Abstract
This paper presents a geometric approach to identify a complete set of object subsurfaces in a cluttered scene, which permit an end-effector to approach and grasp the object. This work proposes the motion constraint graph (mcg) representation for this purpose. It is able to efficiently reason about the space of permissible end-effector poses by placing constraints on the hand's configuration space. These constraints account for the surface area of contact points, end-effector kinematics, as well as collisions with objects in the scene. For a given end-effector and an arbitrary scene, it is possible to formulate and solve a constraint satisfaction problem to compute the set of reachable subsurfaces, which permit valid grasps. The proposed approach is general, and is applicable to any kind of scene geometry, such as arbitrarily cluttered scenes and curved surfaces, as well as complex end-effectors with multiple degrees of freedom. In simulation experiments, the approach increases the success rate of grasp planning, while also reducing total online planning time at the cost of a small amount of precomputation. This approach defines a framework that can be applied in a variety of contexts, from accelerating grasp planning, pruning existing grasping databases, to identifying optimal end-effector designs for a given scene.