Planning for Behaviour-Based Robotic Assembly: A Logical Framework

This thesis describes an approach to robotic assembly planning based on a logic of plan specification formulae. These formulae consist of a term representing the plan, a description of the states before and after the plan is performed (using a simple but structured world model), and a set of literals which can be evaluated as Prolog goals to test preconditions and generate terms to appear in the postconditions. The specifications for the atomic actions are given as axioms and planning takes place by attempting to prove the specification for the desired plan, decomposing it by the reverse application of inference rules that state how the specification for a plan can be deduced from those of its subplans. These are derived from the definitions of the temporal operators that are used to constuct plans. The initial plan may be partially instantiated due to prior constraints on the form of the plan, and the remaining plan variables and atomic action parameters will be instantiated during the course of planning. The planning strategy is expressed using goal-decomposing ‘tactics’. The representation of world states is defined using equational logic and (equational) unification is used to ‘match’ these state specifications. The form of this logic is designed to meet the requirements of behaviour-based assembly systems such as Edinburgh’s SOMASS system, which is described in this thesis, and to allow various temporal languages to be used to represent plans, in order to investigate their uses and to develop appropriate planning strategies.