3D Attack

3D attack research focuses on developing and analyzing methods to adversarially perturb 3D point cloud data, causing misclassification by machine learning models used in applications like autonomous driving and robotics. Current research emphasizes creating imperceptible attacks, even under black-box conditions where model details are unknown, using techniques that manipulate point cloud coordinates, spectral representations, or geometric features. These attacks highlight vulnerabilities in 3D point cloud classifiers and drive the development of more robust models and defense mechanisms, ultimately improving the safety and reliability of AI systems in real-world scenarios.