UAV-CAS — Calibrated Digital-Twin Dataset & Benchmark
Dataset · Code · Paper · Full page
A labeled flow dataset and benchmark for UAV swarm intrusion detection, generated by a Containernet digital twin and calibrated against the NSF AERPAW testbed through a four-layer pipeline.
Scale. 99,492 flows, 1,024 configurations, five attack families, nine collaborative compositions, fifteen label classes.
Fidelity. Simulated-to-real RSS divergence of 0.33 Hellinger — below the 0.73 divergence between two independent real measurement campaigns.
What benchmarking 10 IDS revealed. Binary detection saturates above 0.98, which is why the problem looks solved. Stealth-paired collaborative compositions collapse to 0.51–0.86 AUROC, which is where it isn’t.
My role: dataset design, digital-twin construction, the calibration pipeline, and the ten-baseline benchmark.
Stack: Python, Containernet / Mininet, Docker, GNS3, hping3, AERPAW measurement data.
Links