Academic Journal
Προβολή σε EDS

Enhancing semantic and risk controllability in safety-critical scenario generation: An LLM-guided conditional diffusion method.

Λεπτομέρειες βιβλιογραφικής εγγραφής
Τίτλος: Enhancing semantic and risk controllability in safety-critical scenario generation: An LLM-guided conditional diffusion method.
Συγγραφείς: Zhou H; School of Traffic and Transportation Engineering, Central South University, Changsha, China. Electronic address: hanchuzhou@csu.edu.cn., Li Y; School of Traffic and Transportation Engineering, Central South University, Changsha, China. Electronic address: yaqili19@163.com., Wang C; School of Traffic and Transportation Engineering, Central South University, Changsha, China; Shandong Hi-speed Group Innovation Research Institute, Jinan, China; State Key Lab of Intelligent Transportation System, Jinan, China. Electronic address: wangchengcheng@sdhsg.com., Chang F; School of Resources and Safety Engineering, Central South University, Changsha, China. Electronic address: 222023@csu.edu.cn., Xu P; School of Traffic and Transportation Engineering, Central South University, Changsha, China. Electronic address: pengpengxu@yeah.net., Li C; School of Traffic and Transportation Engineering, Central South University, Changsha, China. Electronic address: 254201017@csu.edu.cn.
Πηγή: Accident; analysis and prevention [Accid Anal Prev] 2026 Oct; Vol. 236, pp. 108648. Date of Electronic Publication: 2026 Jul 11.
Τύπος έκδοσης: Journal Article; Validation Study
Γλώσσα: English
Στοιχεία περιοδικού: Publisher: Pergamon Press Country of Publication: England NLM ID: 1254476 Publication Model: Print-Electronic Cited Medium: Internet ISSN: 1879-2057 (Electronic) Linking ISSN: 00014575 NLM ISO Abbreviation: Accid Anal Prev Subsets: MEDLINE
Imprint Name(s): Publication: Oxford : Pergamon Press
Original Publication: [New York, Pergamon Press]
Ιατρικοί όροι (MeSH): Accidents, Traffic*/prevention & control , Large Language Models* , Semantics* , Autonomous Vehicles* , Safety Management*, Natural Language Processing ; Computer Simulation ; Humans
Περίληψη: Safety-critical scenario generation is essential for autonomous vehicle validation. However, existing methods often struggle to simultaneously achieve semantic controllability, risk-level controllability, and trajectory realism. To address this limitation, we propose a Large Language Model (LLM)-guided conditional diffusion framework for controllable safety-critical scenario generation. Specifically, we first use an LLM to convert natural-language descriptions into structured generation conditions. These conditions are then incorporated into a conditional latent diffusion model to guide scenario generation with controllable semantic and risk features. The generated candidate scenarios are further filtered to retain those with high semantic consistency. Experiments are conducted on the highD dataset for car-following and cut-in scenarios. The results show that the proposed method can generate realistic and risk-controllable scenarios. Compared with representative baseline methods, it achieves a better balance among criticality, realism, and controllability. Ablation experiments further demonstrate that LLM-based semantic conditioning improves both scenario-type and risk-level accuracy. The proposed method provides a flexible and controllable tool to generate safety-critical scenarios for autonomous vehicle testing and safety evaluation.
(Copyright © 2026. Published by Elsevier Ltd.)
Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Contributed Indexing: Keywords: Autonomous vehicles; Diffusion models; Large language models; Safety testing; Traffic safety
Entry Date(s): Date Created: 20260711 Date Completed: 20260924 Latest Revision: 20260924
Update Code: 20260924
DOI: 10.1016/j.aap.2026.108648
PMID: 42435537
Βάση Δεδομένων: MEDLINE
FullText Text:
  Availability: 0
CustomLinks:
  – Url: https://www.doi.org/10.1016/j.aap.2026.108648?
