Benchmarks for Understanding Grasping
In the BURG Project we set out to boost grasping research by focusing on complete tasks and the related object manipulation constraints. In doing so, we need to move from objects to parts, since object parts facilitate the interpretable usage of objects. Parts are essential to know how and where the gripper can grasp given the constraints imposed by the task, e.g., pouring from a container implies grasping from the side. The novelty will come from learning to predict plausible grasps and to link grasps to the parts responsible for selecting each of them. In BURG we will boost grasping research by focusing on complete tasks and the related object manipulation constraints. In doing so, we need to move from objects to parts (Fig. 1), since object parts facilitate the interpretable usage of objects. Parts are essential to know how and where the gripper can grasp given the constraints imposed by the task, e.g., pouring from a container implies grasping from the side. The novelty will come from learning to predict plausible grasps and to link grasps to the parts responsible for selecting each of them.
FUNDING:
The project is funded by FWF – Austrian Science Foundation & CHIST-ERA.