Runtime Safety Filtering for Learned Small UAS Separation Policies under GNSS Degradation
↳ the Digital Avionics Systems Conference (DASC) 2026
// abstract
This study compares action filtering and observation filtering for learned small-UAS separation policies under adversarial GNSS degradation. Both approaches estimate a worst-case traffic state. In the reported experiments, observation filtering reduces near mid-air collisions by 90%, while action filtering provides negligible safety improvement.
// highlights
- Compares action filtering and observation filtering under GNSS degradation.
- Observation filtering reduces near mid-air collisions by 90% in the reported experiments.
// notes
This paper compares filtering a learned policy’s actions with filtering its observations under GNSS degradation. Humbled and honored to have this work as Best of the UAM/UAS Traffic Management and Separation Assurance session, Best of the Air traffic Management Track and 3rd Best of Conference paper awards at DASC 2026.
Presented at DASC 2026. Presentation details.
// cite
Zongo, A., & Wei, P. (2026). "Runtime Safety Filtering for Learned Small UAS Separation Policies under GNSS Degradation." arXiv:2607.10014. Accepted at DASC 2026.