Vision-based grasp planning of 3D objects by extending 2D contour based algorithms

This paper addresses the problem of grasp planning of unmodelled 3D objects with the aide of vision. The approach proposes the extension of fast 2D contour-based grasp planning algorithms, and the use of images of the target object from different points of view to recover critical 3D information like the size, location and the pose. A complete method is described, and experimental results are presented showing its feasibility.