Orienting Deformable Polygonal Parts without Sensors

Parts orienting is an important part of automated manufacturing. Sensorless manipulation has proven to be a useful paradigm in addressing parts orienting, and the manipulation of deformable objects is a growing area of interest. Until now, these areas have remained separate because existing orienting approaches utilize forces that if applied to deformable parts violate the assumptions used by existing algorithms, and could potentially break the part. We introduce a new algorithm and manipulator actions that, when provided with the geometric description and a deformation model of choice for the part, exploits the deformation and generates a Plan that consists of the shortest sequence of manipulator actions guaranteed to orient the part up to symmetry from any unknown initial orientation and pose. Additionally, the algorithm estimates whether a given manipulator is sufficiently precise to perform the actions which guarantee the final orientation. This is dictated by the particular part geometry, deformation model, and the manipulator action path planner which contains simple end-effector constraints and any standard motion planner. We illustrate the success of the algorithm with multiple parts through 192 trials of experiments that were performed with low-precision robot manipulators and six parts made of four types of materials. The experimental trials resulted in 154 successes, which show the feasibility of deformable parts orienting. The analysis of the failures showed that for success the assumptions of zero friction are essential for this work, increased manipulator precision would be beneficial but not necessary, and a simple deformation model can be sufficient. Finally, we note that the algorithm has applications to truly sensorless manipulation of non-deformable parts