Runtime Safety Filtering for Learned sUAS Separation Policies under GNSS Degradation
↳ the Digital Avionics Systems Conference (DASC) 2026
// abstract
This paper addresses real-time separation assurance and tactical deconfliction of small Unmanned Aerial Systems (sUAS) under GNSS degradation. Building on multi-agent reinforcement learning, we introduce a runtime safety filter that screens and corrects learned separation policies when navigation signals are degraded or spoofed, keeping aircraft within safe separation while preserving the efficiency of the learned policy.
// highlights
- Adds a runtime safety filter on top of learned sUAS separation policies.
- Maintains safe separation under GNSS degradation and spoofing while preserving learned-policy efficiency.
// notes
This paper tries to answer the question: how do we certify safety at for autonomous separation assurance data-driven algorithms at runtime despite GPS degradation and spoofing?
This manuscript has been accepted for presentation at DASC 2026.