道路车辆人工智能，ISOPAS8800-RoadVehicles-Safetyandartificialintelligence-2023

需积分: 3 14 浏览量 2024-11-26 15:13:16 上传评论收藏 5.11MB PDF 举报

资源推荐

资源详情

资源评论

ISO/TC 22/SC 32

ISO/CD PAS 8800(en)

Secretariat: JISC

Road Vehicles — Safety and artificial intelligence

Véhicules routiers — Sécurité et intelligence artificielle

ISO/CD PAS 8800:2023(E) 
© ISO 2023 – All rights reserved 
 
iii 
Contents 
Foreword ......................................................................................................................................................................................... viii 
Introduction ....................................................................................................................................................................................... ix 
Scope ....................................................................................................................................................................................... 1 
Normative references ....................................................................................................................................................... 1 
Terms and definitions ...................................................................................................................................................... 2 
1  General AI-related definitions ...................................................................................................................................... 2 
2  Data-related definitions .................................................................................................................................................. 9 
3  General safety-related definitions............................................................................................................................ 10 
4  Safety: Root cause-, error-and failure-related definitions ............................................................................. 13 
5  Miscellaneous definitions ............................................................................................................................................ 15 
Abbreviated terms .......................................................................................................................................................... 18 
Requirements for compliance .................................................................................................................................... 19 
1  Purpose ............................................................................................................................................................................... 19 
2  General requirements ................................................................................................................................................... 19 
3  Interpretations of tables and figures ...................................................................................................................... 19 
AI within the context of road vehicles system safety engineering and basic concepts ..................... 19 
1  Application of the ISO 26262 series for the development of AI systems ................................................. 19 
2  Interactions with encompassing system-level safety activities .................................................................. 20 
3  Mapping of abstraction layers between ISO 26262, ISO/IEC 22989 and this document ................. 24 
4  Example architecture for an AI system .................................................................................................................. 27 
5  Types of AI models ......................................................................................................................................................... 28 
6  AI technologies of a ML model ................................................................................................................................... 28 
7  Error concepts, fault models and causal models ............................................................................................... 29 
7.1  Cause-and-effect chain ............................................................................................................................................. 29 
7.2  Root cause classes ...................................................................................................................................................... 30 
7.3  Error classification based on the safety impact ............................................................................................. 32 
AI safety management ................................................................................................................................................... 33 
1  Objectives ........................................................................................................................................................................... 33 
2  Prerequisites and supporting information ........................................................................................................... 33 
3  General requirements ................................................................................................................................................... 33 
4  Reference AI safety lifecycle ....................................................................................................................................... 36 
5  Iterative development paradigms for AI systems ............................................................................................. 38 
6  Work products ................................................................................................................................................................. 40 
Assurance arguments for AI systems ..................................................................................................................... 40 
1  Objectives ........................................................................................................................................................................... 40 
2  Prerequisites and supporting information ........................................................................................................... 40 
3  General requirements ................................................................................................................................................... 41 

