loop evaluation.

Coordinated control of multiple intersections remains a major challenge in urban traffic systems due to complex local-global dependencies. To address this issue, the MPNNLight framework is proposed, a message-passing graph neural network for cooperative multi-intersection traffic signal control. The core innovation of this work is the proposed Attention-Wavelet Spatial Transformer (AWSformer), a message computation module that combines self-attention mechanisms and 2D discrete wavelet transform (2D-DWT) to adaptively compute inter-agent messages across spatial and frequency domains. This design enables efficient extraction of multi-scale dependencies and enhances coordination among intersections. Experiments on diverse datasets demonstrate that MPNNLight achieves superior performance and efficiency, compared with existing GNN-based traffic signal control methods.

Recent advances in decision-making policies have led to significant progress in fields such as autonomous driving and robotics. However, testing these policies remains crucial with the existence of critical scenarios that may threaten their reliability. Despite ongoing research, challenges such as low testing efficiency and limited diversity persist due to the complexity of the decision-making policies and their environments. To address these challenges, this paper proposes an adaptable Large Language Model (LLM)-driven online testing framework to explore critical and diverse testing scenarios for decision-making policies. Specifically, we design a "generate-test-feedback" pipeline with templated prompt engineering to harness the world knowledge and reasoning abilities of LLMs. Additionally, a multi-scale scenario generation strategy is proposed to address the limitations of LLMs in making fine-grained adjustments, further enhancing testing efficiency. Finally, the proposed LLM-driven method is evaluated on five widely recognized benchmarks, and the experimental results demonstrate that our method significantly outperforms baseline methods in uncovering both critical and diverse scenarios. These findings suggest that LLM-driven methods hold significant promise for advancing the testing of decision-making policies.

Synthetic data mitigates the data scarcity problem in autonomous driving perception. However, the synthetic-to-real gap leads to performance degradation, hindering real-world model generalization. Although current methods leverage diffusion models for photorealistic style transfer to bridge this gap, they critically ignore a practical asymmetry: while synthetic data possesses perfect pixel-level annotations, real-world style reference images generally lack corresponding labels. Consequently, existing methods relying on symmetric semantic guidance suffer from either prohibitive annotation costs or severe semantic misalignment. To address this dilemma, we formally propose a novel task: Asymmetric Style Transfer for Autonomous Driving (ASTAD), which requires semantically consistent transfer using only labeled synthetic content and unlabeled real-world references. We further introduce the ASTModel, a training-free two-stage framework designed to bridge this domain gap under asymmetric constraints. ASTModel first extracts a coarse semantic prior from the unlabeled target, followed by dynamic prior refinement and class-consistent style injection during the denoising process. Extensive experiments demonstrate that ASTModel significantly outperforms existing methods in downstream perception utility and structural fidelity, while offering a 3.2× inference speedup. This work aligns synthetic-to-real adaptation with practical constraints, holding the potential to accelerate the scalable deployment of robust autonomous driving systems. Code: https://github.com/Dingyi-Yao/ASTAD.

The seismic safety of shield tunnels constructed in locally liquefiable sites is highly vulnerable under mainshock-aftershock (MS-AS) sequences due to the coupled effects of strong ground motion and liquefaction-induced soil degradation, while a systematic resilience-based assessment under such sequential loading remains limited. To investigate tunnel performance under such conditions, a validated tunnel-soil interaction numerical model is developed in OpenSees and applied to 200 scaled MS-AS cases to analyze the dynamic response and seismic resilience of shield tunnels subjected to MS-AS sequences. An improved seismic resilience index is proposed by integrating three representative functionality recovery functions, thereby reducing sensitivity to recovery model selection. Based on this index, a data-driven three-level resilience assessment framework is established by coupling damage state probability curves with the evolution of the resilience index, enabling intensity-dependent resilience classification. Results show that aftershocks significantly intensify non-uniform deformation in locally liquefied zones, leading to increased deformation and internal force demands. The horizontal displacement at the tunnel crown and segment joint opening increase by about 40 %, the vertical displacement increases by about 20 %, and the bending moment increases by about 15 %. The exceedance probabilities of slight, moderate, and extensive damage states increase by up to 23 %, 52 %, and 55 %, respectively. From a resilience perspective, the resilience index defined using a positive exponential recovery function decreases by up to 9.8 % under MS-AS sequences. The proposed framework provides a practical tool for seismic resilience assessment and risk control of shield tunnels in liquefiable ground.

