Parts manufactured to tolerances have shape variations. Most work in robotic manipulation assumes that part shape does not vary. Orienting devices such as bowl feeders frequently fail due to variations in part shape. In this paper we develop techniques to orient parts with shape uncertainty. We present a shape uncertainty model and describe the nondeterminism in parts orienting that arises from shape uncertainty. We characterize a class of parts that can be reliably oriented with sensor-based and sensorless orienting plans under shape uncertainty. We present implemented planners that generate orienting plans for the entire variational class of part shapes given a nominal part shape and shape uncertainty bounds. We describe experiments to demonstrate generated plans and outline issues for future work.
Srinivas Akella, Matthew T. Mason