On Adversarial Attacks In Acoustic Drone Localization

TL;DR

This work provides a comprehensive analysis of the effect of PGD adversarial attacks over acoustic drone localization, and develops an algorithm for adversarial perturbation recovery, capable of markedly diminishing the affect of such attacks in this setting.

Abstract

Multi-rotor aerial autonomous vehicles (MAVs, more widely known as "drones") have been generating increased interest in recent years due to their growing applicability in a vast and diverse range of fields (e.g., agriculture, commercial delivery, search and rescue). The sensitivity of visual-based methods to lighting conditions and occlusions had prompted growing study of navigation reliant on other modalities, such as acoustic sensing. A major concern in using drones in scale for tasks in non-controlled environments is the potential threat of adversarial attacks over their navigational systems, exposing users to mission-critical failures, security breaches, and compromised safety outcomes that can endanger operators and bystanders. While previous work shows impressive progress in acoustic-based drone localization, prior research in adversarial attacks over drone navigation only addresses visual sensing-based systems. In this work, we aim to compensate for this gap by supplying a comprehensive analysis of the effect of PGD adversarial attacks over acoustic drone localization. We furthermore develop an algorithm for adversarial perturbation recovery, capable of markedly diminishing the affect of such attacks in our setting. The code for reproducing all experiments will be released upon publication.

Figures (14)

• Figure 1: Adversarial Acoustic Localization Setting -. Localization model input (green) is the drone sound response (blue) perturbed with an external speaker adversarial interference (red).
• Figure 2: Adversarial Pipeline Overview - sound from both drone sound and adversarial source is convolved with NAF-induced RIRs and superimposed at the sensor, prior to being fed into the localization model. Localization loss gradients guide source amplitude optimization.
• Figure 3: Summary of Perturbation Delineation Method - rotor self-sound (blue) under different phase modulations (black arrows depict rotor angular location at $t=0$). In red is the perturbation sound. The two are superimposed to the microphone-sampled signal (green).
• Figure 4: Mean RMS with and without source location optimization - across varying amplitude and power bounds ($\beta \text{ and } \gamma$).
• Figure 5: Clean, perturbed and delineation-recovered spatial error distribution for selected attack bounds - across different scenes (apt., room, and office), measured in mean RMS error across drone orientations. Optimized source location highlighted in blue.