Selected work

Representative Papers

Distribution-Aware Diffusion-LLM for Robust Ultra-Long-Term Time Series Forecasting

Jun 22, 2026

This work addresses the challenges of inadequate probabilistic calibration and difficulties in aligning heterogeneous representations when applying large language models (LLMs) to multivariate time series forecasting. To overcome these limitations, we propose a novel framework that integrates conditional diffusion mechanisms with LLMs. Our approach jointly models the conditional distribution of future data within a shared latent space, enabling semantic alignment and distribution-aware prediction, while incorporating distributional regularization to enhance robustness. As the first study to combine distribution-aware diffusion processes with LLMs for time series forecasting, our method achieves significant performance gains over existing approaches across six long-horizon benchmarks—including ETT, Weather, and ECL—with particularly strong results in ultra-long-term and few-shot forecasting scenarios.

0 citationsRead paper

TriFlow: Generating Artist-Like 3D Mesh Topology via Nearest-Vertex Vector Fields

Jun 18, 2026

This work proposes a novel method for directly generating compact 3D meshes with artist-friendly triangle topology from geometric conditions such as signed distance fields. The key innovation lies in the introduction of a Nearest Vertex Field (NVF)—an implicit representation of surface mesh topology—combined with a latent flow matching model to generate this field. Structured, high-fidelity meshes are then produced end-to-end through NVF-driven region clustering followed by a topology-aware constrained Quadric Error Metric (QEM) simplification algorithm. Compared to existing learning-based approaches, the proposed method achieves substantially improved topological quality and generalization, reducing Chamfer Distance by 90% and accelerating inference by 8×.

0 citationsRead paper

Behavior-Centric Extraction of Scenarios from Highway Traffic Data and their Domain-Knowledge-Guided Clustering using CVQ-VAE

Mar 17, 2026

This work addresses the lack of standardized traffic scenario extraction and the absence of domain knowledge in clustering for autonomous driving validation. To this end, it proposes a behavior-centric scenario extraction method grounded in the “Scenario-as-Specification” paradigm and introduces a class-vector quantized variational autoencoder (CVQ-VAE) that integrates prior driving rules and other domain-specific knowledge to enable interpretable clustering. Evaluated on the highD dataset, the approach achieves standardized, high-fidelity scenario extraction and yields semantically coherent clusters, substantially enhancing the systematic coverage of driving scenarios and the interpretability of the validation pipeline. This provides an efficient and well-structured foundation for autonomous driving testing and verification.

0 citationsRead paper

Foundation Models in Autonomous Driving: A Survey on Scenario Generation and Scenario Analysis

Jun 13, 2025

Autonomous driving faces challenges in generating complex, rare scenarios and achieving high fidelity for safety-critical cases. This paper systematically surveys foundational models—including large language models (LLMs), vision-language models (VLMs), multimodal LLMs, diffusion models, and world models—for scenario generation and analysis. We propose, for the first time, a unified taxonomy tailored to autonomous driving. Our structured framework encompasses methods, datasets, simulation platforms, and evaluation metrics; we introduce novel domain-specific metrics and two analytical dimensions—causal fidelity and safety-critical fidelity. We publicly release an actively maintained literature repository and supplementary materials. Synthesizing over 100 state-of-the-art works, we catalog major open-source resources and explicitly identify key bottlenecks (e.g., insufficient causal modeling, distortion of safety-critical scenarios) alongside promising future directions. This work provides a systematic foundation for enhancing both diversity and realism in autonomous driving scenario generation.

0 citationsRead paper

Assessing the Completeness of Traffic Scenario Categories for Automated Highway Driving Functions via Cluster-based Analysis

Jun 03, 2025

Ensuring safety validation for highway autonomous driving necessitates comprehensive assessment of traffic scenario classification completeness. Method: We propose CVQ-VAE—the first high-dimensional trajectory embedding and clustering method integrating vector quantization with a variational autoencoder—and employ quantitative metrics (e.g., silhouette coefficient) to evaluate clustering quality. Contribution/Results: Evaluated on the highD dataset, CVQ-VAE significantly outperforms existing clustering baselines. Crucially, it reveals an interpretable trade-off between the number of scenario categories and coverage completeness: while increasing category count enhances discriminative capability, it exponentially escalates the data volume required to achieve completeness. This work establishes theoretical bounds and provides an actionable, metric-driven evaluation framework for autonomous driving test case generation.

