Grasp Planning via Decomposition Trees

Abstract

Planning realizable and stable grasps on 3D objects is crucial for many robotics applications, but grasp planners often ignore the relative sizes of the robotic hand and the object being grasped or do not account for physical joint and positioning limitations. We present a grasp planner that can consider the full range of parameters of a real hand and an arbitrary object, including physical and material properties as well as environmental obstacles and forces, and produce an output grasp that can be immediately executed. We do this by decomposing a 3D model into a superquadric 'decomposition tree' which we use to prune the intractably large space of possible grasps into a subspace that is likely to contain many good grasps. This subspace can be sampled and evaluated in GraspIt!, our 3D grasping simulator, to find a set of highly stable grasps, all of which are physically realizable. We show grasp results on various models using a Barrett hand.

Towards Grasp-Oriented Visual Perception for Humanoid Robots

“…Dependent on the primitive shape, one can test several grasp configurations on this shape. This work was continued by Goldfeder et al 63 using more sophisticated shape primitives, known as superquadrics (SQs), which are parameterizable models offering a large variety of different shapes. Most approaches that consider this problem start from point-clouds and synthesizing object shapes by using superquadrics in a bottom-up manner.…”

“…There is a multitude of state-of-the-art approaches based on parameterized superellipsoids for modeling 3D range data with shape primitives. [63][64][65][66] Assuming that an arbitrary point cloud has to be approximated, one SQ is not enough for most objects. The more complex the shape is, the more SQs have to be used to conveniently represent its different parts.…”

“…This is usually done by least square minimization of an insideoutside fitting function, as there is no analytical method to compute the distance between a point and a superquadric. 63 Thus, SQs are though a good trade-off between flexibility and computational simplicity, but sensitive to noise and outliers that will cause imperfect approximations. This is an important issue, as our work is oriented towards the use of dense stereo accompanied by highly distorted and incomplete data.…”

“…Though one could produce polygonal surfaces from 3D point data, for example by the Power Crust algorithm, 70 this introduces another step causing additional effort both in processing time and noise handling. The already mentioned approach by Goldfeder et al 63 combines superquadric representation and decomposition on regularly spaced range data.…”

“…by point clouds, 90 geometric shape primitives, 62,91,92 or superquadric parametrizations. [63][64][65]69 In our earlier work, 86,93 we have shown that even a box approximation of the point cloud can yield grasp hypotheses, as also interfaces to reason about task, viewpoint, graspability, and more. Our representation efficiently approximates a 3D point cloud by a constellation of Minimum Volume Bounding Boxes (MVBBs).…”