Toward autonomous disassembling of randomly piled objects with minimal perturbation

Autonomous capabilities for manipulating randomly piled objects may enhance current methods of path planning and open a new field of development for mobile manipulation and Urban Search And Rescue (USAR) robotics. This paper introduces the challenge of achieving such manipulation capabilities and as a first step presents three algorithms, including a proposed novel solution, for the selection of objects to remove from a pile. The proposed algorithm determines a removability rank for each object according to the degree of its encapsulation within other objects. Using the contact vectors of the examined object, it is possible to obtain the motions that will not violate the object's unilateral contact constraints. The removability rank of the object is proportional to the union of all such motions. All algorithms were tested in simulation in full and partial knowledge modes, and evaluated on a physical robot with a simple manipulator and sensor. This work contributes: the introduction of an important autonomous manipulation challenge, the solution of which will be useful in the field of manipulation in general and USAR in particular; a specific novel algorithm for the construction of disassembly plans for piled objects; and an experimental evaluation of three algorithms targeted at such construction.