Regarding grasp planning, an overwhelming majority of teams used custom approaches, which may appear surprising at first. To explain this, we first note that about 20% of teams opted for suction over grasping, which dramatically simplifies the problem by reducing it to choosing surfaces that are flat and planning to reach them. However, this level of customization is also indicative of more fundamental problems, namely, difficulties in generalizing the grasping problem across mechanical platforms, and difficulties in incorporating uncertainty and environmental context in to grasp planning. The output of "GraspIt!" Designates an end-effector and finger pose for a given object geometry that optimizes some wrench-based grasp metric. This metric ignores environmental context, reachability, pose uncertainty, and nonprehensile strategies, such as pushing, that may be more important than robustness to disturbance wrenches. Further tripping the balance toward custom solutions is that current trends in manipulation include shifting some is that require reasoning into end-effector compliance 5,40 and using under-actuated systems 37,41 -43. As a result, many objects can be grasped using simple rules, such as attempting a power grasp along the medial axis of the object. Compliance can play an important role even in conjuction with suction as indicated by the Unigripper's design with the Rutgers U. Pracsys team, where a foam is introduced between the object and the suction openings so as to help to adapting on the object.
It is important to note that the APC shelves are relatively uncluttered compared to the shelves encountered by human pickers, which may have dozens of object in close contact. This simplification may have biased the teams' choices of grasping strategies toward solutions like suction and standard parallel-jaw grippers, whereas a more complex arrangement of objects may motivate the use of human-like dexterity and grasp planning capabilites.