Subjective Logic method for runtime safety confidence updates
A new method uses Subjective Logic to dynamically update confidence in safety arguments by integrating design-time evidence with runtime Safety Performance Indicators (SPIs). The approach prioritizes safety responsiveness over exact Bayesian updates, increasing confidence when no violations occur and imposing prompt penalties for violations. Demonstrated on a simulation-based construction zone assist function with ML-based cone detection.
Key facts
- Method uses Subjective Logic for runtime confidence updates
- Integrates design-time evidence and windowed runtime SPIs
- Confidence increases with no violations, penalties for violations
- Prioritizes safety responsiveness over exact Bayesian updates
- Demonstrated on simulation-based construction zone assist function
- Focuses on ML-based construction cone detection component
Entities
Institutions
- arXiv