bft-uav-swarm-validation-lab

First public release of the validation testbed accompanying the IEEE Access submission "Byzantine-Fault-Tolerant Hierarchical Navigation for GNSS-Denied UAV Swarms: Architecture, Theoretical Analysis, and Empirical Validation" (Kalynovskyi, 2026; submitted to IEEE Access).

What's archived

The complete nine-phase empirical validation campaign for the manuscript's architecture and theoretical analysis. Every numerical claim in §VI and §VII of the manuscript traces to a specific raw-data file in workspaces/, via the explicit map in CODE_TO_CLAIM_MAPPING.md.

Validation campaign

• Phase 1 — Proof-of-Observation FAR/FRR (AUC = 0.9996)
• Phase 2 — Reputation-driven Byzantine exclusion (30/30 trials, 0 false honest exclusions)
• Phase 3 — CEP versus relay depth (23 m at 3 hops, empirical δ = 0.088)
• Phase 4 — Failover timing (p99 = 6.24 s across 1200 measurements)
• Phase 5 — GNSS-spoof detection (98.3 % detection, 0 false positives, p99 = 0.60 s)
• Phase 6 — Progressive attrition with admission control (14/15 missions survive)
• Phase 7 — Single-pillar ablation (4 pillars stress-tested)
• Phase 8 — Vanilla-Raft baseline + Byzantine adversarial measurement
• Phase 9 — Proof-of-Observation same-region forgery resistance (pre-registered)

Stack pin

ROS 2 Jazzy · Gazebo Harmonic · PX4-Autopilot (SITL) · GTSAM 4.2 · Python 3.12 · CUDA ≥12.4.

Reproduction

docker pull ghcr.io/ok-research/bft-uav-hierarchical-navigation:gpu
git clone https://github.com/ok-research/bft-uav-hierarchical-navigation.git
cd bft-uav-hierarchical-navigation
bash scripts/phase{N}-campaign.sh   # N in 1..9

Per-phase reproducibility documents with Docker image digest, git commit at sweep start, and sweep-parameter manifest are at docs/results/phase-{1..9}.md.

Companion repository

Abstract Monte Carlo simulators for §§VIII–X of the manuscript: github.com/ok-research/bft-uav-swarm-validation (tag v1.2).

License

MIT. See LICENSE.