A Programmable Parts Feeder

An introduction by the author.
A sample animation.

This hypertext document was prepared by Carl Sutter and Ken Goldberg in Spring 1994 based on an article that appeared in the Carnegie Mellon Robotics Institute Annual Report in 1990. More recent articles on related topics can be found here.

• Introduction
  • Parts Feeders and Autonomous Manipulation
• The Parts-Feeding Mechanism
• Example Plan
  • Mechanical Analysis
• The Planning Algorithm
  • Recovering Plans from a Sequence of Preimages
  • Optimality of the Plans
  • Worst-Case Computational Complexity
  • Completeness
• Implementation
• Discussion
  • Stochastic Plans
  • Pushing
  • Future Work
• Acknowledgements
• References

Figure 1: A parts feeder orients parts as they arrive on the left-hand conveyor belt.

The key advantage of a programmable parts feeder is flexibility. Flexible manufacturing systems can be used for a variety of products. Such flexibility reduces the number of machines required and can reduce the engineering overhead and lead time for manufacturing new products. In the Manipulation Laboratory, we pursue basic research related to flexible manufacturing systems and general-purpose robotic manipulation. In particular, we study the mechanics of manipulation and methods for automatically planning manipulation strategies. Our design for a programmable parts feeder is an application of this basic research.

Traditional parts feeders are often inflexible. Although there are a vast number of different techniques for feeding parts, most are hand-crafted mechanisms that depend critically on the shape of the part. When part geometry changes, the feeder must be mechanically redesigned with a trial-and-error process that can require several months [18].

In contrast, a programmable parts feeder can be reprogrammed rather than physically modified when part geometry changes. Flexibility is enhanced by automatically generating the appropriate program (or plan). Our programmable parts feeder has two components.

• Mechanism. A device that can orient parts under software control. We use a modified parallel-jaw gripper. The mechanism requires one controlled degree of freedom to orient the gripper, and one uncontrolled degree of freedom (such as a pneumatic actuator) to open and close the jaws.

• Planning Algorithm. A method for transforming a geometrical part description into a program for the mechanism. We present a planning algorithm that accepts an -sided polygonal part description as input and automatically generates a program (plan) for orienting the part. The algorithm runs in time . We prove that the algorithm is complete, i.e. it works for any polygonal part.

Each plan specifies a sequence of open-loop gripping actions for orienting the part up to symmetry. Three additional components are required: a means for separating jumbled parts into a stream of isolated parts, a conveyor belt for transporting parts in and out of the feeder, and a binary filter that can distinguish between symmetric part orientations (e.g., a silhouette trap).