SLEIGHT: Real-time Closed-Loop Attack Analytics Framework for Holistic Intrusion...
summary
SLEIGHT is a real-time attack analytics framework that trains a conditional GAN with a phase-conditioned RL policy (PPO) plus an attacker-side plausibility scorer to synthesize bounded, physically plausible telemetry perturbations. It runs against Pixhawk hardware-in-the-loop via MAVSDK, enforces a feasibility gate (stealth, amplitude, temporal smoothness, domain validity), and benchmarks IDSs including PID-Piper and ConFIDe. Results show >96% no-alarm on single-channel GPS, >95% vs ConFIDe, >90% vs PID-Piper, ≈11 m single-UAV drift and >60 m in swarm tests.
AI-written summary · about
chapters (15)
- 0:00SLIGHT framework intro
- 1:13Closed-loop UAV operation and intrusion detection systems
- 2:17Limitations of attack models
- 3:42SLIGHT attack analytics overview
- 6:38Offline plausibility scorer training
- 8:04Real-time attack vector generation
- 9:49Feasibility for stealth & impact
- 11:57Policy training with CGAN & RL
- 13:01Hardware-in-the-loop setup
- 13:41Research questions & tests
- 15:08Comparison with adversarial attacks
- 16:12SLIGHT vs different IDS
- 16:56Stealth impact & mission analysis
- 17:42Swarm attack analysis
- 18:24Conclusion & future work
related talks
09mWIP: Hijacking Attacks on UAV Follow-Me Systems in Realistic Scenarios
28mSLEUTH: Real-time Attack Scenario Reconstruction from COTS Audit Data
13mWIP: Learning Adversarial Attacks on Adaptive Traffic Signal Control Systems...
11mWIP: Evaluating the End-to-End Impact of False Localization Attacks on vSLAM-...
27mSecure AI Orchestration: Mitigate Model-centric Attacks with Flyte
49mFight Smarter: Accelerate Your SOC with AI Discovered Attacks
47mRSAC Security Scholar Poster Pitch-Off
38mSecurity for AI Agents Using an Ensemble of Fine-tuned... (Lidan Hazout, Bar Kaduri)