ISO/CD PAS 8800:2023(E) 
© ISO 2023 – All rights reserved 
 
iv 
4  AI system-specific considerations in assurance arguments.......................................................................... 41 
5  Structuring assurance arguments for AI systems.............................................................................................. 43 
5.1  Context of the assurance argument .................................................................................................................... 43 
5.2  Categories of evidence ............................................................................................................................................. 44 
6  The role of quantitative targets and qualitative arguments ......................................................................... 45 
7  Evaluation of the assurance argument ................................................................................................................... 46 
8  Work products ................................................................................................................................................................. 47 
Derivationof AI safetyrequirements ....................................................................................................................... 47 
1  Objectives ........................................................................................................................................................................... 47 
2  Prerequisites and supporting information ........................................................................................................... 48 
3  General requirements ................................................................................................................................................... 48 
4  General workflow for deriving safety requirements ........................................................................................ 49 
5  Deriving AI safety requirements on supervised machine learning ............................................................ 51 
5.1  The need for refinedAI safety requirements .................................................................................................. 51 
5.2  Derivation of refined AI safety requirements to manage uncertainty ................................................. 53 
5.3  Refinement of the input space definition for AI safety lifecycle ............................................................. 56 
5.4  Restricting the occurrence of AI output insufficiencies ............................................................................. 56 
5.5  Metrics, measurements and threshold design ............................................................................................... 59 
5.6  Considerations for deriving safety requirements ......................................................................................... 60 
6  Work products ................................................................................................................................................................. 61 
Selection of AI technologies, architectural and development measures ................................................. 61 
1  Objectives ........................................................................................................................................................................... 61 
2  Prerequisites ..................................................................................................................................................................... 61 
3  General requirements ................................................................................................................................................... 62 
4  Architecture and development process design or refinement..................................................................... 63 
5  Examples of architectural and development measures for AI systems.................................................... 63 
6  Work products ................................................................................................................................................................. 67 
Data-related considerations ....................................................................................................................................... 67 
1  Objectives ........................................................................................................................................................................... 67 
2  Prerequisites and supporting information ........................................................................................................... 68 
3  General requirements ................................................................................................................................................... 68 
4  Dataset lifecycle ............................................................................................................................................................... 69 
4.1  Datasets and the AI safety lifecycle..................................................................................................................... 69 
4.2  Reference dataset lifecycle ..................................................................................................................................... 70 
4.3  Dataset safety analysis ............................................................................................................................................. 71 
4.4  Dataset requirements development ................................................................................................................... 77 
4.5  Dataset design ............................................................................................................................................................. 80 
4.6  Dataset implementation .......................................................................................................................................... 81 
4.7  Dataset verification ................................................................................................................................................... 82 

ISO/CD PAS 8800:2023(E) 
© ISO 2023 – All rights reserved 
 
v 
4.8  Dataset validation ...................................................................................................................................................... 83 
4.9  Dataset maintenance ................................................................................................................................................ 83 
5  Work products ................................................................................................................................................................. 84 
Verification and validation of AI system ............................................................................................................... 84 
1  Objectives ........................................................................................................................................................................... 84 
2  Prerequisites and supporting information ........................................................................................................... 85 
3  General requirements ................................................................................................................................................... 85 
4  AI/ML specific challenges to verification and validation ............................................................................... 87 
5  Verification and validation of the AI system ........................................................................................................ 88 
5.1  Scope of verification and validation of the AI system ................................................................................. 88 
5.2  AI component testing ............................................................................................................................................... 90 
5.3  Methods for testing the AI component .............................................................................................................. 92 
5.4  AI system integration and verification .............................................................................................................. 94 
5.5  Virtual testing vs physical testing ....................................................................................................................... 95 
5.6  Evaluation of the safety-related performance of the AI system ............................................................. 96 
5.7  AI systemsafety validation ..................................................................................................................................... 97 
6  Work products ................................................................................................................................................................. 98 
Safety analysis of AI systems ...................................................................................................................................... 98 
1  Objectives ........................................................................................................................................................................... 98 
2  Prerequisites and supporting information ........................................................................................................... 98 
3  General requirements ................................................................................................................................................... 99 
4  Safety analysis of the AI system ................................................................................................................................ 99 
4.1  Scope of the AI safety analysis .............................................................................................................................. 99 
4.2  Safety analysis based on the results of testing ............................................................................................. 101 
4.3  Safety analysis techniques .................................................................................................................................... 102 
5  Work products ............................................................................................................................................................... 103 
Measures during operation ....................................................................................................................................... 103 
1  Objectives ......................................................................................................................................................................... 103 
2  Prerequisites and supporting information ......................................................................................................... 103 
3  General requirements ................................................................................................................................................. 104 
4  Planning for operation and continuous assurance ......................................................................................... 105 
4.1  Safety risk of the AI system during operation phase ................................................................................. 105 
4.2  Safety activities during the operation phase ................................................................................................ 105 
5  Continual, periodic re-evaluation of the assurance argument ................................................................... 106 
6  Measures to assure safety of the AI system during operation ................................................................... 107 
6.1  General .......................................................................................................................................................................... 107 
6.2  Technical safety measures ................................................................................................................................... 107 
6.3  Safe operation guidance and misuse prevention in the field ................................................................. 109 
7  Field data collection ..................................................................................................................................................... 109 