0 citationsRead paper

Recent publications

Latest Papers

Distribution-Aware Diffusion-LLM for Robust Ultra-Long-Term Time Series Forecasting

Jun 22, 2026

This work addresses the challenges of inadequate probabilistic calibration and difficulties in aligning heterogeneous representations when applying large language models (LLMs) to multivariate time series forecasting. To overcome these limitations, we propose a novel framework that integrates conditional diffusion mechanisms with LLMs. Our approach jointly models the conditional distribution of future data within a shared latent space, enabling semantic alignment and distribution-aware prediction, while incorporating distributional regularization to enhance robustness. As the first study to combine distribution-aware diffusion processes with LLMs for time series forecasting, our method achieves significant performance gains over existing approaches across six long-horizon benchmarks—including ETT, Weather, and ECL—with particularly strong results in ultra-long-term and few-shot forecasting scenarios.

0 citationsRead paper

TriFlow: Generating Artist-Like 3D Mesh Topology via Nearest-Vertex Vector Fields

Jun 18, 2026

This work proposes a novel method for directly generating compact 3D meshes with artist-friendly triangle topology from geometric conditions such as signed distance fields. The key innovation lies in the introduction of a Nearest Vertex Field (NVF)—an implicit representation of surface mesh topology—combined with a latent flow matching model to generate this field. Structured, high-fidelity meshes are then produced end-to-end through NVF-driven region clustering followed by a topology-aware constrained Quadric Error Metric (QEM) simplification algorithm. Compared to existing learning-based approaches, the proposed method achieves substantially improved topological quality and generalization, reducing Chamfer Distance by 90% and accelerating inference by 8×.

0 citationsRead paper

Behavior-Centric Extraction of Scenarios from Highway Traffic Data and their Domain-Knowledge-Guided Clustering using CVQ-VAE

Mar 17, 2026

This work addresses the lack of standardized traffic scenario extraction and the absence of domain knowledge in clustering for autonomous driving validation. To this end, it proposes a behavior-centric scenario extraction method grounded in the “Scenario-as-Specification” paradigm and introduces a class-vector quantized variational autoencoder (CVQ-VAE) that integrates prior driving rules and other domain-specific knowledge to enable interpretable clustering. Evaluated on the highD dataset, the approach achieves standardized, high-fidelity scenario extraction and yields semantically coherent clusters, substantially enhancing the systematic coverage of driving scenarios and the interpretability of the validation pipeline. This provides an efficient and well-structured foundation for autonomous driving testing and verification.

0 citationsRead paper

Foundation Models in Autonomous Driving: A Survey on Scenario Generation and Scenario Analysis

Jun 13, 2025

Autonomous driving faces challenges in generating complex, rare scenarios and achieving high fidelity for safety-critical cases. This paper systematically surveys foundational models—including large language models (LLMs), vision-language models (VLMs), multimodal LLMs, diffusion models, and world models—for scenario generation and analysis. We propose, for the first time, a unified taxonomy tailored to autonomous driving. Our structured framework encompasses methods, datasets, simulation platforms, and evaluation metrics; we introduce novel domain-specific metrics and two analytical dimensions—causal fidelity and safety-critical fidelity. We publicly release an actively maintained literature repository and supplementary materials. Synthesizing over 100 state-of-the-art works, we catalog major open-source resources and explicitly identify key bottlenecks (e.g., insufficient causal modeling, distortion of safety-critical scenarios) alongside promising future directions. This work provides a systematic foundation for enhancing both diversity and realism in autonomous driving scenario generation.

0 citationsRead paper

Assessing the Completeness of Traffic Scenario Categories for Automated Highway Driving Functions via Cluster-based Analysis

Jun 03, 2025

Ensuring safety validation for highway autonomous driving necessitates comprehensive assessment of traffic scenario classification completeness. Method: We propose CVQ-VAE—the first high-dimensional trajectory embedding and clustering method integrating vector quantization with a variational autoencoder—and employ quantitative metrics (e.g., silhouette coefficient) to evaluate clustering quality. Contribution/Results: Evaluated on the highD dataset, CVQ-VAE significantly outperforms existing clustering baselines. Crucially, it reveals an interpretable trade-off between the number of scenario categories and coverage completeness: while increasing category count enhances discriminative capability, it exponentially escalates the data volume required to achieve completeness. This work establishes theoretical bounds and provides an actionable, metric-driven evaluation framework for autonomous driving test case generation.

0 citationsRead paper