Automatic synthesis of perception driven discrete event control laws

A representation for perception-driven discrete-event control laws is described. An artificial intelligence (AI) planning program that can interpret and manipulate the representation to synthesize control laws automatically from symbolic descriptions of the task domain, the individual actions, and the goal condition has been developed. The generated control laws are highly conditional, making heavy use of perceptual information to determine the subgoals and hence the actions that are appropriate at each moment. The representation is rich enough to allow explicit inclusion of the sensing actions that obtain the perceptual information required by the conditional law. Sensing actions may have preconditions that require effector actions. A key outstanding problem is that the planner has yet to be proved sound (correct) and complete (able to find a control law whenever one exists)