AdvGrasp: Adversarial Attacks on Robotic Grasping from a Physical Perspective
Authors: Xiaofei Wang, Mingliang Han, Tianyu Hao, Cegang Li, Yunbo Zhao, Keke Tang
IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments across diverse scenarios validate the effectiveness of Adv Grasp, while real-world validations demonstrate its robustness and practical applicability. |
| Researcher Affiliation | Collaboration | 1Department of Automation, University of Science and Technology of China 2Smart More Corporation 3Cyberspace Institute of Advanced Technology, Guangzhou University 4Institute of Artificial Intelligence, Hefei Comprehensive National Science Center |
| Pseudocode | No | The paper describes the method's steps under Section 4.4 'Implementation of Adv Grasp' with sub-sections like 'Bounding Box Initialization', 'Control Point Placement', and 'Iterative Shape Deformation and Optimization', but it does not include a formal pseudocode block or algorithm figure. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code or a link to a code repository for the methodology described. Figure 6 mentions 'The detailed process is available in the video provided in the supplementary materials.', which refers to a video, not code. |
| Open Datasets | Yes | Object Selection. We select 20 representative objects with diverse shapes from the Grasp Net-1Billion dataset [Fang et al., 2020; Fang et al., 2023], including common items such as CRACKER BOX, TOMATO SOUP CAN, and MUSTARD BOTTLE. |
| Dataset Splits | No | The paper describes the creation of the Adv Grasp-20 benchmark, including object selection and grasp generation, but it does not specify how these grasps or the dataset are split into training, validation, or test sets for their own experimental reproduction. The evaluations are performed on the generated grasps within the benchmark. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments or simulations. It mentions using a 'Robotiq 2-finger 2F-85 gripper' and 'Robotiq 3-finger gripper' for real-world validation, which are robotic components, not the computing hardware itself. |
| Software Dependencies | No | We implement our framework in Python using the Py Bullet simulator [Coumans and Bai, 2016] as the simulation environment. All simulations employ the soft finger contact model [Mahler et al., 2018], with the friction coefficient set to µ = 0.6 and the torsional friction coefficient γ = 0.3. No specific version numbers for Python or PyBullet, or other libraries, are provided. |
| Experiment Setup | Yes | The optimization process utilizes a simulated annealing algorithm, with an initial temperature T0 = 1000, a minimum temperature Tmin = 10 5, a cooling rate α = 0.98, and a perturbation scale ϵ = 0.05 cage size for control points. To balance the contributions of lift capability, grasp stability, and shape regularization in Adv Grasp, we set the weighting parameters to λ1 = 10000 and λ2 = 50. All simulations employ the soft finger contact model [Mahler et al., 2018], with the friction coefficient set to µ = 0.6 and the torsional friction coefficient γ = 0.3. |