    Name: ScienceDirect (all content) (s7799221)
    Category: fullText
    Text: View record from ScienceDirect
    MouseOverText: View record from ScienceDirect
  – Url: https://resolver.ebsco.com/c/fiv2js/result?sid=EBSCO:cmedm&genre=article&issn=18792057&ISBN=&volume=236&issue=&date=20261001&spage=108648&pages=108648&title=Accident; analysis and prevention&atitle=Enhancing%20semantic%20and%20risk%20controllability%20in%20safety-critical%20scenario%20generation%3A%20An%20LLM-guided%20conditional%20diffusion%20method.&aulast=Zhou%20H&id=DOI:10.1016/j.aap.2026.108648
    Name: Full Text Finder (for New FTF UI) (ns324271)
    Category: fullText
    Text: Full Text Finder
    MouseOverText: Full Text Finder
Header DbId: cmedm
DbLabel: MEDLINE
An: 42435537
AccessLevel: 3
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Enhancing semantic and risk controllability in safety-critical scenario generation: An LLM-guided conditional diffusion method.
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AU" term="%22Zhou+H%22">Zhou H</searchLink>; School of Traffic and Transportation Engineering, Central South University, Changsha, China. Electronic address: hanchuzhou@csu.edu.cn.<br /><searchLink fieldCode="AU" term="%22Li+Y%22">Li Y</searchLink>; School of Traffic and Transportation Engineering, Central South University, Changsha, China. Electronic address: yaqili19@163.com.<br /><searchLink fieldCode="AU" term="%22Wang+C%22">Wang C</searchLink>; School of Traffic and Transportation Engineering, Central South University, Changsha, China; Shandong Hi-speed Group Innovation Research Institute, Jinan, China; State Key Lab of Intelligent Transportation System, Jinan, China. Electronic address: wangchengcheng@sdhsg.com.<br /><searchLink fieldCode="AU" term="%22Chang+F%22">Chang F</searchLink>; School of Resources and Safety Engineering, Central South University, Changsha, China. Electronic address: 222023@csu.edu.cn.<br /><searchLink fieldCode="AU" term="%22Xu+P%22">Xu P</searchLink>; School of Traffic and Transportation Engineering, Central South University, Changsha, China. Electronic address: pengpengxu@yeah.net.<br /><searchLink fieldCode="AU" term="%22Li+C%22">Li C</searchLink>; School of Traffic and Transportation Engineering, Central South University, Changsha, China. Electronic address: 254201017@csu.edu.cn.
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="JN" term="%221254476%22">Accident; analysis and prevention</searchLink> [Accid Anal Prev] 2026 Oct; Vol. 236, pp. 108648. <i>Date of Electronic Publication: </i>2026 Jul 11.
– Name: TypePub
  Label: Publication Type
  Group: TypPub
  Data: Journal Article; Validation Study
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: TitleSource
  Label: Journal Info
  Group: Src
  Data: <i>Publisher: </i><searchLink fieldCode="PB" term="%22Pergamon+Press%22">Pergamon Press </searchLink><i>Country of Publication: </i>England <i>NLM ID: </i>1254476 <i>Publication Model: </i>Print-Electronic <i>Cited Medium: </i>Internet <i>ISSN: </i>1879-2057 (Electronic) <i>Linking ISSN: </i><searchLink fieldCode="IS" term="%2200014575%22">00014575 </searchLink><i>NLM ISO Abbreviation: </i>Accid Anal Prev <i>Subsets: </i>MEDLINE
– Name: PublisherInfo
  Label: Imprint Name(s)
  Group: PubInfo
  Data: <i>Publication</i>: Oxford : Pergamon Press<br /><i>Original Publication</i>: [New York, Pergamon Press]
– Name: SubjectMESH
  Label: MeSH Terms
  Group: Su
  Data: <searchLink fieldCode="MM" term="%22Accidents%2C+Traffic%22">Accidents, Traffic*</searchLink>/<searchLink fieldCode="MM" term="%22Accidents%2C+Traffic+prevention+%26+control%22">prevention & control</searchLink> <br /><searchLink fieldCode="MM" term="%22Large+Language+Models%22">Large Language Models*</searchLink> <br /><searchLink fieldCode="MM" term="%22Semantics%22">Semantics*</searchLink> <br /><searchLink fieldCode="MM" term="%22Autonomous+Vehicles%22">Autonomous Vehicles*</searchLink> <br /><searchLink fieldCode="MM" term="%22Safety+Management%22">Safety Management*</searchLink><br /><searchLink fieldCode="MH" term="%22Natural+Language+Processing%22">Natural Language Processing</searchLink> ; <searchLink fieldCode="MH" term="%22Computer+Simulation%22">Computer Simulation</searchLink> ; <searchLink fieldCode="MH" term="%22Humans%22">Humans</searchLink>
