CAGE: Context-Aware Grasping Engine

摘要: Semantic grasping is the problem of selecting stable grasps that are functionally suitable for specific object manipulation tasks. In order robots to effectively perform manipulation, a broad sense contexts, including and task constraints, needs be accounted for. We introduce Context-Aware Grasping Engine, which combines novel semantic representation grasp contexts with neural network structure based on Wide & Deep model, capable capturing complex reasoning patterns. quantitatively validate our approach against three prior methods dataset consisting 14,000 44 objects, 7 tasks, 6 different states. Our outperformed all baselines by statistically significant margins, producing new insights into importance balancing memorization generalization grasping. further demonstrate effectiveness robot experiments in presented model successfully achieved 31 32 grasps. The code data available at: https://github.com/wliu88/railsemanticgrasping