GRASPING UNKNOWN OBJECTS

摘要

This paper describes a complete robotic system which is capable of removing unmodeled objects from a heap, one by one. As it relies on geometric information only, the use of range data is a natural choice. The objects are to be grasped by a two-fingered gripper, thus it is mandatory that the system can see opposite faces on the objects. Two range views from opposite sides are acquired and analyzed. We have succeeded in integrating the range data from the two views in a consistent way. Other accomplishments of the system are: controlled approximation of the scene by planar patches, creation of a representation of the scene which is rich enough to support segmentation of it into object hypotheses and the determination and ranking of grasping possibilities.