Talk Proposal: Towards the Realistic Evaluation of Evasion Attacks using CARLA
This work addresses the need for more realistic security evaluations in autonomous driving systems, though it appears incremental by applying an existing attack method to new simulation scenarios.
The authors tackled the problem of evaluating evasion attacks on object detectors by introducing ShapeShifter, a content-preserving attack method, and demonstrated its application in realistic urban driving scenarios using the CARLA simulator.
In this talk we describe our content-preserving attack on object detectors, ShapeShifter, and demonstrate how to evaluate this threat in realistic scenarios. We describe how we use CARLA, a realistic urban driving simulator, to create these scenarios, and how we use ShapeShifter to generate content-preserving attacks against those scenarios.