Robotic Manipulation for Parts Transfer and Orienting: Mechanics, Planning, and Shape Uncertainty

Ph.D. thesis, The Robotics Institute, Carnegie Mellon University, December 1996, Robotics Institute Technical Report CMU-RI-TR-96-38.

Srinivas Akella

Abstract

Robots can modify their environment by manipulating objects. To fully exploit this ability, it is important to determine the manipulation capabilities of a given robot. Such characterization in terms of the physics and geometry of the task has important implications for manufacturing applications, where simpler hardware leads to cheaper and more reliable systems. This thesis develops techniques for robots to {\em transfer\/} parts from a known position and orientation to a goal position and orientation, and to {\em orient\/} parts by bringing them from an unknown initial orientation to a goal orientation. This {\em parts feeding\/} process is an important aspect of flexible assembly. Designing automatic planners that capture the task mechanics and geometry leads to flexible parts transfer and orienting systems. The implemented parts feeding systems use simple effectors that allow manipulation of a broad class of parts, and simple sensors that are robust and inexpensive.

The main research issues are to identify a set of actions for the robot that is {\em complete\/} for the task and to develop {\em automatic planners\/} that share this completeness property. That is, the actions should enable the robot to successfully execute the task, and the planners should automatically generate such sequences of actions.

To illustrate this approach, the thesis describes a set of parts transfer and orienting tasks, their mechanics, and planning techniques to solve them. The first example is a parts transfer system that automatically identifies a sensorless sequence of pushes for a robot to move any polygonal part to any goal position and orientation in the plane. The second system demonstrates that a one-joint robot can transfer any polygon to a specified goal position and orientation by pushing it on a conveyor. We present automatic planners that use mathematical programming formulations for these tasks. The thesis then describes a one-joint robot system to perform sensorless orienting of parts. The last system, also for parts orienting, demonstrates the speedup resulting from using inexpensive photosensors in combination with actions. The sensors provide partial information on a part's orientation by measuring its width; the actions rotate the part to orientations the sensors can identify. This system can orient multiple part shapes with a single plan. Further, the thesis analyzes the effects of shape uncertainty arising from manufacturing tolerances on parts orienting and identifies conditions under which we can orient parts with shape uncertainty. Planners for these systems have been implemented and experimentally demonstrated on industrial robots.