Dynamic watermarking detects replay attacks in vision controllers
Original title: Dynamic Watermarking and Action-Conditioned Latent Dynamics for Secure Vision-Based Controllers
RL policy balances watermark energy against control distortion on UR5 robot.
Adaptive dynamic-watermarking framework adds Gaussian watermarks to control inputs for vision-based systems. Latent dynamics networks predict image evolution to detect replay attacks via discrepancy. RL optimizes covariance using detector confidence.
- Private Gaussian watermark to nominal inputs
- LDN predicts latent response from observations
- RL adjusts watermark covariance
- Evaluated on UR5 reaching target avoiding human arm
From the original article
1 min read at arxiv.orgReplay attacks are difficult to detect in vision-based control systems because previously recorded observations can remain plausible while the physical system follows a different trajectory. This paper develops an adaptive dynamic-watermarking framework for detecting replay attacks against vision-based controllers.
By Aditya Parameshwaran · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.