Logic based modelling of goal-directed behavior

Abstract

We address the problem of characterizing goal-directed robotic behavior using a logic of actions and change. Our approach is based on distinguishing two kinds of actions: procedural actions which are defined in a mechanistic way, and goal-directed actions which are performed through a process involving tries, possibly failures, and corrective action and new tries until the goal has been reached. (The definition of procedural actions may be done external to the logic, for example through differential equations, or through a conventional programming language). For both kinds of actions, the logic expresses explicitly whether the action succeeds or fails. Each execution of a goal-directed action is also characterized by a number of breakpoints where some sub-action has been completed and a new sub-action for getting to the desired goal is selected. The logic is used for characterizing the selection of sub-actions at breakpoints, and the success or failure of the goal-directed action in terms of the success or failure of the sub-actions. The article describes how goal-directed actions can be modelled by an extension of existing results on logics of actions and change.