Automatic Grasp Planning in the Presence of Uncertainty

Abstract

This paper presents an algorithm for automatic planning of robot grasping motions that are insensitive to bounded uncertainties in the object's location. The algorithm plans parallel-jaw grasping motions for arbitrary two-dimensional polygonal objects, which need not be of uniform density. Grasping motions are viewed as parameterized operations, where the parameter values that describe an individual operation define an operation space of all possible operations. By combining an analysis of object geometry and the physics of friction, the planning algorithm divides the operation space into regions, where all operations within a given region produce the same final grasping configuration. Task uncertainties are then included by shrinking these regions by the amount of uncertainty present. The smaller regions that remain after shrinking indicate all those operations that will successfully result in a given grasping configuration, even if the worst-case combination of errors occurs.

Moreover, the grasping operations presented in this paper intrinsically reduce task uncertainty. If an operation is chosen from a region and executed, then two degrees of uncertainty will be removed from the object's position when the operation is completed.

It is shown that simple squeeze-grasp operations are not sufficient for grasping all possible objects, and offset-grasp and push-grasp operations are added to increase the scope of the planner. The planner has been implemented, and the resulting program has been tested with an industrial manipulator.