driving scene configuration

This work addresses critical challenges in autonomous driving development: the absence of standardized evaluation criteria for traffic datasets and simulators, inconsistent annotation practices, scenario coverage bias, and insufficient geographic and environmental robustness. To tackle these issues, we conduct a systematic survey and empirical evaluation, constructing a comprehensive assessment matrix encompassing over 120 real-world datasets and 30+ simulation platforms—marking the first effort to unify annotation standards and delineate simulator functional boundaries. We propose a standardized annotation pipeline framework to quantitatively analyze annotation quality, data distribution characteristics, and environmental adversarial effects. Our analysis reveals the pivotal role of synthetic data in alleviating real-data bottlenecks. Results demonstrate that high-fidelity, geographically diverse, and semantically consistent synthetic data-driven paradigms constitute a key pathway to enhancing generalization and reliability across the full-stack perception–prediction–planning pipeline.

To address the scarcity of rare safety-critical scenarios in autonomous driving motion planning, the high cost and limited coverage of manual annotation for long-tail risks, this paper proposes a traffic scenario generation method leveraging in-context learning (ICL) with large language models (LLMs). The approach requires no model fine-tuning or handcrafted programming; instead, it synthesizes executable CARLA simulation scripts directly from natural-language scenario descriptions, enabling low-cost, highly diverse, and customizable critical scenario construction. To our knowledge, this is the first work to apply ICL to script-based traffic scenario modeling, substantially improving the realism and generalizability of synthetic data. Experimental results demonstrate that motion planners trained on the synthesized data achieve a 23.6% improvement in success rate on real-world critical-risk scenario evaluation, with marked gains in safety and robustness.

This work addresses the limitations of current scenario-based autonomous driving testing, which often relies on imperative definitions that struggle to efficiently generate comprehensive and specification-compliant simulation scenarios. The authors propose RoadLogic, a novel approach that, for the first time, enables automatic instantiation of executable simulations directly from declarative OpenSCENARIO domain-specific language (DSL) specifications. By integrating Answer Set Programming for abstract reasoning and motion planning, RoadLogic generates feasible vehicle trajectories while ensuring correctness through formal specification monitoring. Implemented within the CommonRoad framework, RoadLogic automatically produces diverse, realistic, and specification-conforming simulation trajectories for multiple OpenSCENARIO 2 (OS2) scenario types within minutes, significantly enhancing both the systematic coverage and behavioral diversity of autonomous driving testing.

To address insufficient traffic simulation fidelity and ambiguous requirement specifications in driving simulators, this paper proposes a systematic traffic simulation requirement analysis method based on sub-goal decomposition. The experimental objective is hierarchically decomposed into verifiable sub-goals—including microscopic traffic modeling, agent behavioral modeling, and visual rendering—thereby establishing a structured, traceable mapping from research objectives to simulation configuration. This method establishes, for the first time, an explicit linkage between traffic simulation design and underlying experimental goals, significantly enhancing simulation fidelity, experimental validity, and participant immersion. Empirical evaluation demonstrates that the proposed framework supports high-fidelity development and human–autonomy interaction testing of autonomous driving systems.

Existing driving simulators suffer from systematic limitations in multimodal perceptual fidelity, scene realism, closed-loop evaluation capability, traffic diversity, multi-agent collaboration, and computational efficiency. This paper proposes SimFusion, a high-fidelity closed-loop driving simulation framework that, for the first time, jointly models background/foreground-decoupled 3D scene reconstruction and multimodal view synthesis—powered by Neural Radiance Fields (NeRF)—to unify perceptual authenticity with geometric accuracy. It introduces a closed-loop simulation engine supporting non-reactive scenario generation, safety-critical testing, and multi-agent coordination. Leveraging real-world multimodal sensor data, SimFusion significantly narrows the perception gap between simulation and reality. Evaluated on multiple benchmarks, it enhances the reliability of driving policy assessment while enabling scalable, low-cost, and highly diverse closed-loop testing.

