Safety-by-Design Method Defines AI Operational Conditions from Data
A new Safety-by-Design method enables a posteriori definition of Operational Design Domains (ODD) for AI-based systems using previously collected data. The approach, validated via Monte Carlo methods and a real-world aviation collision-avoidance use case, addresses the challenge of incomplete ODD descriptions that hinder certification in safety-critical applications. Traditional ODD creation relies on early-stage expert knowledge and standards, but this method uses a multi-dimensional kernel-based representation to derive conditions from data. The paper is published on arXiv (2601.22118) and highlights the growing role of AI in safety-critical domains.
Key facts
- Paper arXiv:2601.22118 proposes Safety-by-Design method for ODD definition
- Method uses multi-dimensional kernel-based representation from collected data
- Validated through Monte Carlo methods and aviation collision-avoidance use case
- Addresses incomplete ODD descriptions in safety-critical AI systems
- Traditional ODD relies on early expert knowledge and standards
- AI is increasingly used in safety-critical applications
- Method enables a posteriori ODD definition from data
- Published on arXiv as a replace announcement
Entities
Institutions
- arXiv