AI and ML Design Challenges for Autonomous Systems Dependability
A new arXiv paper (2604.27807) examines design challenges in ensuring dependability of safety-critical autonomous systems integrating AI and ML. The study highlights that traditional reliability, safety, and security methods fail to address dynamic behaviors from AI/ML components under real-time, power, and safety constraints. It calls for holistic approaches spanning abstraction layers and combining design-time and run-time assurance.
Key facts
- arXiv paper 2604.27807 published
- Focuses on autonomous systems dependability
- Addresses AI/ML integration challenges
- Traditional methods insufficient for AI/ML dynamics
- Requires holistic multi-layer assurance approach
- Considers real-time, power, safety constraints
- AI/ML offer predictive and adaptive capabilities
- Non-determinism and lack of formal guarantees are key issues
Entities
Institutions
- arXiv