A unified interactive simulation platform for smart mobility: Adversarial attack generation in connected autonomous vehicles

Autonomous vehicles rely on complex perception and communication systems that are susceptible to adversarial attacks. We present an open-source simulation framework for reproducible multi-domain adversarial evaluation targeting both perception and communication layers. The framework enables high-fidelity, configuration-driven co-simulation of physical environments, traffic, and V2X networking, with seamless ROS 2 integration. Using this framework, we benchmark diverse adversarial attacks on 2D and 3D perception models, demonstrating realistic and effective attack scenarios. A teleoperation interface supports human-in-the-loop cooperative driving under adversarial conditions, highlighting the framework’s practicality, reproducibility, and extensibility for robustness assessment in autonomous driving. Experiments on state-of-the-art 2D and 3D models demonstrate substantial performance degradation under adversarial conditions. Within this context, a representative teleoperation scenario is realized to showcase the framework’s capability to support cooperative driving and human-in-the-loop operation in the presence of active adversarial attacks, highlighting its practicality, reproducibility, and extensibility for comprehensive robustness assessment in autonomous driving. Videos illustrating the proposed framework and the teleoperation scenario are available at https://doi.org/10.6084/m9.figshare.30951002 .