3D Gaussian Splat Vulnerabilities
This work addresses safety risks for applications like autonomous navigation by exposing new adversarial threats in 3DGS, though it is incremental as it adapts existing attack methods to a new domain.
The paper tackles security vulnerabilities in 3D Gaussian Splatting (3DGS) by introducing CLOAK, which embeds adversarial content visible only from specific viewpoints using view-dependent appearances, and DAGGER, a targeted attack that perturbs 3D Gaussians to deceive object detectors like Faster R-CNN.
With 3D Gaussian Splatting (3DGS) being increasingly used in safety-critical applications, how can an adversary manipulate the scene to cause harm? We introduce CLOAK, the first attack that leverages view-dependent Gaussian appearances - colors and textures that change with viewing angle - to embed adversarial content visible only from specific viewpoints. We further demonstrate DAGGER, a targeted adversarial attack directly perturbing 3D Gaussians without access to underlying training data, deceiving multi-stage object detectors e.g., Faster R-CNN, through established methods such as projected gradient descent. These attacks highlight underexplored vulnerabilities in 3DGS, introducing a new potential threat to robotic learning for autonomous navigation and other safety-critical 3DGS applications.