Floating visual grasp of unknown objects using an elastic reconstruction surface

In this paper a new method for fast visual grasp of unknown objects is presented. The method is composed of an object surface reconstruction algorithm and of a local grasp planner, evolving in a parallel way. The reconstruction algorithm makes use of images taken by a camera carried by the robot, mounted in an eyein-hand configuration. An elastic reconstruction sphere, composed by masses interconnected each other by springs, is virtually placed around the object. The sphere is let to evolve dynamically under the action of external forces, which push the masses towards the object centroid. To smoothen the surface evolution, spatial dampers are attached to each mass. The reconstruction surface shrinks toward its center of mass until some pieces of its surface intercept the object visual hull, and thus local rejection forces are generated to push out the reconstruction points until they stay into the visual hull. This process shapes the sphere around the unknown object. Running in parallel to the reconstruction algorithm, the grasp planner moves the fingertips, floating on the current available reconstructed surface, according to suitable quality measures. The fingers keep moving towards local minima depending on the evolution of the reconstruction surface deformation. The process stops when the object has been completely reconstructed and the planner reaches a local minimum. Quality measures considering both hand and grasp proprieties are adopted. Simulations are presented, showing the effectiveness of the proposed algorithm.