Mapping kinematic interactions between objects for robot motion planning

Abstract

Knowledge about the possible interactions between objects is needed by a robot if it is to plan tasks employing those objects. In this paper, we focus on methods to represent the ways in which the degrees of freedom of a rigid body interlocked with another change as the relative pose of the two objects changes, in such a way so as to allow efficient planning queries about the objects. Narrow passages and occlusions make this a difficult problem for usual approaches in robotics involving sample based planners and 3D shape reconstruction. Instead, we propose a data structure (called a DoF map) that can be constructed from tactile information alone, and which, once constructed for a pair of rigid bodies, enables fast planning queries for that particular pair. The data structure is also sufficiently abstract and allows reuse even for objects with different geometry, as long as the new pair of objects is such that, if its DoF map were constructed, it would be isomorphic to the DoF map of the original pair. The DoF map then provides a possible criterion for object pair classification that is more general than exact geometric shape but more informative than simple topology. We describe a method to construct DoF maps, as well as a method to reuse a known one. We test these ideas in simulation.