Recently, pre-trained encoders have gained widespread use due to their strong capability in representation extraction. However, they are vulnerable to downstream-agnostic attacks (DAAs). Existing DAA methods operate under a permissive threat model, where an attack is successful if the generated downstream-agnostic adversarial examples (DAEs) change the original prediction, without requiring a specific target. In this paper, we propose a Targeted DAA (TDAA) method under a stricter threat model requiring the attack to be both targeted and downstream-agnostic. Since the downstream task is unknown and encoders do not directly produce predictions, achieving a targeted attack is particularly challenging. To address this, we introduce a novel component termed the 'threat image', pre-selected by the attacker as the target. Specifically, a generator is designed to produce example-specific adversarial perturbations that compel the victim encoder to output identical features for both the DAEs and the threat image. Unlike previous DAA methods that generate a single shared perturbation for all samples, which often fails due to image diversity, our method adopts an example-specific paradigm. This generates tailored perturbations for each image to ensure a high attack success rate and invisibility. By leveraging the threat image as a feature-level anchor, our method builds a task-agnostic bridge to reveal the vulnerabilities of the victim encoder. Extensive experiments on 10 self-supervised methods across 3 benchmark datasets demonstrate the effectiveness of our approach and reveal the pronounced vulnerability of pre-trained encoders. The code will be made publicly available after the review period.

Ensuring the reliability of autonomous driving perception systems requires extensive environment-based testing, yet real-world execution is often impractical. Synthetic datasets have therefore emerged as a promising alternative, offering advantages such as cost-effectiveness, bias free labeling, and controllable scenarios. However, the domain gap between synthetic and real-world datasets remains a major obstacle to model generalization. To address this challenge from a data-centric perspective, this paper introduces a profile extraction and discovery framework for characterizing the style profiles underlying both synthetic and real image datasets. We propose Style Embedding Distribution Discrepancy (SEDD) as a novel evaluation metric. Our framework combines Gram matrix-based style extraction with metric learning optimized for intra-class compactness and inter-class separation to extract style embeddings. Furthermore, we establish a benchmark using publicly available datasets. Experiments are conducted on a variety of datasets and sim-to-real methods, and the results show that our method is capable of quantifying the synthetic-to-real gap. This work provides a standardized profiling-based quality control paradigm that enables systematic diagnosis and targeted enhancement of synthetic datasets, advancing future development of data-driven autonomous driving systems.

The safe deployment of autonomous driving (AD) systems is fundamentally hindered by the long-tail problem, where rare yet critical driving scenarios are severely underrepresented in real-world data. Existing solutions including safety-critical scenario generation and closed-loop learning often rely on rule-based heuristics, resampling methods and generative models learned from offline datasets, limiting their ability to produce diverse and novel challenges. While recent works leverage Vision Language Models (VLMs) to produce scene descriptions that guide a separate, downstream model in generating hazardous trajectories for agents, such two-stage framework constrains the generative potential of VLMs, as the diversity of the final trajectories is ultimately limited by the generalization ceiling of the downstream algorithm. To overcome these limitations, we introduce VILTA (VLM-In-the-Loop Trajectory Adversary), a novel framework that integrates a VLM into the closed-loop training of AD agents. Unlike prior works, VILTA actively participates in the training loop by comprehending the dynamic driving environment and strategically generating challenging scenarios through direct, fine-grained editing of surrounding agents' future trajectories. This direct-editing approach fully leverages the VLM's powerful generalization capabilities to create a diverse curriculum of plausible yet challenging scenarios that extend beyond the scope of traditional methods. We demonstrate that our approach substantially enhances the safety and robustness of the resulting AD policy, particularly in its ability to navigate critical long-tail events.

