This paper focuses on the study of task-oriented grasp synthesis using partial point cloud data. The authors present a method for selecting a grasp so that the object is not lost during manipulation and adequate force/moment can be applied to perform the task. This is done by using a neural network to approximate a function for predicting the grasp quality metric on a cuboid shape. The proposed method does not require manually labeled data or grasping simulator, making it efficient to implement and integrate with motion planners. The effectiveness of this approach is demonstrated through simulation and experimental results.
Publication date: 22 Sep 2023
Project Page: https://irsl-sbu.github.io/Task-Oriented-Gra
Paper: https://arxiv.org/pdf/2309.11689
