On the automatic generation of plans for mechanical assembly

Abstract

The problem of moving a set of parts from a given initial configuration to a given final configuration has been well studied. However, this classical problem differs in several important respects from the type of assembly planning problem most commonly encountered in manufacturing applications. In manufacturing, a worksite is normally configured specifically for each task, so the layout of the workspace and the initial configuration of the parts must be treated as outputs of the planning process, rather than as inputs. A system to solve such problems must be capable of reasoning about a larger variety of part insertion trajectories and must be able to optimize with respect to a more complex set of criteria than existing robot planning systems allow. Furthermore, such a system must be flexible enough to h and le the wide variety of different criteria which arise in different assembly applications. After an analysis of the assembly planning problem, this dissertation describes an experimental assembly planner meeting these requirements which is capable of rapidly generating plans for building most assemblies. Only plans using no subassemblies or temporary positions are generated. It solves the problem of being able to reason about a large variety of trajectories by proposing them a priori and uses plug-in criteria modules to guide its search for optimal plans.