Intelligent grasp planning for robot programming

Abstract

In an attempt to experiment with the next generation of programming methodologies, but restrict the domain to a tractable and therefore achievable subset, this work explores the concept of automatic grasp planning for robot programming. Prior efforts in this field have restricted the research domain to include the part and grasping it in place. Our effort extends the scope of research to cover the intended use of the part to be manipulated, and provides for a more comprehensive collision avoidance computation than in previous attempts. In addition this research extends the notion of grasp planning from the actual grasp point to include the approach and departure both at the pick up point and drop off locations. These results are achieved by embedding the grasp selection system directly in a three dimensional solid modeling system developed by IBM. The selection of the GDP system is explored and criterion for such a system are developed and explained. We have introduced new concepts for grasp planning which define a universal grasping set, and a sequence of manual and automatic operations which refine the initial set of possibilities according to a set of filters into a prioritized set of possible grip locations. The system built for research permits a spectrum of interactions from completely automatic to manual in the way it explores the set of possible grasp points and selects a final candidate. The work explores in depth the grip selection problem and develops strategies to cope with a myriad of problems including the requirements to define an intermediate drop off and pick up location for some tasks. The theoretical discussions in the written thesis extend beyond the scope of the demonstration system implementation which was itself extensive. Our experience indicates that the division of effort must be adjustable and will change based on circumstance and experience. Beyond the notion of automatic grasp planning we study the embedding of this capability into a next generation task oriented robot programming methodology. Accordingly, we define a set of equivalence classes for the grasping task and explore syntactic forms for the inclusion of automatic grasp selection in a next generation programming language. (Abstract shortened with permission of author.)