Counter-UAS Research
Research · UAV Security · 2026
Abstract—This project presents RAC-SRPL for detecting formation-preserving GNSS spoofing through randomized authenticated anchor audits, and VERGE-CUAS for uncertainty-aware, collision-conscious clearance of partially responsive drone swarms. Seeded position-domain simulations provide reproducible evidence; hardware validation remains future work.
Problem
A coordinated GNSS spoofing attack can move an entire drone swarm while preserving pairwise distances, so relative range checks still look normal. Even a fixed audited leader can become a blind spot: rotate the formation around that sentinel and the leader remains unchanged while the rest of the swarm drifts toward sensitive airspace.
Approach
The research now has two reproducible tracks in one GitHub package. RAC-SRPL detects and bounds absolute swarm displacement with sparse randomized authenticated ground-anchor challenges, then only permits recovery when the Swarm Rigid-Transform Protection Level is finite. VERGE-CUAS models the defensive side: bounded control nudges responsive intruders toward verified exits, identifies resistant members, and issues clearance only when the full protected zone is certified under uncertainty and packet loss.
Features
- Formation-preserving threat model — Shows why pairwise inter-UAV ranging cannot observe a common rigid translation or rotation of the swarm.
- Fixed-sentinel blind spot — Constructs the case where a permanently audited leader stays unchanged while the rest of the formation rotates around it.
- Randomized anchor challenges — Selects both the audited swarm member and anchor subset unpredictably, then compares GNSS against surveyed-anchor position estimates.
- SRPL-gated recovery — Computes an absolute Swarm Rigid-Transform Protection Level and treats it as infinite until the audit history makes the transform observable.
- Reproducible RAC-SRPL evidence — Position-domain Monte Carlo results report random-member auditing detecting rigid attacks that relative-only and fixed-sentinel checks miss.
- VERGE-CUAS clearance simulator — Models risk-aware eviction of responsive drones from a protected zone while preserving separation and flagging resistant members.
- 7,100 seeded VERGE-CUAS trials — Includes main comparisons, ablations, and sensitivity sweeps; the strongest result is the verification layer, while static route-risk assignment is reported as mixed.
- Safety-first validation boundary — The package stays in simulation, recorded data, cabled RF, or shielded-test territory and does not provide RF spoofing or jamming instructions.
Outcome
The latest repository packages two IEEE-style research manuscripts and reproducibility materials. RAC-SRPL contributes a randomized anchor-audit detector and absolute swarm protection bound for formation-preserving GNSS spoofing. VERGE-CUAS contributes a verifiable clearance simulator whose completed computational study covers 7,100 deterministic seeded trials. The repo is explicit that these are position-domain simulations, with hardware-in-the-loop testing, calibrated tracking errors, institutional approval, and independent review still required before publication or deployment claims.
Stack
Python, NumPy, pandas, Matplotlib, LaTeX, Monte Carlo simulation
Links
GitHub · Paper (PDF)
All projects · Harshil Patel