UniAda: Universal Adaptive Multi-objective Adversarial Attack for End-to-End Autonomous Driving Systems

2026-04-25

AI Summary

This work addresses a critical limitation in existing end-to-end autonomous driving systems, whose adversarial attacks predominantly focus on steering angle deviations while neglecting speed control and the imperceptibility of perturbations, thereby failing to provide a comprehensive safety assessment. To overcome this, the authors propose a general, adaptive, multi-objective white-box attack method that simultaneously disrupts both steering and speed control through an image-agnostic adversarial perturbation generation mechanism. Central to this approach is a multi-objective optimization function incorporating an Adaptive Weighting Strategy (AWS), which, for the first time, enables joint, efficient, and imperceptible attacks on both control signals. Experiments on both simulated and real-world driving datasets demonstrate that the proposed method induces substantial deviations—averaging 3.54°–29° in steering angle and 11–22 km/h in speed—significantly outperforming five baseline attack methods.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Multiagent Systems: Adversarial Agents
• Machine Learning: Adversarial Learning & Robustness

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Security and Privacy: Large-scale security measurements
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Adversarial attacks play a pivotal role in testing and improving the reliability of deep learning (DL) systems. Existing literature has demonstrated that subtle perturbations to the input can elicit erroneous outcomes, thereby substantially compromising the security of DL systems. This has emerged as a critical concern in the development of DL-based safety-critical systems like Autonomous Driving Systems (ADSs). The focus of existing adversarial attack methods on End-to-End (E2E) ADSs has predominantly centered on misbehaviors of steering angle, which overlooks speed-related controls or imperceptible perturbations. To address these challenges, we introduce UniAda, a multi-objective white-box attack technique with a core function that revolves around crafting an image-agnostic adversarial perturbation capable of simultaneously influencing both steering and speed controls. UniAda capitalizes on an intricately designed multi-objective optimization function with the Adaptive Weighting Scheme (AWS), enabling the concurrent optimization of diverse objectives. Validated with both simulated and real-world driving data, UniAda outperforms five benchmarks across two metrics, inducing steering and speed deviations from 3.54 degrees to 29 degrees and 11 km per hour to 22 km per hour on average. This systematic approach establishes UniAda as a proven technique for adversarial attacks on modern DL-based E2E ADSs.

Problem

• adversarial attack
• autonomous driving systems
• multi-objective
• steering control
• speed control

Innovation

• multi-objective adversarial attack
• adaptive weighting scheme
• image-agnostic perturbation
• end-to-end autonomous driving
• steering and speed control