MAST: Multi-Agent Security Test Bed
uncoordinated_3_attackers.mp4Case Studies
Using the MAST platform with the Carla simulator, we performed multiple case studies using varying numbers of coordinated and un-coordinated attackers. See the case studies tab for additional case study videos
Key Contributions
We design MAST, the multi-agent security testbed, using ROS2 for security-analysis of MS/MA autonomy.
We integrate MAST with the recently-released multiagent dataset pipeline [17] for the first real-time playback of simulated MS/MA autonomy data.
We build the first bridge between AVstack [19] and ROS to integrate state-of-the-art AV algorithms into a scalable, near-real-time simulation.
We design an automated build process for AV algorithm pipelines using high-level configuration files to enable rapid prototyping of MS/MA configurations.
We provide a baseline framework for security analysis of MS/MA algorithms in MAST with (un)coordinated models of adversary nodes that can affect the sensing, perception, tracking, and/or communication of AVs.
We execute security-focused case studies and Monte Carlo evaluations with centralized sensor fusion and motivate the need for security-aware collaborative autonomy