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: Safety-critical scenario generation is essential for autonomous vehicle validation. However, existing methods often struggle to simultaneously achieve semantic controllability, risk-level controllability, and trajectory realism. To address this limitation, we propose a Large Language Model (LLM)-guided conditional diffusion framework for controllable safety-critical scenario generation. Specifically, we first use an LLM to convert natural-language descriptions into structured generation conditions. These conditions are then incorporated into a conditional latent diffusion model to guide scenario generation with controllable semantic and risk features. The generated candidate scenarios are further filtered to retain those with high semantic consistency. Experiments are conducted on the highD dataset for car-following and cut-in scenarios. The results show that the proposed method can generate realistic and risk-controllable scenarios. Compared with representative baseline methods, it achieves a better balance among criticality, realism, and controllability. Ablation experiments further demonstrate that LLM-based semantic conditioning improves both scenario-type and risk-level accuracy. The proposed method provides a flexible and controllable tool to generate safety-critical scenarios for autonomous vehicle testing and safety evaluation.<br /> (Copyright © 2026. Published by Elsevier Ltd.)
– Name: Abstract
  Label: Competing Interests
  Group: Ab
  Data: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
– Name: SubjectMinor
  Label: Contributed Indexing
  Group:
  Data: <i>Keywords: </i>Autonomous vehicles; Diffusion models; Large language models; Safety testing; Traffic safety
– Name: DateEntry
  Label: Entry Date(s)
  Group: Date
  Data: <i>Date Created: </i>20260711 <i>Date Completed: </i>20260924 <i>Latest Revision: </i>20260924
– Name: DateUpdate
  Label: Update Code
  Group: Date
  Data: 20260924
– Name: DOI
  Label: DOI
  Group: ID
  Data: 10.1016/j.aap.2026.108648
– Name: AN
  Label: PMID
  Group: ID
  Data: 42435537
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=cmedm&AN=42435537
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1016/j.aap.2026.108648
    Languages:
      – Code: eng
        Text: English
    PhysicalDescription:
      Pagination:
        StartPage: 108648
    Subjects:
      – SubjectFull: Natural Language Processing
        Type: general
      – SubjectFull: Computer Simulation
        Type: general
      – SubjectFull: Humans
        Type: general
      – SubjectFull: Accidents, Traffic prevention & control
        Type: general
      – SubjectFull: Large Language Models
        Type: general
      – SubjectFull: Semantics
        Type: general
      – SubjectFull: Autonomous Vehicles
        Type: general
      – SubjectFull: Safety Management
        Type: general
    Titles:
      – TitleFull: Enhancing semantic and risk controllability in safety-critical scenario generation: An LLM-guided conditional diffusion method.
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Zhou H
      – PersonEntity:
          Name:
            NameFull: Li Y
      – PersonEntity:
          Name:
            NameFull: Wang C
      – PersonEntity:
          Name:
            NameFull: Chang F
      – PersonEntity:
          Name:
            NameFull: Xu P
      – PersonEntity:
          Name:
            NameFull: Li C
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 10
              Text: 2026 Oct
              Type: published
              Y: 2026
          Identifiers:
            – Type: issn-electronic
              Value: 1879-2057
          Numbering:
            – Type: volume
              Value: 236
          Titles:
            – TitleFull: Accident; analysis and prevention
              Type: main
ResultId 1