Reliable safety evaluation is essential for the development and large-scale deployment of autonomous vehicles (AVs). However, the complexity of high-dimensional traffic environments and the rarity of safety-critical events make direct testing extremely inefficient and costly, requiring an enormous number of simulation miles. Importance sampling has been employed to accelerate this process, but most existing methods rely on known or explicit traffic behavior distributions, which are often inaccessible in practice. To address this limitation, we propose Deep-IIS, which extends implicit importance sampling by utilizing rejection sampling to handle unknown distributions without requiring explicit density functions. By integrating dense deep reinforcement learning to adaptively optimize key sampling parameters, Deep-IIS intelligently concentrates sampling on high-risk regions, significantly improving testing efficiency while maintaining evaluation reliability. We validate the proposed method in two representative traffic scenarios, demonstrating its superior ability to uncover safety-critical cases and to produce accurate performance estimates under implicit distributions.

Abstract This study evaluates the real-world emission impacts associated with the transition from China V to China VI standards for heavy-duty diesel trucks (HDDTs). On-road measurements from 24 HDDTs were obtained using Portable Emission Measurement Systems to quantify particle number (PN) and nitrogen oxides (NOx) emission factors (EFs) under diverse operating conditions. Comparisons were conducted among China V, China VI compliant, and China VI non-compliant vehicles, alongside an assessment of selective catalytic reduction (SCR) and exhaust gas recirculation (EGR) system performance. Fleet-level results indicate that the average PN and NOx EFs of China VI HDDTs were 5.48 and 0.75 times those of China V vehicles, primarily due to a substantial share of non-compliant trucks exceeding regulatory limits. Non-compliant HDDTs exhibited increases of 1,695.6% in PN and 85.6% in NOx EFs relative to China V, whereas compliant vehicles achieved reductions of 51.5% and 59.6%, respectively. Operating mode-specific analysis revealed inadequate PN control at low-to-medium speeds and NOx exceedances under high vehicle-specific power. SCR systems reduced NOx by 76.6-96.8%, with performance varying by compliance and operating mode, while EGR consistently lowered NOx emissions. These findings highlight the necessity for strengthened in-use compliance monitoring and enforcement under China VI regulations.

This study proposes a hierarchical Mixed-Integer Linear Programming optimization framework to determine the optimal allocation of unidirectional and bidirectional electric vehicle chargers in a Vehicle-to-Building enabled office building, in order to minimize daily operational costs. Simulations are conducted across two Vehicle-to-Building settings, each defined by a different compensation scheme, and evaluated under six scenarios representing varying levels of electric vehicle user participation. The results demonstrate that strategic Vehilce-to-Building deployment can achieve cost reductions of 0.29-4.86% and grid energy demand reduction up to 7.98% compared to conventional unidirectional Vehicle-One-Grid charging systems in the case study building. However, low participation rates render investments economically not viable. Both the extent of cost savings and the optimal investment in charger infrastructure are highly sensitive to the level of electric vehicle user participation.

Spatio-temporal graphs are powerful tools for modeling complex dependencies in traffic time series. However, the distributed nature of real-world traffic data across multiple stakeholders poses significant challenges in modeling and reconstructing inter-client spatial dependencies while adhering to data locality constraints. Existing methods primarily address static dependencies, overlooking their dynamic nature and resulting in suboptimal performance. In response, we propose Federated Spatio-Temporal Graph with Dynamic Inter-Client Dependencies (FedSTGD), a framework designed to model and reconstruct dynamic inter-client spatial dependencies in federated learning. FedSTGD incorporates a federated nonlinear computation decomposition module to approximate complex graph operations. This is complemented by a graph node embedding augmentation module, which alleviates performance degradation arising from the decomposition. These modules are coordinated through a client-server collective learning protocol, which decomposes dynamic inter-client spatial dependency learning tasks into lightweight, parallelizable subtasks. Extensive experiments on four real-world datasets demonstrate that FedSTGD achieves superior performance over state-of-the-art baselines in terms of RMSE, MAE, and MAPE, approaching that of centralized baselines. Ablation studies confirm the contribution of each module in addressing dynamic inter-client spatial dependencies, while sensitivity analysis highlights the robustness of FedSTGD to variations in hyperparameters.

