Automatic Decompositions of Assembly Sequence Plans

This paper describes an approach to robotic task sequence planning which decomposes task sequences into operations sequences for an assembly workcell. The approach provides a framework for robust execution of tasks through properties of traceability - implicit mapping of operations to task representation, and viability - retaining multiple paths for execution. Given the descriptions of the objects in this system and all feasible geometric configurations and relationships among these objects and combinations of objects, an AND/OR net which describes the relationships of all feasible geometric states and associated feasibility criteria for net transitions is generated. This AND/OR net is mapped into a Petri net which incorporates all feasible sequences of high level operations. The resulting Petri net is then decomposed in a stepwise manner into a lower level Petri net of which each transition can be directly implemented by control commands or command sequences in an object-oriented framework to control and coordinate the operations of devices and objects in the system. All possible task sequences are found using a systolic stack algorithm. Each feasible sequence is pumped out of the stack and could be stored sequentially. A shortest sequence may be chosen as an output of the hierarchical planning system to efficiently implement the desired tasks.