SA-Attack: Speed-adaptive stealthy adversarial attack on trajectory prediction

TL;DR

SA-Attack tackles adversarial vulnerabilities in neural trajectory predictors for autonomous driving by introducing a speed-adaptive, stealthy attack. It first identifies model-sensitive trajectory shapes through an unconstrained search, then reconstructs feasible paths with a continuous-curvature clothoid model and pure-pursuit, guided by future trajectory information to ensure smooth junctions. Empirical results on nuScenes and ApolloScape show substantial degradation in ADE and FDE across Trajectron++ and Grip++, outperforming a baseline attack and producing realistic, hard-to-detect perturbations. The work underscores robustness gaps in current predictors and provides a practical benchmark and methodology for evaluating defenses against stealthy trajectory-level adversaries.

Abstract

Trajectory prediction is critical for the safe planning and navigation of automated vehicles. The trajectory prediction models based on the neural networks are vulnerable to adversarial attacks. Previous attack methods have achieved high attack success rates but overlook the adaptability to realistic scenarios and the concealment of the deceits. To address this problem, we propose a speed-adaptive stealthy adversarial attack method named SA-Attack. This method searches the sensitive region of trajectory prediction models and generates the adversarial trajectories by using the vehicle-following method and incorporating information about forthcoming trajectories. Our method has the ability to adapt to different speed scenarios by reconstructing the trajectory from scratch. Fusing future trajectory trends and curvature constraints can guarantee the smoothness of adversarial trajectories, further ensuring the stealthiness of attacks. The empirical study on the datasets of nuScenes and Apolloscape demonstrates the attack performance of our proposed method. Finally, we also demonstrate the adaptability and stealthiness of SA-Attack for different speed scenarios. Our code is available at the repository: https://github.com/eclipse-bot/SA-Attack.

Figures (4)

• Figure 1: Speed-adaptive stealthy adversarial attack (SA-Attack) methodology overview
• Figure 2: Comparison of different attack methods.
• Figure 3: Quantitative comparison of adversarial trajectories at maximum acceleration.
• Figure 4: Qualitative comparison of the feasibility of adversarial trajectories.