5.5 Hypothesize and test analysis

Necessary variations on our naive, basic tooling-design approach include: abstraction to a small number of relatively thick layers in the tool; superposition of adjacent part outlines to compute non-interfering thick-layered finger contours; computing left and right envelopes of part outlines for finger-stroke clearance; computing vertical-approach clearances (possibly leading to tool contour modifications); eroding finger widths resulting in only partial encirclement of grasped parts; superimposing multiple part outlines to compute multipurpose gripper contours; and incorporating contour perturbations for guiding imprecisely approached parts into precisely known grasp poses. Each of these variations violates the assurance of form (or frictional form) closure associated with the naive, basic approach.

Our more sophisticated tooling-design process offers many prospective rewards, but at the expense of losing any guarantees of grasp stability. To generate feasible tooling designs automatically, we are developing a hypothesize-and-test approach. Candidate finger shapes will be generated automatically, with consideration of multiple part handling, limited stroke length, approach-ability, and approach uncertainty. Automatically generated tool designs will be tested computationally for grasp form closure or frictional form closure (e.g. by invoking the linear-programming analysis reported by Trinkle in [20]). Tool designs will be iterated automatically within the space of design variables described above, resulting in practical designs which can be fabricated cheaply and quickly.

Prev, Up, Next