Numerous studies on adversarial attacks targeting self-driving policies fail to incorporate
realistic-looking adversarial objects, limiting real-world applicability. Building upon prior …