@MISC{Dejong_explanation-basedcontrol, author = {Gerald Dejong}, title = {Explanation-Based Control of the Acrobot}, year = {} }

Share

OpenURL

Abstract

In our approach intelligent contrd integrates symbolic reasoning of artificial intelligence (AI) into the control prcr.edurĀ¢. We describe how this can be acc.~iished by a straightforward abstraction of the conventional notion of the central model. We apply the techniques to the problem of swing-up contrd of the acrobot. "lhe resulting Explanation-Based Gantrd strategy is compared with two mare-con~ealtionally-derived contrd strategies. We briefly discuss the strengths and weaknesses of the new appr oach. Model-Based Reasoning and Model-Based Control We wish to draw a parallel between model-based reasoning in AI and model-based control. We adopt symbolic inference, one of the most common reasoning frameworks in AI. One begins by writing a set of axioms in some logic (e.g., first order predicate calculus) which captures aspects of interest about the world. We identify this axiom set as the model of the AI system. Reasoning consists of drawing conclusions from the model according to some formal inference laws. Inferences (or theorems) are explicit statements of properties that were previously only implicit in the axioms. To reason about the control of mechanisms, one can write symbolic axioms describing the dynamical properties of the mechanism. The inference process then makes explicit previously implicit properties of the system. From a suitably abstract point of view this is just what happens in conventional control theory. The only difference is in the choice of notational vocabulary used to specify the model. In control theory, the model is the set of equations describing some world behavior. The inferences are explicit statements of