For multi-intersection control, existing research mainly adopts multiagent frameworks to tackle scalability issues. However, the traffic signal control (TSC) problem necessitates a single-agent framework, as a single control center monitors traffic conditions across all roads in the study area and coordinates the control of all intersections. This work proposes a novel single-agent RL-based urban corridor ATSC model: It abandons complex multiagent coordination and uses a single agent to centrally orchestrate signal timings across multiple intersections. Notably, the model is highly applicable to real-world settings. It defines state and reward functions based on a queue length metric—one that correlates with congestion and can be reliably estimated using probe vehicle data. Since probe vehicle data have become highly prevalent, this feature enables rapid, large-scale deployment. The single-agent framework primarily relies on a unique design of state, action, and reward. To facilitate learning and manage congestion, both state and reward functions are defined based on queue length, with actions designed to modulate queue dynamics. The queue length definition used in this study deviates slightly from conventional definitions but is closely correlated with congestion states. The method was comprehensively evaluated using the SUMO simulation platform under various traffic patterns. Experimental results show that the PPO algorithm demonstrates significantly faster learning than the DQN algorithm. The model effectively alleviates urban corridor congestion through coordinated multi-intersection control: During the simulation period of the entire scenario, the queue length did not exceed 50 vehicles, and instances where it exceeded 30 vehicles were relatively rare. Compared with the baseline scenario where queue lengths exceeded 150 vehicles, the proposed method significantly reduces road congestion. The work in this paper demonstrates the feasibility of controlling multiple intersections under a single-agent framework, and the control scope will be further expanded in the future.

This paper studies the safe and resilient control of Connected and Automated Vehicles (CAVs) operating in mixed traffic environments where they must interact with Human-Driven Vehicles (HDVs) under uncertain dynamics and exponentially unbounded false data injection (EU-FDI) attacks. These attacks pose serious threats to safety-critical applications. While resilient control strategies can mitigate adversarial effects, they often overlook collision avoidance requirements. Conversely, safety-focused approaches tend to assume nominal operating conditions and lack resilience to adversarial inputs. To address these challenges, we propose a control framework that integrates event-driven Control Barrier Functions (CBFs) and Control Lyapunov Functions (CLFs) with adaptive attack-resilient control. The framework further incorporates data-driven estimation of HDV behaviors to ensure safety and resilience against EU-FDI attacks. Specifically, we focus on the lane-changing maneuver of CAVs in the presence of unpredictable HDVs and EU-FDI attacks on acceleration inputs. The event-driven approach reduces computational load while maintaining real-time safety guarantees. Simulation results, including comparisons with pure event-driven methods lacking resilience, validate the effectiveness and robustness of the proposed EDSR framework in achieving collision-free maneuvers, stable velocity regulation, and resilient operation under adversarial conditions.

The advancement of autonomous vehicles hinges significantly on addressing safety concerns and obtaining reliable evaluation results. Testing the safety of autonomous vehicles is challenging due to the complexity of the high-dimensional traffic environment and the rarity of safety- critical events, often requiring billions of miles to achieve comprehensive validation, which is inefficient and costly. Current approaches, such as accelerated testing using importance sampling, aim to provide unbiased estimates of the performance of autonomous vehicles by generating a new distribution of background vehicles' behaviors based on an initial nominal distribution. However, these methods require knowledge of the original distribution of traffic behaviors, which is often difficult to obtain in practice. In response to these challenges, we introduce a novel methodology termed implicit importance sampling (IIS). Unlike traditional methods, IIS is designed to generate intelligent driving environments based on implicit distributions of traffic behaviors where the true distributions are unknown or not explicitly defined. IIS method leverages accept-reject sampling to construct an unnormalized proposal distribution, which increases the likelihood of sampling adversarial cases. Through applying importance sampling technique with unnormalized proposal distribution, IIS enhances testing efficiency and obtains reliable and representative evaluation results as well. The bias caused by unnormalization is also proved to be controlled and bounded.