To address insufficient diversity and limited coverage of critical scenarios in natural language–driven traffic scene generation for autonomous driving simulation, this paper proposes the first large language model (LLM)–driven end-to-end text-to-scene generation framework. The method integrates semantic parsing, vector-based retrieval, multi-factor road ranking, and joint planning of dynamic road networks and agent behaviors—thereby overcoming reliance on predefined trajectories. It enables semantically controllable generation of both routine and high-risk driving scenarios and seamlessly interfaces with the CARLA simulator. Evaluated on the SafeBench benchmark, the framework reduces the average collision rate from 8.0% to 3.5%, while significantly improving narrative coherence and causal reasoning in scene descriptions. This work establishes a scalable, interpretable paradigm for safety-critical scenario generation in autonomous driving validation.

Autonomous driving in the real world faces significant challenges, including data scarcity, stringent safety constraints, and limited generalization across diverse environments. This work presents a systematic review of synthetic data and virtual simulation techniques applied to perception, planning, and validation in autonomous systems. It proposes an integrated three-dimensional framework that combines synthetic data generation, digital twin–based validation, and domain adaptation, further enhanced by vision–language models to improve simulation fidelity and semantic generalization. By establishing a comprehensive taxonomy of current methodologies, the study identifies critical research directions—such as safety verification, cooperative autonomy, and simulation-driven policy learning—to advance the development of scalable, safe, and generalizable autonomous driving systems.

This work addresses the limitations of existing autonomous driving testing methods, which rely on handcrafted templates or fixed models and struggle to efficiently reproduce complex real-world failure scenarios. The authors propose a modular synthesis framework powered by large language models (LLMs) that, for the first time, incorporates natural language descriptions and contextual information from real accident reports into the scenario generation process. Operating under predefined testing constraints, the framework automatically constructs diverse driving scenarios. Integrated with the MetaDrive platform and validated using NHTSA crash data, the approach successfully generates test cases covering four road types, three non-ego vehicle motion patterns, and construction-zone anomalies. Remarkably, only 20 generated scenarios effectively expose latent system failures, significantly outperforming conventional testing methodologies.

This work addresses the challenge of efficiently translating real-world multimodal accident reports—comprising textual descriptions and hand-drawn sketches—into high-fidelity, executable simulation scenarios for autonomous driving safety validation, a bottleneck that hinders scalable testing. The authors propose a framework leveraging a large language model (GPT-4o mini) and an extended version of the Scenic domain-specific language, introducing a probabilistic intermediate representation that decouples high-level semantic understanding from low-level scene rendering. This enables automated extraction of semantic scene configurations and generation of simulatable test cases. Evaluated on the NHTSA CIREN dataset, the approach achieves 100% accuracy in reconstructing environmental and road network attributes and 97%–98% accuracy in trajectory extraction. Furthermore, across 2,000 generated scenario variants, the method consistently triggers the intended traffic violations, demonstrating robustness, effectiveness, and scalability.

Existing simulation-based testing methods for autonomous driving often rely on templates, manual construction, or random generation when verifying formal safety requirements, leading to inefficient coverage and potential omission of critical scenarios. This work proposes the first framework that directly translates LTLf specifications into a structured, exhaustive test scenario space. By integrating the SCENEFLOW language with a systematic generation strategy, the approach enables efficient and comprehensive coverage of behaviors mandated by the specification. The method drastically reduces manual intervention, achieving up to twice the test coverage of baseline approaches across diverse LTLf specifications, improving coarse-grained metrics by 75%, and attaining equivalent coverage with only one-sixth the number of simulations.

Traditional on-road testing of autonomous driving systems (ADS) suffers from prohibitive costs and insufficient scenario coverage, hindering rigorous safety validation. Method: This study systematically reviews scene generation techniques from 2015–2025, with focused analysis on AI-driven generative approaches (2023–2025). It integrates large language models, diffusion models, GANs, reinforcement learning, ontologies, and naturalistic driving data. Contribution/Results: We propose (i) a multimodal extended taxonomy; (ii) an ethics- and safety-aware design checklist; (iii) an ODD coverage map; and (iv) a scenario difficulty spectrum—achieving, for the first time, a standardized evaluation framework, explicit human-factor modeling, and multimodal co-generation of scenarios. The framework delivers a reusable methodology for academia and a reproducible testing guideline for industry, significantly improving efficiency in generating safety-critical scenarios and enhancing ODD adaptability—thereby supporting regulatory compliance and deployment of L4+ ADS.