A representation for serial robotic tasks

Abstract

The representation for serial robotic tasks proposed in this thesis is a language of temporal constraints derived directly from a model of the space of serial plans. It was specifically designed to encompass problems that include disjunctive ordering constraints. This guarantees that the proposed language can completely and, to a certain extent, compactly represent all possible serial robotic tasks. The generality of this language carries a penalty. The proposed language of temporal constraints is NP-Complete. Specific methods have been demonstrated for normalizing constraints posed in this language in order to make subsequent sequencing and analysis more tractable. Using this language, the planner can specify necessary and alternative orderings to control undesirable interactions between steps of a plan. For purposes of analysis, the planner can factor a plan into strategies, and decompose those strategies into essential components. Using properly normalized constraint expressions the sequencer can derive admissible sequences and admissible next operations. Using these facilities, a robot can be given the specification of a task and it can adapt its sequence of operations according to run-time events and the constraints on the operations to be performed.