The paper introduces 3D Depth Fool (3D²Fool), a novel 3D texture-based adversarial attack designed to manipulate monocular depth estimation (MDE) models in autonomous driving. Unlike previous 2D adversarial patches, 3D²Fool generates robust 3D textures that can affect the depth estimation of vehicles from various viewpoints and under different weather conditions. The attack is optimized to be applicable to a wide range of target vehicles, including cars, buses, and pedestrians, and is designed to be more effective in challenging scenarios such as rain and fog. The method involves two main components: texture conversion (TC) and physical augmentation (PA). TC converts a 2D adversarial texture seed into a 3D camouflage texture, while PA simulates various weather conditions to improve the attack's robustness. Experimental results show that 3D²Fool outperforms existing 2D and 3D texture-based attacks in terms of depth estimation error and the ratio of affected regions. Real-world experiments on a scaled Tesla Model Y car further validate the effectiveness of the attack, demonstrating that it can cause an MDE error of over 10 meters. The code for 3D²Fool is available at <https://github.com/Gandolfczjh/3D2Fool>.The paper introduces 3D Depth Fool (3D²Fool), a novel 3D texture-based adversarial attack designed to manipulate monocular depth estimation (MDE) models in autonomous driving. Unlike previous 2D adversarial patches, 3D²Fool generates robust 3D textures that can affect the depth estimation of vehicles from various viewpoints and under different weather conditions. The attack is optimized to be applicable to a wide range of target vehicles, including cars, buses, and pedestrians, and is designed to be more effective in challenging scenarios such as rain and fog. The method involves two main components: texture conversion (TC) and physical augmentation (PA). TC converts a 2D adversarial texture seed into a 3D camouflage texture, while PA simulates various weather conditions to improve the attack's robustness. Experimental results show that 3D²Fool outperforms existing 2D and 3D texture-based attacks in terms of depth estimation error and the ratio of affected regions. Real-world experiments on a scaled Tesla Model Y car further validate the effectiveness of the attack, demonstrating that it can cause an MDE error of over 10 meters. The code for 3D²Fool is available at <https://github.com/Gandolfczjh/3D2Fool>.