This paper presents a closed-loop framework for conflict-free routing and scheduling of multi-aircraft in Terminal Manoeuvring Areas (TMA), aimed at reducing congestion and enhancing landing efficiency. Leveraging data-driven arrival inputs (either historical or predicted), we formulate a mixed-integer optimization model for real-time control, incorporating an extended TMA network spanning a 50-nautical-mile radius around Changi Airport. The model enforces safety separation, speed adjustments, and holding time constraints while maximizing runway throughput. A rolling-horizon Model Predictive Control (MPC) strategy enables closed-loop integration with a traffic simulator, dynamically updating commands based on real-time system states and predictions. Computational efficiency is validated across diverse traffic scenarios, demonstrating a 7-fold reduction in computation time during peak congestion compared to onetime optimization, using Singapore ADS-B dataset. Monte Carlo simulations under travel time disturbances further confirm the framework's robustness. Results highlight the approach's operational resilience and computational scalability, offering actionable decision support for Air Traffic Controller Officers (ATCOs) through real-time optimization and adaptive replanning.

Traditional traffic control methods at signalised intersections primarily focused on optimising traffic signals without adequately addressing the coordination between signals and vehicles. Fortunately, with the advent of advanced vehicular communication technologies, real-time bidirectional communication between roadside infrastructure and vehicles has become feasible, significantly improving coordination between these elements. This paper presents a bi-level optimisation method for signalised intersections, enhancing arrive-on-green (AOG) performance in connected vehicles. By extending optimal control from a single dimension - whether spatial or temporal - to a two-dimensional spatial-temporal approach, we develop a comprehensive bi-level control framework. The framework includes outer-layer signal optimisation for maximising green utilisation and inner-layer platoon trajectory optimisation. Intermediate parameters and extended planning-time are proposed to improve solution finding. The effectiveness of the proposed joint optimisation method was evaluated through simulation case studies conducted in SUMO. The results showed increased efficiency and reduced stops, with stable, accurate control.

This study addresses the multi-bus dispatching and boarding management problem for a ring-shaped structured bus transport system, where the dispatching of multiple buses is allowed and incorporates volume dynamics on bus stops and buses. A mathematical model of this problem aims to minimize the waiting time of passengers and space in buses during bus service operations. In the literature, the solution to this problem is obtained using a commercial solver named GUROBI, which suffers from the issue of computational inefficiency while solving the large scale of this problem. To tackle this computational inefficiency challenge and to provide a realistic solution methodology, this paper develops three advanced metaheuristic algorithms based on the sine cosine algorithm, grey wolf optimizer, and harmony search algorithm. These algorithms are developed by embedding a search strategy to tackle the binary and integer variables of the problem, and a simple constraint handling scheme to handle a large number of linear and non-linear constraints of the problem. To validate the performance of the proposed algorithms, eight different case studies are conducted, where the number of stops, buses, and prediction horizons are varied to analyze the computational efficiency of the proposed algorithms. Performance comparison of algorithms indicates that the proposed metaheuristic based on the harmony search algorithm can resolve the computational inefficiency issue comparatively more effectively, especially for large scales, compared to the other algorithms and commercial solvers. Therefore, beyond the methodological contributions, this study highlights the potential applications of the proposed enhanced versions of the metaheuristic algorithms in improving the planning for real-world urban bus transport by reducing passengers’ waiting times and enhancing overall service reliability.