Reproduce “Deep Reinforcement Learning Based Systems for Safety Critical Applications in Aerospace” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Abedin Sherifi
Abedin Sherifi, affiliated with the University of Texas at Austin and Pratt and Whitney, explores the integration of Deep Reinforcement Learning (DRL) into safety-critical aerospace applications, detailing how existing certification standards are insufficient and proposing new verification strategies and runtime assurance mechanisms to enable safe deployment.
Key insights
- Safe deployment of DRL in aerospace necessitates a fundamental shift from traditional deterministic assurance methods to new frameworks that account for AI's unique characteristics.
- Runtime assurance, featuring continuous monitoring and a fail-safe switch to classical control, is a pragmatic and essential component for mitigating DRL's inherent risks in safety-critical environments.
- Emerging regulatory guidance from bodies like EASA and NASA (e.g., High-Level Properties, Overarching Properties) provides a foundational structure for formalizing DRL system intent, correctness, and innocuity arguments for certification.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.