Adversarial RL finds sparse DoS attacks to destabilize self-triggered control systems

Researchers have developed a novel adversarial reinforcement learning approach to identify vulnerabilities in self-triggered control systems. This method focuses on finding the sparsest denial-of-service (DoS) attack schedules that can destabilize a system, mirroring the safety certificates used by defenders. The study proves a lower bound on the jam count needed for an adversary to cause a crash and empirically demonstrates the effectiveness of the learned adversary against various controllers on simulation environments like Pendulum, CartPole, and Quadrotor2D. AI

IMPACT This research could lead to more robust control systems by identifying and mitigating novel attack vectors.

RANK_REASON The cluster contains a research paper detailing a novel adversarial reinforcement learning method for analyzing control system vulnerabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• CartPole
• denial-of-service attack
• hold-last medium-access-control
• linear-quadratic regulator
• Lyapunov-increase
• Pendulum
• Quadrotor2D
• reinforcement learning
• Run time assurance of application-level requirements in wireless sensor networks
• Self-triggered Control of Multiple Loops over IEEE 802.15.4 Networks

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →